<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by rafat adebanjo on Medium]]></title>
        <description><![CDATA[Stories by rafat adebanjo on Medium]]></description>
        <link>https://medium.com/@aderafat?source=rss-f2026239515------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*ixl2_LpnehnvMjFq.jpg</url>
            <title>Stories by rafat adebanjo on Medium</title>
            <link>https://medium.com/@aderafat?source=rss-f2026239515------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 10:57:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@aderafat/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Raw Trade Data to Meaningful Insights: My Week 8 Journey with Pandas, Seaborn and SQL]]></title>
            <link>https://medium.com/@aderafat/from-raw-trade-data-to-meaningful-insights-my-week-8-journey-with-pandas-seaborn-and-sql-efe970103278?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/efe970103278</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Sun, 01 Nov 2026 20:43:18 GMT</pubDate>
            <atom:updated>2026-11-01T20:43:18.238Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EANFj991-fXgU7s251Bixw.jpeg" /></figure><p>This week at DataRaflow, I explored how to transform raw datasets into meaningful information using Python, Pandas, Seaborn, Matplotlib and SQL.</p><p>The focus was on <strong>grouping and aggregating data, analysing trade patterns, identifying major trading partners, comparing imports and exports, and presenting findings through visualisations.</strong></p><p>For the practical exercises, I worked with two international trade datasets from the United Nations Comtrade database:</p><ol><li>United Kingdom milk and cream imports and exports from January to May 2015.</li><li>Nigeria’s crude and refined petroleum imports and exports from January to December 2023.</li></ol><p>Working with these datasets helped me understand how the same analytical techniques can be applied to different commodities, countries and periods.</p><p>I also completed five SQL practice exercises, which allowed me to apply some of the concepts I had learnt with Pandas using SQL queries.</p><p>Here is a breakdown of my learning experience, the tasks I completed, the challenges I encountered and some of the insights I gained.</p><h3>1. Understanding the United Nations Comtrade Dataset</h3><p>The first part of the assignment involved obtaining monthly international trade data from the United Nations Comtrade database.</p><p>The database provides information about international trade, including reporting countries, trading partners, commodities, imports, exports and trade values.</p><p>For the UK exercise, I worked with two milk and cream commodity categories:</p><ul><li><strong>0401:</strong> Milk and cream, neither concentrated nor sweetened.</li><li><strong>0402:</strong> Milk and cream, concentrated or sweetened.</li></ul><p>The dataset covered January to May 2015.</p><p>For my individual project, I selected Nigeria as the reporting country and petroleum as my commodity of interest. I focused on:</p><ul><li><strong>2709:</strong> Crude petroleum oils.</li><li><strong>2710:</strong> Refined petroleum oils and oils obtained from bituminous minerals.</li></ul><p>I downloaded the monthly import and export records for all twelve months of 2023.</p><p>This gave me an opportunity to investigate Nigeria’s petroleum trade relationships, including the countries it imports from, the countries it exports to and how the trade values change over time.</p><h3>2. Loading and Combining Data with Pandas</h3><p>One of the first things I needed to do was load the downloaded CSV files into Pandas.</p><p>For Nigeria’s petroleum analysis, I downloaded crude and refined petroleum data separately. I therefore started by importing Pandas and loading both files.</p><pre>import pandas as pd</pre><pre>crude = pd.read_csv(<br>    &#39;crude_petroleum.csv&#39;,<br>    dtype={&#39;cmdCode&#39;: str},<br>    index_col=False<br>)</pre><pre>refined = pd.read_csv(<br>    &#39;refined_petroleum.csv&#39;,<br>    dtype={&#39;cmdCode&#39;: str},<br>    index_col=False<br>)</pre><p>I used dtype={&#39;cmdCode&#39;: str} to ensure that commodity codes were treated as text rather than numerical values.</p><p>Commodity codes are identifiers, so treating them as strings helps preserve their original format.</p><p>I also used index_col=False because the downloaded CSV files contained an extra field compared with the header. Without this, some columns could be incorrectly interpreted.</p><h3>Combining the two datasets</h3><p>After loading the files, I checked their column names to confirm that both datasets had the same structure.</p><pre>crude.columns == refined.columns</pre><p>The result confirmed that the columns matched.</p><p>I then combined the two DataFrames using concat().</p><pre>df = pd.concat(<br>    [crude, refined],<br>    ignore_index=True<br>)</pre><p>To confirm that the combination worked, I checked the number of rows:</p><pre>print(len(crude), len(refined), len(df))</pre><p>Output:</p><pre>245 366 611</pre><p>The crude dataset contained 245 rows, the refined dataset contained 366 rows, and the combined DataFrame contained 611 rows.</p><p>This was my first practical reminder to validate data after combining files. Rather than assuming that the operation worked, I checked whether the resulting row count matched the expected total.</p><h3>3. Data Preparation and Cleaning</h3><p>The original Comtrade dataset contained 47 columns. Many of these columns were not needed for my analysis.</p><p>I first renamed the columns to make them more descriptive and consistent with the UK milk dataset.</p><p>For example, Comtrade used:</p><ul><li>period instead of Period.</li><li>flowDesc instead of Trade Flow.</li><li>partnerDesc instead of Partner.</li><li>cmdCode instead of Commodity Code.</li><li>primaryValue instead of Trade Value (US$).</li></ul><p>I also created a readable month description from the period column.</p><pre>df[&#39;Period Desc.&#39;] = pd.to_datetime(<br>    df[&#39;Period&#39;].astype(str),<br>    format=&#39;%Y%m&#39;<br>).dt.strftime(&#39;%b %Y&#39;)</pre><p>This converted values such as 202301 into Jan 2023.</p><p>After renaming the columns, I selected the relevant fields for my analysis.</p><pre>COLUMNS = [<br>    &#39;Year&#39;,<br>    &#39;Period&#39;,<br>    &#39;Trade Flow&#39;,<br>    &#39;Reporter&#39;,<br>    &#39;Partner&#39;,<br>    &#39;Commodity&#39;,<br>    &#39;Commodity Code&#39;,<br>    &#39;Trade Value (US$)&#39;<br>]</pre><pre>df_new = df[COLUMNS]</pre><p>This reduced the dataset to eight columns.</p><h3>Removing the World records</h3><p>One important cleaning step was removing records where the partner was recorded as World.</p><p>The World records represent the total trade with all partners. Since I wanted to calculate trade values for individual countries, retaining these records would cause double-counting when I summed the individual country values.</p><p>I removed them using:</p><pre>df_new = df_new[<br>    df_new[&#39;Partner&#39;] != &#39;World&#39;<br>]</pre><p>The dataset reduced from 611 rows to 575 rows.</p><pre>print(df_new.shape)</pre><p>Output:</p><pre>(575, 8)</pre><p>I also checked for missing values and examined the dataset using describe().</p><p>This helped me understand the structure of the data before proceeding with the analysis.</p><p><strong>My takeaway:</strong> Data cleaning requires more than removing empty cells. It also requires understanding the meaning of records and identifying information that could distort the results.</p><h3>4. Grouping and Aggregating Data</h3><p>This was one of the most important concepts I practised during the week.</p><p>Grouping allows us to organise records based on one or more columns, while aggregation allows us to calculate summary values for each group.</p><p>For example, instead of looking through hundreds of individual trade records, I could group the records by trade flow and calculate the total value of imports and exports.</p><h3>Grouping by commodity code</h3><p>I also wanted to understand which petroleum product contributed more to the overall trade value.</p><pre>group_by_comm_code = (<br>    df_new.groupby(&#39;Commodity Code&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>)</pre><pre>group_by_comm_code</pre><p>The results showed:</p><p>Commodity codeTotal trade value2709 — Crude petroleum$48,979,144,8762710 — Refined petroleum$20,223,626,504</p><p>However, these totals combine imports and exports. To understand the direction of trade, I grouped the data by trade flow, commodity code and year.</p><pre>GROUP_BY_TRADE_COMMODITY_YEAR = [<br>    &#39;Trade Flow&#39;,<br>    &#39;Commodity Code&#39;,<br>    &#39;Year&#39;<br>]</pre><pre>groupings = df_new.groupby(<br>    GROUP_BY_TRADE_COMMODITY_YEAR<br>)[&#39;Trade Value (US$)&#39;].sum()</pre><pre>groupings</pre><p>The result was particularly interesting:</p><p>The data showed that crude petroleum accounted for almost all recorded exports, while refined petroleum accounted for all recorded imports.</p><p>There were no recorded crude petroleum imports in this dataset.</p><p>This helped me understand why grouping by more than one column is useful. A general total can hide important differences between categories.</p><h3>5. Identifying Nigeria’s Major Trading Partners</h3><p>Another question I wanted to answer was:</p><p><strong>Which countries does Nigeria import petroleum from, and which countries buy petroleum from Nigeria?</strong></p><p>To answer this, I separated the dataset into imports and exports.</p><pre>imports = df_new[<br>    df_new[&#39;Trade Flow&#39;] == &#39;Import&#39;<br>]</pre><pre>exports = df_new[<br>    df_new[&#39;Trade Flow&#39;] == &#39;Export&#39;<br>]</pre><p>I then grouped each DataFrame by partner country and calculated the total trade value.</p><h3>Nigeria’s top import partners</h3><pre>top_import = (<br>    imports.groupby(&#39;Partner&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>    .head(10)<br>)</pre><pre>print(top_import)</pre><p>The five largest import partners were:</p><p>Country: Belgium, Netherlands, India, Malta and Russian Federation</p><p>Belgium was Nigeria’s largest import partner for the selected petroleum products in 2023.</p><p>The analysis also showed that Nigeria imported from 56 countries.</p><h3>Nigeria’s top export partners</h3><p>I applied the same method to the export records.</p><pre>top_export = (<br>    exports.groupby(&#39;Partner&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>    .head(10)<br>)</pre><pre>print(top_export)</pre><p>The five largest export destinations were:</p><p>Netherlands, India, Spain, Indonesia, USA</p><p>The Netherlands was Nigeria’s largest export destination for the selected petroleum products.</p><p>Nigeria exported to 39 countries.</p><p>An interesting observation was that the Netherlands and India appeared among both the major import and export partners.</p><p>This demonstrated why it is important to analyse imports and exports separately. A country can be an important destination for exports and also a significant source of imports.</p><h3>6. Measuring Dependence on Major Trading Partners</h3><p>Identifying the largest trading partners was useful, but I wanted to go further.</p><p>I wanted to understand how much Nigeria depended on its biggest partners compared with the total value of trade.</p><p>To do this, I calculated the combined value of the top three partners and compared it with the total value of imports or exports.</p><h3>Top three import partners compared with total imports</h3><pre>import_by_partner = (<br>    df_new[df_new[&#39;Trade Flow&#39;] == &#39;Import&#39;]<br>    .groupby(&#39;Partner&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>)</pre><pre>top3_import = import_by_partner.head(3)<br>world_import = import_by_partner.sum()</pre><pre>plot_import = pd.concat([<br>    top3_import,<br>    pd.Series({&#39;World total&#39;: world_import})<br>]) / 1_000_000_000</pre><pre>plot_import.plot(kind=&#39;bar&#39;)</pre><pre>plt.title(&#39;Nigeria: Top 3 Petroleum Import Partners vs Total Imports&#39;)<br>plt.xlabel(&#39;Country&#39;)<br>plt.ylabel(&#39;Trade value (US$ billions)&#39;)<br>plt.xticks(rotation=0)<br>plt.tight_layout()<br>plt.show()</pre><pre>print(<br>    &#39;Top 3 share of total imports:&#39;,<br>    f&#39;{top3_import.sum() / world_import:.1%}&#39;<br>)</pre><p>The top three import partners were Belgium, the Netherlands and India.</p><p>Together, they accounted for approximately <strong>52.1% of Nigeria’s total petroleum imports</strong>.</p><p>This suggested that Nigeria’s imports of the selected petroleum products were relatively concentrated among a small number of suppliers.</p><h3>Top three export partners compared with total exports</h3><p>I repeated the same analysis for exports.</p><pre>export_by_partner = (<br>    df_new[df_new[&#39;Trade Flow&#39;] == &#39;Export&#39;]<br>    .groupby(&#39;Partner&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>)</pre><pre>top3_export = export_by_partner.head(3)<br>world_export = export_by_partner.sum()</pre><pre>plot_export = pd.concat([<br>    top3_export,<br>    pd.Series({&#39;World total&#39;: world_export})<br>]) / 1_000_000_000</pre><pre>plot_export.plot(kind=&#39;bar&#39;)</pre><pre>plt.title(&#39;Nigeria: Top 3 Petroleum Export Partners vs Total Exports&#39;)<br>plt.xlabel(&#39;Country&#39;)<br>plt.ylabel(&#39;Trade value (US$ billions)&#39;)<br>plt.xticks(rotation=0)<br>plt.tight_layout()<br>plt.show()</pre><pre>print(<br>    &#39;Top 3 share of total exports:&#39;,<br>    f&#39;{top3_export.sum() / world_export:.1%}&#39;<br>)</pre><p>The Netherlands, India and Spain were the three largest export partners.</p><p>Together, they accounted for approximately <strong>31.7% of total exports</strong>.</p><p>Comparing both results revealed an interesting difference:</p><ul><li>The top three import partners accounted for 52.1% of imports.</li><li>The top three export partners accounted for 31.7% of exports.</li></ul><p>This suggests that Nigeria’s imports were more concentrated among its largest suppliers than its exports were among its largest buyers.</p><h3>7. Identifying Regular Trading Partners</h3><p>I also wanted to answer another question:</p><p><strong>Which countries bought petroleum from Nigeria in every month of 2023?</strong></p><p>Initially, I considered counting the number of rows for each country. However, this was not the correct approach because a country could have more than one record in the same month, for example, when it purchased both crude and refined petroleum.</p><p>Instead, I needed to count the number of distinct months in which each partner appeared.</p><p>First, I selected the export records:</p><pre>petroleum_export = df_new[<br>    df_new[&#39;Trade Flow&#39;] == &#39;Export&#39;<br>]</pre><p>Then I created a function:</p><pre>def buysEveryMonth(group):<br>    return group[&#39;Period&#39;].nunique() == 12</pre><p>The function checks whether a trading partner has records in all twelve months.</p><p>I grouped the data by partner and applied the function:</p><pre>grouped = petroleum_export.groupby(&#39;Partner&#39;)</pre><pre>regular = grouped.filter(buysEveryMonth)</pre><p>Finally, I extracted the regular customers:</p><pre>regular_customers = sorted(<br>    regular[&#39;Partner&#39;].unique()<br>)</pre><pre>print(len(regular_customers))<br>print(regular_customers)</pre><p>The result identified ten regular trading partners:</p><ul><li>Areas, nes</li><li>Canada</li><li>Côte d’Ivoire</li><li>France</li><li>India</li><li>Indonesia</li><li>Italy</li><li>Netherlands</li><li>Spain</li><li>United Kingdom</li></ul><p>One important detail is that Areas, nes is a catch-all trade category, not an individual country. Therefore, the result represents nine named countries and one additional partner category.</p><p>Together, these regular partners accounted for <strong>68.4% of Nigeria’s total petroleum export value</strong>.</p><p>This was an interesting finding because it showed that a relatively small group of regular partners accounted for a substantial share of export value.</p><p>It also taught me the importance of choosing the correct measurement when working with repeated records. Counting distinct months was more appropriate than simply counting rows.</p><h3>8. Analysing Bi-directional Trade</h3><p>Another question I explored was:</p><p><strong>Which countries both import petroleum from Nigeria and export petroleum to Nigeria?</strong></p><p>This is known as bi-directional trade.</p><p>I used a Pandas pivot table to display imports and exports side by side for each trading partner.</p><pre>countries = pd.pivot_table(<br>    df_new,<br>    index=&#39;Partner&#39;,<br>    columns=&#39;Trade Flow&#39;,<br>    values=&#39;Trade Value (US$)&#39;,<br>    aggfunc=&#39;sum&#39;<br>)</pre><p>The pivot table produced one row per partner, with separate columns for imports and exports.</p><p>Some countries had a value in only one direction, while others had values in both columns.</p><p>I used dropna() to keep only the countries with recorded trade in both directions.</p><pre>countries_both = countries.dropna()</pre><pre>countries_both</pre><p>The result identified <strong>29 partners with bi-directional trade</strong>.</p><p>Some of the countries included Brazil, Canada, China, Côte d’Ivoire, France, Germany, India, Indonesia, Italy, Malaysia, the Netherlands and Spain.</p><p>The Netherlands and India were among the largest two-way trading partners by combined value.</p><p>However, the presence of trade in both directions does not necessarily mean that the values are balanced. For example, Canada purchased approximately $3.5 billion in petroleum exports from Nigeria, while Nigeria’s recorded imports from Canada were only about $56,326.</p><p>This helped me understand that bi-directional trade describes the direction of trade, not equality in trade value.</p><h3>9. Analysing Monthly Import and Export Trends</h3><p>I wanted to understand how Nigeria’s petroleum trade changed throughout 2023.</p><p>Specifically, I wanted to identify:</p><ul><li>The month with the highest import value.</li><li>The month with the highest export value.</li><li>Whether there was an apparent monthly trend.</li></ul><h3>Monthly imports</h3><p>I grouped the import records by period and calculated the total trade value.</p><pre>petrol_grouping_import = (<br>    df_new[df_new[&#39;Trade Flow&#39;] == &#39;Import&#39;]<br>    .groupby(&#39;Period&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>)</pre><pre>print(petrol_grouping_import)</pre><p>The results showed that January had the highest import value, at approximately <strong>$2.33 billion</strong>.</p><p>June had the lowest import value, at approximately <strong>$872 million</strong>.</p><h3>Monthly exports</h3><p>I repeated the same operation for exports.</p><pre>petrol_grouping_export = (<br>    df_new[df_new[&#39;Trade Flow&#39;] == &#39;Export&#39;]<br>    .groupby(&#39;Period&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>)</pre><pre>print(petrol_grouping_export)</pre><p>December recorded the highest export value, at approximately <strong>$6.22 billion</strong>.</p><p>The monthly results showed that exports were fairly steady from January to August, generally between $3.0 billion and $4.0 billion. From September, exports increased each month, reaching approximately $6.22 billion in December.</p><h3>Visualising monthly trade trends</h3><p>A line chart is useful for comparing changes over time.</p><p>I prepared the monthly data in calendar order and used Seaborn to visualise the trends.</p><pre>import seaborn as sns<br>import matplotlib.pyplot as plt</pre><pre>monthly_trade = (<br>    df_new.groupby([&#39;Period&#39;, &#39;Trade Flow&#39;])[<br>        &#39;Trade Value (US$)&#39;<br>    ]<br>    .sum()<br>    .reset_index()<br>)</pre><pre>monthly_trade[&#39;Value (US$ billions)&#39;] = (<br>    monthly_trade[&#39;Trade Value (US$)&#39;] / 1_000_000_000<br>)</pre><pre>monthly_trade[&#39;Month&#39;] = pd.to_datetime(<br>    monthly_trade[&#39;Period&#39;].astype(str),<br>    format=&#39;%Y%m&#39;<br>).dt.strftime(&#39;%b %Y&#39;)</pre><pre>plt.figure(figsize=(12, 6))</pre><pre>sns.lineplot(<br>    data=monthly_trade,<br>    x=&#39;Month&#39;,<br>    y=&#39;Value (US$ billions)&#39;,<br>    hue=&#39;Trade Flow&#39;,<br>    marker=&#39;o&#39;<br>)</pre><pre>plt.title(&#39;Nigeria Petroleum Imports and Exports, 2023&#39;)<br>plt.xlabel(&#39;Month&#39;)<br>plt.ylabel(&#39;Trade value (US$ billions)&#39;)<br>plt.xticks(rotation=45)<br>plt.grid(True)<br>plt.tight_layout()<br>plt.show()</pre><p>The monthly analysis showed that imports fluctuated throughout the year, with no clear seasonal pattern.</p><p>Exports, however, showed a noticeable increase towards the end of the year.</p><p>I was careful not to describe this as a confirmed seasonal trend. Since the dataset covers only one year, more years of data would be needed to establish whether the same pattern repeats.</p><p>This was an important lesson in data interpretation: an observed pattern is not automatically proof of a recurring trend.</p><h3>10. Exploring Grouping and Aggregation with the UK Milk Dataset</h3><p>After completing the individual Nigeria analysis, I also worked through the UK milk and cream dataset.</p><p>This allowed me to practise the same concepts using a different dataset.</p><p>The UK dataset covered January to May 2015 and included imports and exports for processed and unprocessed milk and cream.</p><h3>Identifying regular UK export partners</h3><p>For this exercise, the dataset contained two commodity categories and five months.</p><p>I defined a regular customer as a country that imported both categories in all five months.</p><p>That meant each regular partner needed ten records: two commodity categories multiplied by five months.</p><pre>def buysEveryMonth(group):<br>    return len(group) == 10</pre><pre>grouped = exports.groupby(&#39;Partner&#39;)</pre><pre>regular = grouped.filter(buysEveryMonth)</pre><p>The notebook’s analysis showed that 16 countries met this definition.</p><p>Together, these regular customers accounted for approximately <strong>75.8% of total UK milk and cream exports</strong>.</p><p>This was another useful example of how the definition of a regular customer depends on the structure of the dataset and the question being asked.</p><h3>Identifying countries with bi-directional trade</h3><p>I used a pivot table to compare imports and exports for each UK trading partner.</p><pre>countries = pd.pivot_table(<br>    milk,<br>    index=&#39;Partner&#39;,<br>    columns=&#39;Trade Flow&#39;,<br>    values=&#39;Trade Value (US$)&#39;,<br>    aggfunc=&#39;sum&#39;<br>)</pre><pre>countries_both = countries.dropna()</pre><p>This identified 20 countries with recorded imports and exports in both directions.</p><p>Ireland, France, the Netherlands, Germany and Belgium were among the leading partners by combined trade value.</p><p>The exercise helped reinforce the use of pivot tables when comparing two or more categories for the same group.</p><h3>11. Data Visualisation with Seaborn and Matplotlib</h3><p>Data analysis does not end with calculating totals.</p><p>It is also important to communicate findings clearly. This is where data visualisation becomes useful.</p><p>During this part of the assignment, I used Seaborn and Matplotlib to create five visualisations from the UK milk and cream dataset.</p><h3>Plot 1: Comparing total imports and exports</h3><p>The first question was:</p><p><strong>Did the UK export more milk and cream than it imported between January and May 2015?</strong></p><p>I grouped the data by trade flow and calculated the total value in millions of US dollars.</p><pre>plot_df = milk.copy()</pre><pre>plot_df[&#39;Value (US$ millions)&#39;] = (<br>    plot_df[&#39;Trade Value (US$)&#39;] / 1_000_000<br>)</pre><pre>totals = (<br>    plot_df.groupby(&#39;Trade Flow&#39;)[&#39;Value (US$ millions)&#39;]<br>    .sum()<br>    .reset_index()<br>)</pre><pre>sns.barplot(<br>    data=totals,<br>    x=&#39;Trade Flow&#39;,<br>    y=&#39;Value (US$ millions)&#39;,<br>    hue=&#39;Trade Flow&#39;,<br>    legend=False<br>)</pre><pre>plt.title(&#39;UK Milk Exports vs Imports, Jan-May 2015&#39;)<br>plt.xlabel(&#39;Trade flow&#39;)<br>plt.ylabel(&#39;Trade value (US$ millions)&#39;)<br>plt.tight_layout()<br>plt.show()</pre><p>The results showed approximately $265 million in exports and $156 million in imports.</p><p>The UK therefore recorded a positive trade balance for the selected milk and cream products during the period.</p><h3>Plot 2: Top five UK export partners</h3><p>The second visualisation identified the five largest destinations for UK milk and cream exports.</p><pre>exports = milk[<br>    milk[&#39;Trade Flow&#39;] == &#39;Exports&#39;<br>]</pre><pre>top_5 = (<br>    exports.groupby(&#39;Partner&#39;)[&#39;Trade Value (US$)&#39;]<br>    .sum()<br>    .sort_values(ascending=False)<br>    .head(5)<br>)</pre><pre>sns.barplot(<br>    x=top_5.values / 1_000_000,<br>    y=top_5.index<br>)</pre><pre>plt.title(&#39;Top 5 Countries the UK Exports Milk To&#39;)<br>plt.xlabel(&#39;Trade value (US$ millions)&#39;)<br>plt.ylabel(&#39;Country&#39;)<br>plt.tight_layout()<br>plt.show()</pre><p>Ireland was the leading export destination by a considerable margin, followed by the Netherlands, China, Germany and China, Hong Kong SAR.</p><p>A horizontal bar chart made the country names easier to read and compare.</p><h3>Plot 3: Top five bi-directional trading partners</h3><p>For the third plot, I wanted to identify the five countries with the largest combined import and export values.</p><p>I first created a pivot table and removed countries without trade in both directions.</p><pre>countries = pd.pivot_table(<br>    milk,<br>    index=&#39;Partner&#39;,<br>    columns=&#39;Trade Flow&#39;,<br>    values=&#39;Trade Value (US$)&#39;,<br>    aggfunc=&#39;sum&#39;<br>)</pre><pre>milk_countries = countries.dropna().copy()</pre><pre>milk_countries[&#39;Total&#39;] = (<br>    milk_countries[&#39;Exports&#39;] +<br>    milk_countries[&#39;Imports&#39;]<br>)</pre><pre>top_5 = milk_countries.sort_values(<br>    &#39;Total&#39;,<br>    ascending=False<br>).head(5)</pre><p>I then plotted imports and exports side by side.</p><pre>top_5[[&#39;Exports&#39;, &#39;Imports&#39;]].plot(<br>    kind=&#39;bar&#39;<br>)</pre><pre>plt.title(<br>    &#39;Top 5 Countries the UK Both Imports From and Exports To&#39;<br>)</pre><pre>plt.xlabel(&#39;Country&#39;)<br>plt.ylabel(&#39;Trade value (US$)&#39;)<br>plt.xticks(rotation=0)<br>plt.tight_layout()<br>plt.show()</pre><p>The leading countries were Ireland, France, the Netherlands, Germany and Belgium.</p><p>This visualisation made it easier to compare the direction and value of trade with each partner.</p><h3>Plot 4: Five smallest bi-directional trading partners</h3><p>I also wanted to examine the countries with the smallest combined trade values.</p><p>Rather than sorting the DataFrame in descending order, I sorted it in ascending order.</p><pre>bottom_5_both = (<br>    milk_countries.sort_values(<br>        &#39;Total&#39;,<br>        ascending=True<br>    ).head(5)<br>)</pre><pre>bottom_5_both[[&#39;Exports&#39;, &#39;Imports&#39;]].plot(<br>    kind=&#39;bar&#39;<br>)</pre><pre>plt.title(<br>    &#39;Least 5 Countries the UK Both Imports From and Exports To&#39;<br>)</pre><pre>plt.xlabel(&#39;Country&#39;)<br>plt.ylabel(&#39;Trade value (US$)&#39;)<br>plt.xticks(rotation=0)<br>plt.tight_layout()<br>plt.show()</pre><p>The five countries were Latvia, Slovakia, Austria, Hungary and Lithuania.</p><p>This exercise showed me that sorting can be used not only to identify the largest values but also to investigate the smallest ones.</p><p>It also raised an interesting point: very small trade values may deserve further investigation before drawing conclusions about the importance of a trading relationship.</p><h3>Plot 5: Monthly imports and exports</h3><p>The final visualisation examined how the UK’s milk and cream imports and exports changed between January and May 2015.</p><p>I grouped the data by period and trade flow, then converted the period into readable month labels.</p><pre>group_per_flow = (<br>    milk.groupby([&#39;Period&#39;, &#39;Trade Flow&#39;])[<br>        &#39;Trade Value (US$)&#39;<br>    ].sum()<br>)</pre><pre>monthly = group_per_flow.reset_index()</pre><pre>monthly[&#39;Value (US$ millions)&#39;] = (<br>    monthly[&#39;Trade Value (US$)&#39;] / 1_000_000<br>)</pre><pre>monthly[&#39;Month&#39;] = pd.to_datetime(<br>    monthly[&#39;Period&#39;].astype(str),<br>    format=&#39;%Y%m&#39;<br>).dt.strftime(&#39;%b %Y&#39;)</pre><p>I used a line chart to compare both flows.</p><pre>sns.lineplot(<br>    data=monthly,<br>    x=&#39;Month&#39;,<br>    y=&#39;Value (US$ millions)&#39;,<br>    hue=&#39;Trade Flow&#39;<br>)</pre><pre>plt.title(&#39;UK Milk Imports and Exports by Month&#39;)<br>plt.xlabel(&#39;Month&#39;)<br>plt.ylabel(&#39;Trade value (US$ millions)&#39;)<br>plt.grid(True)<br>plt.tight_layout()<br>plt.show()</pre><p>The monthly results were:</p><p>MonthExports (US$ millions)Imports (US$ millions)January5234February4630March5232April5530May5931</p><p>Exports were higher than imports in every month.</p><p>Exports were lowest in February and highest in May, while imports remained relatively steady.</p><p>This was a good example of how a line chart can reveal changes over time more clearly than a long table of figures.</p><h3>12. Challenges I Faced and How I Overcame Them</h3><p>This week’s assignment involved several practical challenges.</p><p><strong>1. Understanding the Comtrade CSV structure</strong></p><p>The downloaded Comtrade files used different column names from the UK milk dataset.</p><p>For example, the trade value was stored as primaryValue, rather than Trade Value (US$).</p><p>I addressed this by inspecting the column names and renaming the fields I needed.</p><p>I learnt that it is important to inspect a dataset before writing analysis code. Assuming that columns have familiar names can lead to errors.</p><p><strong>2. Combining the crude and refined petroleum files</strong></p><p>I had to confirm that both CSV files had the same column structure before combining them.</p><p>I compared their columns and checked the final row count after using concat().</p><p>This helped me verify that the files had been combined correctly.</p><p><strong>3. Understanding the World records</strong></p><p>The World records initially required some thought.</p><p>I learnt that these records represented totals across all partners. Including them alongside individual countries would inflate my calculated totals.</p><p>Removing them before aggregation ensured that my partner-level analysis was based on individual trading partners.</p><p><strong>4. Identifying regular trading partners</strong></p><p>My first approach to identifying regular customers was to count rows.</p><p>However, this was not reliable because a country could have multiple commodity records in the same month.</p><p>I changed my approach to count distinct months using nunique().</p><p>This was a valuable lesson: the method used to measure something must match the actual definition of the question.</p><p><strong>5. Understanding pivot tables</strong></p><p>I needed to distinguish between grouping records to calculate totals and using pivot tables to compare categories side by side.</p><p>For example, groupby() helped me calculate total trade values, while pivot_table() helped me display imports and exports for each country in separate columns.</p><p><strong>6. Working with SQL window functions</strong></p><p>The ranking exercise was another learning point.</p><p>I had to understand why a window function could not simply be filtered in the same query’s WHERE clause.</p><p>Using a subquery allowed me to calculate the rank first and then filter for the top three partners in the outer query.</p><p>Working through the exercise helped me understand the order in which these SQL operations are applied.</p><h3>13. What I Would Like to Explore Next</h3><p>The analysis answered several questions, but it also raised new ones.</p><p>Some questions I would like to investigate in future projects include:</p><ol><li>How did Nigeria’s petroleum exports and imports change between 2022, 2023 and 2024?</li><li>Are the increases in exports towards the end of 2023 repeated in other years?</li><li>How dependent is Nigeria on its five largest petroleum export partners?</li><li>Which countries have increased or reduced their petroleum trade with Nigeria?</li><li>How do Nigeria’s petroleum trading partners compare across Europe, Asia, Africa and the Americas?</li><li>What is the average trade value per unit or per kilogram for each partner?</li><li>Which countries purchase crude petroleum from Nigeria while supplying refined petroleum to Nigeria?</li><li>How does Nigeria’s petroleum trade compare with its trade in other major commodities?</li></ol><p>These questions would help me move beyond descriptive analysis and explore more detailed trade patterns.</p><h3>14. Key Lessons from Week 8</h3><p>Looking back at this week’s work, there are several lessons I will take forward.</p><p><strong>Data preparation comes first.</strong> Before analysing a dataset, I need to understand its structure, inspect its columns, check for missing values and remove records that could distort the results.</p><p><strong>Grouping makes large datasets easier to understand.</strong> With groupby(), I can summarise hundreds of records by country, commodity, trade flow or month.</p><p><strong>Aggregation helps answer specific questions.</strong> Calculating sums, averages, counts and percentages allows me to move from individual records to useful findings.</p><p><strong>Pivot tables make comparisons easier.</strong> They allow different categories to be displayed side by side, making it easier to compare imports and exports.</p><p><strong>Visualisation improves communication.</strong> A well-labelled chart can make a pattern or comparison easier to understand than a long table of numbers.</p><p><strong>Pandas and SQL complement each other.</strong> I can use both tools to perform similar analytical tasks, although their syntax and approaches differ.</p><p><strong>Interpretation matters as much as calculation.</strong> Finding a pattern is one thing; explaining what it means, what it does not mean and what further investigation is needed is another.</p><h3>Conclusion</h3><p>Week 8 was a practical learning experience that helped me strengthen my understanding of data analysis using Pandas, Seaborn, Matplotlib and SQL.</p><p>I worked with two different international trade datasets, cleaned and combined CSV files, grouped records, calculated trade values, identified major trading partners, analysed monthly trends and created visualisations.</p><p>One of my most interesting findings was from the Nigeria petroleum analysis.</p><p>The selected 2023 dataset showed approximately <strong>$49.09 billion in petroleum exports and $20.11 billion in imports</strong>, giving a trade surplus of approximately <strong>$28.98 billion</strong>.</p><p>I also discovered that the top three import partners accounted for 52.1% of imports, while the top three export partners accounted for 31.7% of exports.</p><p>Another important finding was that ten partner categories appeared in all twelve months of Nigeria’s export records, accounting for 68.4% of total export value.</p><p>The UK milk and cream exercises also helped me practise these concepts from another perspective, especially through monthly comparisons, regular trading partners, bi-directional trade and SQL reporting.</p><p>Beyond the results, I appreciated the opportunity to understand how different tools can work together. Pandas helped me prepare and analyse the data, Seaborn and Matplotlib helped me communicate the findings, and SQL allowed me to perform similar operations directly on database tables.</p><p>Most importantly, I am learning that data analysis is not just about writing code or producing charts. It is about asking meaningful questions, selecting the right methods, interpreting the results carefully and communicating what the data tells us.</p><p>Another week of learning, practising and becoming more confident in my data journey.</p><p><strong>On to the next week!</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=efe970103278" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Dates to Data Stories: My Week of Learning SQL at DataRaflow]]></title>
            <link>https://medium.com/@aderafat/from-dates-to-data-stories-my-week-of-learning-sql-at-dataraflow-97d47841f436?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/97d47841f436</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 22:19:01 GMT</pubDate>
            <atom:updated>2026-10-25T22:19:01.393Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*er5lADCcQLOegYgUNC6_Uw.jpeg" /></figure><p>This week at DataRaflow, I learnt that writing an SQL query is only one part of working with data. Getting the correct result requires understanding the data, applying the right conditions, checking the output, and being willing to investigate when something does not work as expected.</p><p>My practice focused on SQL date conversion, filtering, joining tables, sorting results, and using window functions. These concepts helped me see how SQL can be used to answer practical questions rather than simply retrieve records from a database.</p><h3>Working with Dates That Were Stored as Text</h3><p>One of my tasks involved London’s 2014 weather data. The challenge was to retrieve the maximum and minimum temperatures between 8 and 12 December, inclusive.</p><p>The date column, GMT, was stored as text rather than as a proper date. This meant I needed to convert it before filtering the records by date.</p><p>I used TRY_CONVERT(date, GMT) to convert the values and BETWEEN to select the required date range.</p><pre>SELECT TRY_CONVERT(date, GMT) AS day,<br>       [Max TemperatureC],<br>       [Min TemperatureC]<br>FROM week07.london_2014<br>WHERE TRY_CONVERT(date, GMT)<br>      BETWEEN &#39;2014-12-08&#39; AND &#39;2014-12-12&#39;<br>ORDER BY day;</pre><p>This exercise helped me understand why data types matter. A value may look like a date to a human, but the database may still treat it as text. Converting it to the appropriate data type makes date-based operations clearer and more reliable.</p><p>I also learnt that getting the expected number of rows does not automatically mean a query is correct. When the practice checker rejected my result despite the correct shape, I realised that the actual values, column order, sorting and handling of missing data also matter.</p><h3>Bringing Information Together with JOIN</h3><p>Another task involved combining two World Bank tables: one containing GDP data for 2013 and the other containing life expectancy.</p><p>The goal was to identify ten countries with the highest life expectancy among those with GDP below 10 billion US dollars.</p><p>I used an INNER JOIN on the country column to bring the two datasets together. I then applied a GDP filter, excluded records with unknown life expectancy, converted GDP into billions of US dollars, and rounded the results to the required decimal places.</p><pre>SELECT TOP 10<br>       g.country,<br>       ROUND(g.[NY.GDP.MKTP.CD] / 1000000000.0, 2)<br>           AS gdp_bn_usd,<br>       ROUND(l.[SP.DYN.LE00.IN], 1)<br>           AS life_expectancy<br>FROM week08.wb_gdp_2013 AS g<br>INNER JOIN week08.wb_le_2013 AS l<br>    ON g.country = l.country<br>WHERE g.[NY.GDP.MKTP.CD] &lt; 10000000000<br>  AND l.[SP.DYN.LE00.IN] IS NOT NULL<br>ORDER BY l.[SP.DYN.LE00.IN] DESC,<br>         g.country ASC;</pre><p>This task helped me understand that a join is more than combining tables. The join condition determines how records relate, while the filters and sorting determine which records answer the question.</p><p>I also learnt to pay attention to units. GDP expressed in US dollars must be divided by one billion to produce a value in billions. Rounding the result to two decimal places makes it easier to read, while life expectancy is rounded to one decimal place.</p><h3>Finding the Driest Week with Window Functions</h3><p>The most interesting task involved finding the seven-day period with the least total rainfall in London during the summer of 2014.</p><p>The period had to begin on a date in June, July or August and include that date and the following six days. If multiple periods had the same rainfall total, the earliest start date had to be selected.</p><p>This introduced me to the practical use of SQL window functions.</p><p>The expression below calculates a rolling rainfall total for each starting date:</p><pre>SUM(rain) OVER (<br>    ORDER BY day<br>    ROWS BETWEEN CURRENT ROW AND 6 FOLLOWING<br>)</pre><p>Unlike a normal aggregate that combines values into one result, a window function calculates a value for each row while retaining the individual rows.</p><p>I used a common table expression (CTE) to organise the query into stages. The first stage selected the summer weather records and converted the dates. The next calculated the seven-row rainfall totals. The final stage selected a complete seven-day window and ordered the results from the lowest rainfall total to the highest, using the start date to break ties.</p><p>One challenge was ensuring that the window contained seven rows. Another was understanding the difference between counting rows and counting non-null rainfall values. COUNT(*) counts rows, while COUNT(rain) counts only rows where the rainfall value is not null.</p><p>My first attempts returned the expected number of rows and columns, but the checker still rejected the result. Rather than assuming that the query was correct because its shape matched, I began inspecting the underlying results to investigate the mismatch.</p><p>That experience reminded me that window functions require careful attention to the ordering of rows, the completeness of each window, and the treatment of missing values.</p><h3>The Challenges Behind the Queries</h3><p>Not every task worked on the first attempt. Some of the most useful learning happened when I received an error or a result that did not match the expected answer.</p><p>Three challenges stood out:</p><p><strong>1. Distinguishing a correct shape from a correct answer</strong></p><p>A result can contain the correct number of rows and columns but still have the wrong values or order. I learnt to look beyond the shape and inspect the actual records.</p><p><strong>2. Understanding how missing values affect calculations</strong></p><p>SQL functions do not all handle null values in the same way. For example, SUM() ignores null values, and COUNT(*) behaves differently from COUNT(column). Understanding these differences is essential when working with incomplete datasets.</p><p><strong>3. Debugging systematically</strong></p><p>Repeatedly changing a query without understanding the mismatch can waste time. I started separating the diagnostic query from the submission query so that I could inspect intermediate results without repeatedly submitting an answer to the checker.</p><p>I also learnt to keep SQL strings clean in my Python notebook. Unwanted characters and formatting copied into a query can cause execution errors, even when the SQL logic itself looks reasonable.</p><h3>LL: Lessons Learnt This Week</h3><p>Here are my key lessons from this week’s practice:</p><ul><li><strong>Understand the data types first.</strong> Dates stored as text may need conversion before date filtering.</li><li><strong>Read the question carefully.</strong> Details such as inclusive date ranges, units, decimal places and tie-breaking rules can change the expected answer.</li><li><strong>Use joins deliberately.</strong> Identify the correct relationship between tables before combining them.</li><li><strong>Understand window frames.</strong> ROWS BETWEEN CURRENT ROW AND 6 FOLLOWING refers to the current row and the next six rows, not automatically to seven consecutive calendar dates.</li><li><strong>Check missing values.</strong> Decide how null values should be handled before relying on counts and totals.</li><li><strong>Inspect the results.</strong> A successful query execution or matching result shape does not prove that the answer is correct.</li><li><strong>Debug one issue at a time.</strong> Small diagnostic queries can make it easier to locate a problem than rewriting an entire query repeatedly.</li></ul><h3>What This Means for My Data Journey</h3><p>As someone transitioning from teaching into data analytics and data science, I am learning that technical skills require patience as much as knowledge.</p><p>Teaching has trained me to break difficult ideas into smaller steps, explain processes clearly and keep trying different approaches when something is not understood. I am finding those same habits useful when debugging SQL.</p><p>I am also beginning to see how these techniques can support my longer-term interest in healthcare analytics. Joining datasets, handling missing values, filtering records and calculating results over time are useful foundations for analysing health indicators, patient records and public health trends.</p><p>I still have more to learn, but this week’s exercises have helped me understand SQL more deeply. My goal is not just to write queries that run. I want to understand why they work, how to validate their results and how to use them to answer meaningful questions.</p><p>Every error is an opportunity to investigate, and every query is another step towards becoming a more confident data professional.</p><p><strong>One query at a time, one lesson at a time.</strong> 📊</p><p>#DataRaflow #SQL #DataAnalytics #DataScience #LearningInPublic #WomenInTech</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=97d47841f436" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 6 at DataRaflow: The Week That Tested My Resilience]]></title>
            <link>https://medium.com/@aderafat/week-6-at-dataraflow-the-week-that-tested-my-resilience-4aa8a2cc16d8?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/4aa8a2cc16d8</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Fri, 23 Oct 2026 07:59:40 GMT</pubDate>
            <atom:updated>2026-10-23T07:59:40.271Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IG0-M-YRqRPH__D-_Ew2iQ.png" /><figcaption>How I looked like at the beginning of week 6</figcaption></figure><p>I thought the previous weeks at DataRaflow were tough, but Week 6 proved to be tougher. Between intermediate SQL, NumPy, Pandas, database connection problems, home tasks and timed assessments, this week tested not just my technical knowledge but also my patience and resilience.</p><p>There were moments when things did not work as expected, and I had to troubleshoot instead of simply moving on to the next task. Thankfully, I got through it. Looking back, I can say that this was one of those weeks that reminded me why consistency, problem-solving and resilience matter when learning data analysis.</p><p>Here is a reflection on what I learnt, the challenges I faced, how I worked through them and one strategy that helped me prepare for the timed quiz.</p><h3>1. What I Learnt</h3><h3>Intermediate SQL for Data Analysis</h3><p>This week, I worked through <em>Intermediate SQL for Data Analysts</em>. The lessons took me beyond basic queries and introduced more ways to retrieve, transform and analyse data.</p><p>Some of the concepts I studied included:</p><ul><li>Exploring database tables and selecting relevant columns.</li><li>Filtering records using WHERE.</li><li>Handling missing values with NULL.</li><li>Converting data types using CAST and TRY_CONVERT.</li><li>Working with date, string and mathematical functions.</li><li>Using CASE expressions to create categories.</li><li>Performing aggregations with COUNT, SUM, AVG, MIN and MAX.</li><li>Understanding GROUP BY and HAVING.</li><li>Combining tables using joins and combining query results with set operators.</li><li>Writing subqueries and Common Table Expressions (CTEs).</li><li>Learning about window functions, PIVOT and UNPIVOT.</li><li>Understanding good practices when working with SQL and Python together.</li></ul><p>One important lesson was the difference between WHERE and HAVING. Although both filter data, WHERE filters individual rows before grouping, while HAVING filters groups after aggregation.</p><p>I also learnt how joining tables on non-unique keys can produce more rows than expected. This is sometimes called <em>row explosion</em>, and it can affect calculations if the results are not checked carefully.</p><p>Another interesting detail was integer division. In SQL Server, 7 / 2 returns 3, while 7 / 2.0 returns 3.5. This matters when calculating averages, ratios and percentages.</p><p>These concepts reminded me that writing a query is only part of the work. I also need to understand what the query is doing and whether the result makes sense.</p><h3>NumPy: Working with Arrays</h3><p>I continued practising NumPy and learnt more about working with numerical data in Python.</p><p>The topics included creating one-dimensional, two-dimensional and three-dimensional arrays; reshaping arrays with reshape(); changing data types with astype(); and combining arrays using concatenate() and stack().</p><p>I also practised splitting arrays with array_split(), searching sorted arrays with searchsorted(), and creating arrays from binary data using frombuffer().</p><p>Other concepts included checking for missing values with isnan(), checking complex and real numbers with iscomplex() and isreal(), identifying scalars with isscalar(), and understanding how the axis parameter affects array operations.</p><p>One distinction I paid attention to was the difference between concatenate() and stack(). concatenate() joins arrays along an existing axis, while stack() combines arrays along a new axis.</p><p>I also learnt how reshape() changes the shape of an array and how using -1 allows NumPy to infer a dimension automatically when the number of elements permits it.</p><h3>Pandas: Working with Series and DataFrames</h3><p>With Pandas, I practised creating a Series from a list, converting a Series into a DataFrame, converting data into lists or arrays, sorting values and checking data types.</p><p>I learnt to distinguish between a Series, which is one-dimensional, and a DataFrame, which is two-dimensional and organised into rows and columns.</p><p>I also practised using methods such as head() to inspect the first few rows of a DataFrame and iloc to select data by its integer position.</p><p>These are important skills because before analysing a dataset, I need to understand its structure, inspect its contents and identify the information relevant to the question I want to answer.</p><h3>2. Preparing for the Timed Quiz: 40 AI-Generated Practice Questions</h3><p>One thing I want to highlight from this week is how I prepared for the timed quiz.</p><p>With so many concepts to cover, I knew that simply reading through the course materials might not be enough. I wanted to test my understanding before attempting the actual assessment, so I used <strong>ChatGPT and Claude to generate 40 practice questions</strong> based on the topics I was studying.</p><p>I read through the questions and used them to revise the concepts, test myself and identify areas where I needed more practice. This gave me an opportunity to engage with the material actively rather than just read it passively.</p><p>The practice questions became an important part of my preparation. They helped me revisit the topics and approach the actual quiz with more confidence.</p><p>The result was encouraging: <strong>I scored 96 out of 100 after attempting the 100-question quiz.</strong></p><p>I was pleased with this result because it showed me the value of preparing deliberately before an assessment. The questions did not replace the course materials or my own understanding; they helped me practise what I had learnt and check where I still needed improvement.</p><p>This experience also taught me to use AI tools more intentionally. Generating questions was useful, but I still had to read, think through the answers, check corrections and understand the concepts behind them.</p><p>My goal is to graduate among the top 10 students in the DataRaflow cohort. Preparing with practice questions is one strategy I intend to continue using, alongside consistent study and hands-on practice.</p><h3>3. The Technical Challenges That Tested Me</h3><h3>The .env File Error</h3><p>One of the first problems I encountered was an error related to my .env file.</p><p>At first, I had to troubleshoot to understand what was wrong. After investigating, I discovered that there were trailing spaces after some of the values in the file. Those extra spaces were causing errors.</p><p>It was a small detail, but it took attention to identify. Once I removed the trailing spaces, I resolved that particular problem.</p><p>This reminded me that debugging does not always involve complicated code. Sometimes, the cause of an error is something as simple as an extra space in a configuration file. It is important to check the small details instead of assuming the problem must be complex.</p><h3>The Database Connection Issue</h3><p>Even after fixing the .env file, I still had connection issues with the Azure SQL database.</p><p>This time, the problem was not the trailing spaces I had already corrected. The database connection was being affected by IP access restrictions. I needed my IP address to be approved before I could connect successfully.</p><p>I reached out for help, and <strong>DataRaflow resolved the IP connection issue</strong>. This allowed me to move forward with the work.</p><p>The experience helped me understand that troubleshooting requires separating one problem from another. Fixing the configuration file did not automatically resolve the database connection issue because the two problems had different causes.</p><p>It also reminded me that asking for support when necessary is part of solving technical problems. I could investigate what was within my control and seek help when the issue required access or intervention from the programme team.</p><h3>SQL Errors and Unexpected Results</h3><p>The SQL exercises brought their own learning moments.</p><p>For example, I learnt that checking for missing values requires IS NULL, not = NULL. I also encountered a column name in the London dataset that had a leading space: [ Events].</p><p>This showed me why it is important to inspect the actual database schema rather than assume column names are written exactly as expected.</p><p>I also learnt to be careful with placeholder table names. A query using week06.some_table will not work unless that table actually exists. I need to identify the correct table before running the query.</p><p>These experiences helped me become more deliberate about checking the data, verifying names and understanding error messages.</p><h3>4. An Additional Task: The Niagara Succession Signal Founder Challenge</h3><p>Alongside my regular Week 6 learning, I also worked on the <em>Niagara Succession Signal</em> Founder Challenge.</p><p>This involved retrieving data from Statistics Canada through its Web Data Service (WDS) API and bringing it into Python.</p><p>Unlike working with a dataset that has already been provided, using an API involves obtaining data from an external source and preparing it for further work. This introduced me to another part of the data workflow: data collection.</p><p>It also gave me an opportunity to connect my Python learning with a practical task involving an external data source.</p><p>Although this challenge was separate from my main Week 6 home tasks, it added to an already demanding week and reinforced how different technical skills can come together in a data project.</p><h3>5. How I Got Through the Week</h3><p>This week required a lot of troubleshooting, revision and persistence.</p><p>For the .env issue, I investigated the error, found the trailing spaces and corrected them. For the database connection problem, I recognised that the remaining issue was different and sought support from DataRaflow.</p><p>For the SQL exercises, I paid closer attention to syntax, table structures, column names and the behaviour of different functions.</p><p>For the timed quiz, I prepared by reading through 40 practice questions generated with AI tools. This helped me test my understanding before attempting the assessment.</p><p>I also learnt that breaking a problem into smaller steps can make it easier to solve. Rather than changing several things at once, it is better to identify the error, investigate its likely cause, test a solution and check whether the result is correct.</p><p>Electricity disruptions over the weekend also affected my ability to study as consistently as I wanted. Balancing several tasks under those circumstances was not easy, but I kept working through the material.</p><p>Above all, <strong>resilience got me through this week</strong>. I did not understand everything immediately, and not every problem was something I could resolve alone. But I kept investigating, practising and asking for help when necessary.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hi78hx1_mCEnob-ixehl5g.png" /><figcaption>How I looked like after reading, understanding and solving week 6 tasks</figcaption></figure><h3>6. My Biggest Takeaway from Week 6</h3><p>I thought the previous weeks were tough, but this week was tougher. It pushed me technically and tested how well I could manage several demands at once.</p><p>I learnt more about SQL, NumPy and Pandas, but I also learnt lessons that go beyond the technical content. I learnt to check small configuration details, distinguish between different causes of errors, use AI tools as study assistants and ask for help when an issue requires it.</p><p>The 96/100 quiz result was an encouraging highlight, especially because I had taken time to prepare with practice questions beforehand. It showed me that deliberate revision can make a difference.</p><p>I am also beginning to understand how the tools I am learning fit together. SQL helps me retrieve and summarise data, NumPy supports numerical operations, and Pandas helps me inspect and manipulate tabular data. Working with APIs adds another layer by showing me how data can be collected from external sources.</p><p>There is still a lot to learn, but I am grateful that I got through a week that initially felt overwhelming. I am learning that progress does not always mean getting everything right the first time. Sometimes, progress means staying with a problem long enough to understand it and continuing even when the work becomes difficult.</p><p>My goal remains to finish among the top 10 students in the DataRaflow cohort. Week 6 was intense, but it also reminded me that resilience, preparation and a willingness to learn from mistakes will remain essential throughout this journey.</p><p><strong>Have you ever had a week of learning that tested you more than you expected? What helped you push through it?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4aa8a2cc16d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learning Under Pressure: My Week 6 Experience at DataraFlow]]></title>
            <link>https://medium.com/@aderafat/learning-under-pressure-my-week-6-experience-at-dataraflow-0b82ae77c050?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/0b82ae77c050</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Mon, 19 Oct 2026 22:09:24 GMT</pubDate>
            <atom:updated>2026-10-19T22:09:24.527Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u3ygmoBtFSIEQjMSOvVa8A.jpeg" /><figcaption>How I looked like at the beginning of week 6.</figcaption></figure><p>Week 6 at DataraFlow started with a lot of activity because there were so many tasks to complete. To make matters more challenging, our quiz was timed, and the score now contributes to our overall assessment.</p><p>I found myself wondering, <em>How do I learn these topics, prepare well enough to pass the quiz, and still have time to complete the home task?</em></p><p>I knew that simply reading through the materials might not be enough. I needed a way to practise, test my understanding and identify the areas where I needed improvement.</p><p>My solution was to use AI agents as study assistants. I would copy a topic into an AI tool and ask it to generate at least 40 practice questions. I would answer the questions myself, check the corrections and revisit the areas where I made mistakes.</p><p>I continued practising with this method, and when I attempted a set of 100 questions, I missed only four. Scoring 96 out of 100 in my practice was encouraging. It showed me how useful active practice could be when preparing for an assessment.</p><p>However, I also learnt that AI-generated questions should not replace the original learning materials. It is important to check the questions and corrections to make sure they are accurate and relevant to what we have been taught.</p><h3>Working Through the Week 6 Home Task</h3><p>The Week 6 home task gave me the opportunity to put several concepts into practice, particularly in NumPy, Pandas and SQL.</p><h3>Strengthening my NumPy skills</h3><p>I worked on creating and examining arrays, checking for missing values, identifying complex and real numbers, comparing values and checking how much memory an array occupies.</p><p>One thing I learnt was that similar functions can perform different tasks. For example, np.equal() checks whether values are exactly equal, while np.isclose() checks whether they are sufficiently close within a specified tolerance.</p><p>Understanding these differences is important because choosing the wrong function can affect the results of an analysis.</p><h3>Practising data manipulation with Pandas</h3><p>I also worked with Pandas Series and DataFrames. The exercises involved creating Series, converting them into lists or arrays, performing arithmetic operations, changing data types, sorting values and adding data to an existing Series.</p><p>These exercises helped me understand how Pandas can be used to organise, inspect and manipulate data before analysis.</p><p>I also learnt the importance of data types. When values contain a mixture of numbers and text, I need to check whether they can be used directly for numerical calculations or whether they require conversion.</p><h3>Exploring datasets and practising SQL</h3><p>Another part of the week’s work involved querying datasets from the DataraFlow cohort database. I practised retrieving tables using SQL and examining the resulting data in Python.</p><p>The SQL tasks covered filtering records, handling missing values, grouping data, joining tables, combining datasets, using conditional statements, writing common table expressions and applying window functions.</p><p>One important lesson was that SQL uses IS NULL rather than = NULL to identify missing values. I also became more aware of how integer division can affect calculations when decimal results are expected.</p><p>Beyond learning the syntax, I began to appreciate the importance of checking whether a query answers the actual question and whether its output makes sense.</p><h3>An Additional Challenge: The Founder Challenge</h3><p>Alongside the Week 6 home task, my team and I also worked on the DataraFlow Founder Challenge, <em>The Niagara Succession Signal</em>.</p><p>My task included pulling data through an API so that we could work with it in Python. This gave me an opportunity to explore another part of the data science workflow: collecting data from an external source rather than relying only on datasets already available in a notebook or database.</p><p>I had to understand how an API request works, how the response is returned and how the data can be brought into Python for further processing.</p><p>This experience reminded me that data analysis does not begin with building charts or training models. First, we need to find the right data, access it correctly and understand its structure.</p><p>It also connected with my Week 6 learning. Whether I retrieve data using SQL or an API, I still need to inspect the dataset, understand its columns and prepare it before analysis.</p><h3>The Challenges and How I Managed Them</h3><p>The biggest challenge was balancing the timed quiz, the home task and my contribution to the Founder Challenge. There was a lot to learn, and I had to find a way to make progress without becoming overwhelmed.</p><p>Using AI-generated practice questions helped me prepare for the quiz more actively. Instead of reading a topic repeatedly, I could test myself, identify mistakes and concentrate on the areas that needed more attention.</p><p>Breaking the programming tasks into smaller parts also helped me understand the concepts one at a time. Rather than focusing only on getting an answer, I tried to understand what each function or SQL statement was doing.</p><p>The API task introduced another learning experience because data collection requires a different set of steps from working with a dataset that is already available. It requires understanding the source, making the request, handling the response and checking the resulting data.</p><p>Looking back, I could improve my approach by starting my revision earlier, keeping a record of questions I missed and documenting my data collection process more carefully.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T5191CuXewFq8CFu2f9zwA.jpeg" /><figcaption>How I look like after reading, understanding and solving week 6 tasks</figcaption></figure><h3>What I Am Taking Away from Week 6</h3><p>This week taught me that learning effectively is not just about spending more time with the materials. It is also about finding methods that help me test my understanding and learn from my mistakes.</p><p>I strengthened my understanding of NumPy and Pandas, practised SQL queries and explored API-based data collection through the Founder Challenge. Each activity helped me see how different skills fit into the broader data science workflow.</p><p>My experience with the practice questions was particularly encouraging. Missing only four out of 100 questions showed me that active recall and reviewing corrections can be useful ways to prepare for assessments.</p><p>There is still a lot to learn, but I am beginning to understand that progress comes from consistent practice, asking questions and being willing to revisit concepts I do not fully understand.</p><p>Week 6 was demanding, but it reminded me that even when the workload feels overwhelming, having a structured approach can make it more manageable.</p><p>One topic, one practice question and one task at a time — that is how I am learning to keep moving forward.</p><p>#DataraFlow #DataScience #Python #NumPy #Pandas #SQL #LearningJourney</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0b82ae77c050" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From the Classroom to the Code: Week 5 at DataRaflow]]></title>
            <link>https://medium.com/@aderafat/from-the-classroom-to-the-code-week-5-at-dataraflow-685d7cba67ea?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/685d7cba67ea</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 14:46:27 GMT</pubDate>
            <atom:updated>2026-10-13T14:46:27.424Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lcY_vrAvOI7KxIPuiwYs6A.jpeg" /></figure><p><em>How a teacher learning data science began exploring whether AI can explain why students are at risk of falling behind.</em></p><p>I have spent years in the classroom, teaching students and watching them learn at different speeds. Some understand a topic almost immediately. Others need more time, a different explanation, or a little extra support before things begin to make sense.</p><p>That experience has made me curious about a question: <strong>what if data could help teachers identify students who need support before it is too late?</strong></p><p>This was the thinking behind my Week 5 research work at DataRaflow. I started developing a proposal titled <em>“Do Early-Warning Explanations Travel? Cross-Context Stability of Explainable Academic Risk Models.”</em></p><p>The title may sound like a lot, but the idea is straightforward. If an AI model predicts that a student is at risk of failing or dropping out, can we trust that prediction? And if we move the model to a different course or group of students, will it still work, and will the reasons behind its predictions remain consistent?</p><p>Welcome to Week 5, where my teaching experience met my growing interest in machine learning research.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0ogCJM5x59mODnWBgtKQ6A.jpeg" /></figure><h3>Why This Problem?</h3><p>My background is in Computer Science, but I have also spent more than ten years teaching ICT at the secondary-school level. Moving into data science has given me an opportunity to look at education from a different angle.</p><p>Teachers make decisions every day about where students need help. We use test scores, classroom participation, attendance and other observations to understand how students are progressing.</p><p>Machine learning offers another way to examine these patterns. A model can learn from historical student data and estimate which students may be at risk of an undesirable academic outcome.</p><p>But there is a catch.</p><p>A model that performs well in one school, course or academic session may not perform as well in another. Even when its predictions remain reasonably accurate, the factors influencing those predictions may change.</p><p>For example, a model might rely heavily on online learning activity in one course but place more importance on assessment performance in another.</p><p>That made me realise that asking whether a model is accurate is only part of the question. I also want to know whether its behaviour remains reliable when the context changes.</p><h3>Why Accuracy Is Not the Whole Story</h3><p>Imagine building a model that predicts whether a student is likely to drop out.</p><p>You test it and discover that it makes good predictions. Great! But suppose a teacher asks, “Why has this student been flagged?”</p><p>Can the model provide a useful explanation?</p><p>And what happens if the same model is used for another group of students? Does it identify similar warning signs, or does it suddenly rely on different features?</p><p>This is where Explainable Artificial Intelligence, or XAI, comes in.</p><p>Explainable AI refers to methods that help us understand how a machine learning model arrives at its predictions. One technique I plan to explore in my research is SHAP (SHapley Additive exPlanations).</p><p>SHAP helps estimate how individual features contribute to a model’s prediction. It can show which features are influential across a dataset and how different features contribute to an individual prediction.</p><p>However, there is an important distinction: <strong>a feature that influences a model’s prediction is not necessarily the cause of the outcome.</strong></p><p>If a model identifies assessment performance as an important predictor of academic risk, that does not automatically mean poor assessment performance is the root cause of a student’s difficulties.</p><p>As a teacher, I find this distinction particularly important. An AI system should help educators investigate where support may be needed, not replace their judgement or label students permanently.</p><h3>The Research Gap I Want to Explore</h3><p>I did not want to choose a research topic simply because machine learning is popular. I wanted a question that could be investigated systematically.</p><p>Existing research has examined student-success prediction and explainable learning analytics, including the stability of explanations across different contexts. So I am not claiming that explainability or explanation stability is a completely new research area.</p><p>Instead, my proposal focuses on a narrower question:</p><p><strong>When an academic-risk model is tested in a different educational context, how do its predictive performance and explanations change together? Does measuring explanation stability tell us anything that ordinary performance metrics miss?</strong></p><p>This is the gap I intend to investigate further through my literature review.</p><p>I find this question interesting because a model can appear successful when judged by its predictive performance alone, while changes in the reasons behind its predictions may reveal something important about how it behaves in a new setting.</p><p>The goal is not to prove in advance that explanation stability improves a model. It is to investigate whether measuring it provides useful additional information.</p><h3>My Idea in Three Steps</h3><p>My proposed study will use the Open University Learning Analytics Dataset (OULAD), a publicly available dataset containing anonymised student records, assessment information, course presentations and virtual learning environment activity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*glgZoSjmfEEHm4lB2ihY2w.png" /></figure><p>Here is how I plan to approach the research.</p><h3>1. Prepare the data</h3><p>First, I will inspect the dataset, understand its variables and identify an academic outcome that can be defined consistently across the contexts being compared.</p><p>I will also examine missing values, class imbalance and the suitability of the available features.</p><p>One issue I need to avoid is data leakage. If I want to predict whether a student is at risk at an early stage, I cannot use information that would only become available later. For example, using final assessment results to predict an outcome at a point before those results exist would give the model an unfair advantage.</p><h3>2. Train and test three models</h3><p>I propose using three machine learning algorithms:</p><ul><li><strong>Logistic Regression:</strong> a relatively interpretable baseline for classification.</li><li><strong>Random Forest:</strong> a model that can capture nonlinear relationships and interactions between features.</li><li><strong>XGBoost:</strong> a gradient-boosting algorithm suited to structured data.</li></ul><p>I will train the models on selected course presentations and test them on a different presentation that was not used to develop the models.</p><p>This will allow me to compare performance within a familiar context against performance in a held-out context.</p><p>I will also need to ensure that the training and testing strategy prevents information from the test context from influencing model selection.</p><h3>3. Compare performance with explanation stability</h3><p>Once the models have been evaluated, I plan to use SHAP to examine their explanations.</p><p>I will compare the rankings of important features, measure how much the most influential features overlap between contexts, and investigate changes in feature contributions where meaningful comparisons are possible.</p><p>For model performance, I propose using metrics such as precision, recall, F1-score, ROC-AUC and PR-AUC, alongside probability calibration.</p><p>The interesting part will be comparing the findings.</p><p>Does performance remain stable while explanations change? Do both deteriorate? Or do the models behave consistently across the selected contexts?</p><p>I do not know the answers yet. That is what the experiments are intended to establish.</p><h3>The Part That Made Me Think Differently</h3><p>One of the most important lessons from developing this proposal is that a model can be good at predicting an outcome without necessarily providing a complete or dependable picture of the situation.</p><p>Imagine two models that perform similarly on a test dataset. One consistently identifies a similar group of important features across the contexts being compared, while the other produces substantially different feature rankings.</p><p>That difference deserves investigation.</p><p>It does not automatically mean the second model is wrong. The contexts may genuinely differ, or the features may behave differently in each group. But it does mean that predictive performance alone may not tell the full story.</p><p>This is why I want to study both dimensions rather than assume that a high performance score is enough.</p><p>Another lesson is that an explanation can be stable without being correct, meaningful or useful to a teacher. Stability is something to evaluate, not a guarantee of quality.</p><p>For me, this is an important part of learning to approach machine learning research critically.</p><h3>My Greatest Challenges</h3><h3>The research topic had to be specific enough</h3><p>Academic performance prediction is already a well-studied field. Explainable AI is also receiving attention in educational research.</p><p>I had to move beyond a broad topic such as using machine learning to predict student failure and identify a question that could be investigated more precisely.</p><p>The solution was to focus on the relationship between predictive performance and explanation stability when the educational context changes.</p><p>I also learnt to be careful about claiming a research gap before reviewing the relevant studies thoroughly.</p><h3>The technical concepts are connected, but not interchangeable</h3><p>Predictive performance, calibration, feature importance and explanation stability all tell us different things.</p><p>For instance, a model may rank students reasonably well but produce probabilities that do not accurately reflect their actual risk. Another model may perform well but provide explanations that vary across contexts.</p><p>I therefore had to think about which evaluation measures answer which research questions instead of treating every metric as another way to measure accuracy.</p><h3>Connecting the research to Nigeria</h3><p>This is probably the part that matters most to me personally.</p><p>My long-term interest includes using data science to support education in Nigeria. However, the dataset I propose to use comes from UK higher education.</p><p>I cannot assume that patterns learned from UK university students will apply to Nigerian secondary-school pupils.</p><p>Our curricula, assessment practices, school resources, attendance records and access to digital learning systems may differ considerably.</p><p>The approach I have taken is to treat the study as an initial investigation into model reliability, not as proof that the resulting models can be deployed in Nigerian schools.</p><p>A future study would need appropriate Nigerian school data, local validation, attention to fairness and privacy, and input from teachers and other stakeholders.</p><h3>A Beginner’s Glossary</h3><p>Here are some of the technical terms I encountered while developing the proposal, explained in simpler language.</p><ul><li><strong>Machine learning:</strong> Teaching a computer to learn patterns from data and use them to make predictions.</li><li><strong>Classification:</strong> Predicting a category, such as whether a student belongs to a defined academic-risk group.</li><li><strong>Explainable AI (XAI):</strong> Methods that help us understand how a model arrives at its predictions.</li><li><strong>SHAP:</strong> A technique for estimating how individual features contribute to a model’s prediction.</li><li><strong>Feature:</strong> A piece of information used by a model, such as previous assessment performance or recorded learning activity.</li><li><strong>Cross-context evaluation:</strong> Testing a model in a different setting from the one used to train it.</li><li><strong>Data leakage:</strong> When information that would not legitimately be available at prediction time influences model training or evaluation.</li><li><strong>Calibration:</strong> How closely predicted probabilities correspond to observed outcome frequencies.</li><li><strong>Baseline model:</strong> A reference model used to determine whether a more sophisticated approach provides meaningful improvement.</li><li><strong>Explanation stability:</strong> How consistently a model’s explanations behave when the evaluation context changes.</li></ul><h3>What I Am Not Claiming</h3><p>I think it is important to be honest about the limitations of a research idea, especially when working with AI.</p><p>First, I have developed the proposal, but the proposed experiments still need to be conducted. I do not yet have results showing which algorithm performs best or how stable its explanations will be.</p><p>Second, OULAD represents selected UK higher-education courses. Testing across its course presentations will provide evidence about those contexts, not proof of transferability to other countries.</p><p>Third, SHAP describes how a model uses features; it does not establish that those features cause academic failure or dropout.</p><p>Finally, even a reliable prediction does not prove that an intervention based on it will improve a student’s outcome. That requires further research involving educators, students and the educational system.</p><p>These limitations do not make the research less valuable. They help define what the study can reasonably contribute.</p><h3>What Is Next?</h3><p>My next steps are to review the relevant literature, inspect OULAD and confirm that the data support the planned comparisons.</p><p>I will then define the prediction target, prepare the data, establish the evaluation strategy, train the selected models and compare their performance across contexts.</p><p>After that, I can use SHAP to investigate the stability of their explanations and determine whether those measurements add useful information to the performance results.</p><p>I also want to continue building my practical skills in Python and machine learning so I can move from describing a research method to implementing and evaluating it.</p><p>The longer-term motivation remains close to my teaching experience: understanding how data-driven tools might help educators identify students who need support, while recognising that local evidence and human judgement must remain central.</p><h3>Final Thoughts</h3><p>I started this journey to develop my data science skills, but each research question is teaching me to think more carefully about the problems I want to solve.</p><p>Week 5 reminded me that machine learning is not simply about selecting an algorithm, training it and reporting a score. It is about understanding the data, designing fair tests, interpreting results and being honest about uncertainty.</p><p>As a teacher transitioning into data science, I am interested in technology that can support better decisions without losing sight of the people affected by those decisions.</p><p>My biggest takeaway this week is this:</p><p><strong>A model should not only be evaluated by how well it predicts. We should also investigate how reliably it behaves, what drives its predictions and whether its explanations remain useful when the context changes.</strong></p><p>I am still learning, and the real work of implementing this study lies ahead. But I am excited to turn this proposal into a practical research project, one step at a time.</p><p>If you work in data science, education or machine learning research, I would love to hear your thoughts: <strong>when evaluating an AI model, how much importance should we give to the stability of its explanations alongside its predictive performance?</strong></p><h3>Read my research proposal</h3><p>I have developed a research proposal titled <em>Do Early-Warning Explanations Travel? Cross-Context Stability of Explainable Academic Risk Models.</em></p><p>It outlines the research questions, proposed methodology, evaluation measures and relevance to future research in Nigerian education.</p><p><a href="https://docs.google.com/document/d/1Kc2OScKi3zMor2ZF5vrmLtAeRjzOcL923hqVTTMs8ig/edit?usp=sharing"><strong>Read the full research proposal here.</strong></a></p><p>#DataRaflow #DataScience #MachineLearning #ExplainableAI #SHAP #EducationalDataMining #LearningInPublic</p><h3>References</h3><ul><li>Kuzilek, J., Hlosta, M., &amp; Zdrahal, Z. (2017). <a href="https://doi.org/10.1038/sdata.2017.171">Open University Learning Analytics dataset</a>. <em>Scientific Data, 4</em>, 170171.</li><li><a href="https://doi.org/10.1016/j.dss.2024.114229">Explainable Learning Analytics: Assessing the Stability of Student Success Prediction Models by Means of Explainable AI</a>. (2024). <em>Decision Support Systems, 182</em>, 114229.</li><li><a href="https://doi.org/10.1109/ACCESS.2025.3643645">Systematic Literature Review on Explainable Learning Analytics and Educational Data Mining</a>. (2025). <em>IEEE Access, 13</em>, 214387–214408.</li><li>Alfredo, R., et al. (2024). <a href="https://doi.org/10.1016/j.caeai.2024.100215">Human-centred learning analytics and AI in education: A systematic literature review</a>. <em>Computers and Education: Artificial Intelligence, 6</em>, 100215.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=685d7cba67ea" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 4: From Weeks of Python to a Research Question]]></title>
            <link>https://medium.com/@aderafat/week-4-from-weeks-of-python-to-a-research-question-91bbe4b67dd3?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/91bbe4b67dd3</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Sun, 04 Oct 2026 21:05:38 GMT</pubDate>
            <atom:updated>2026-10-04T21:05:38.372Z</atom:updated>
            <content:encoded><![CDATA[<p>I got really interested in DataraFlow when I learned we would be taught how to write research, and that we would be required to propose a research topic. After weeks of learning Python, I can now create a project that I can document academically and, hopefully, publish.</p><p>That sounded exciting. It also turned out to be harder than I expected.</p><h3>A topic is not a research idea</h3><p>My first instinct was to look for a topic that sounded impressive. Something with “machine learning” and “Africa” in the title. This week taught me that a good title is not clever. It should signal the variables and the context, so a reader knows what you are studying and where.</p><p>The harder part is the problem behind the title. A research topic has to answer questions like these:</p><ul><li>What exactly is the problem?</li><li>Who is affected, and why does it matter?</li><li>What do we already know, and what is still missing?</li></ul><h3>The shape of a research proposal</h3><p>The lesson walked us through the structure of a proposal: title, abstract, introduction, literature review, problem statement, research questions, objectives, methodology, significance, limitations, timeline and references.</p><p>A few things changed how I think about them.</p><p><strong>The abstract is written last.</strong> It is the whole proposal in miniature, so you can’t write it before you know what the proposal says.</p><p><strong>The literature review has a job.</strong> It isn’t a list of everything I have read. It exists to justify the study by building toward the gap, so the reader reaches the conclusion that no one has yet examined this particular question.</p><p><strong>The problem statement is the gap, stated plainly.</strong> If I can’t say it in a short, clear paragraph, I don’t understand it yet.</p><p><strong>Research questions must be narrow.</strong> A good one is narrow enough that you can imagine what its answer would look like. “Can we improve health outcomes?” is a wish. “Which household factors are most strongly associated with a child missing all routine vaccinations?” is a question.</p><p><strong>Methodology is the longest and most scrutinised section.</strong> This is where you explain your design, your data, your sampling, your instruments, your analysis and your ethics. It’s also where I realised how much I still have to learn: train and validation splits, cross-validation, evaluation metrics, statistical testing and reproducibility.</p><h3>My working topic</h3><p>The topic I’m working towards is:</p><p><strong>Who Is Being Missed? Mapping Zero-Dose Children in Nigeria with Interpretable, Spatially Aware Machine Learning</strong></p><p>Zero-dose children are children who have not received even the first dose of a routine vaccine. Nigeria is consistently among the countries with the highest numbers. The practical question is where they are and what they have in common, so that outreach can reach them.</p><p>The questions I want to investigate are:</p><ol><li>Which household, maternal and community factors are most strongly linked to zero-dose status?</li><li>Can a machine learning model identify high-risk children or areas better than a simple logistic regression baseline, and does it stay fair across regions and wealth groups?</li><li>Does a model trained on an earlier survey round still work on a later one?</li></ol><p>I plan to use public survey data from the Demographic and Health Surveys, which means registering and requesting access first. The plan is to start with a simple baseline, then compare it with models like Random Forest and XGBoost, and explain the results so that someone who is not a data scientist can understand them.</p><h3>Why this topic</h3><p>There are two reasons.</p><p>The first is that I’m interested in health data, and I want my projects to move in that direction. Public health data lets me practise real skills, like handling survey weights, choosing metrics for imbalanced outcomes and checking whether a model is fair, on a problem that matters.</p><p>The second comes from my background. I spent more than ten years teaching ICT, and teaching taught me to explain things clearly. That skill matters in research, because a model that nobody can understand or trust is unlikely to change anything.</p><h3>What this week really taught me</h3><p>I came into Week 4 thinking I was learning how to write a research paper. I think I was learning how to think like a researcher: to question assumptions, to find out what has already been done, to define a problem clearly and to design a method before jumping to a solution.</p><p>I also learned that I’m not an expert yet. I haven’t finished my literature review. My methodology is still a draft. I’m still learning concepts like cross-validation and statistical testing. But for the first time, I can see a path from the Python I’ve been learning to something I can document, defend and, I hope, publish.</p><p>So I have a research direction. Now the real work begins.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91bbe4b67dd3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week 3: Modules, Dates, JSON, and Environments ‍]]></title>
            <link>https://medium.com/@aderafat/week-3-modules-dates-json-and-environments-475a851e30d7?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/475a851e30d7</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Fri, 25 Sep 2026 10:30:53 GMT</pubDate>
            <atom:updated>2026-09-25T10:30:53.216Z</atom:updated>
            <content:encoded><![CDATA[<p>Three weeks into the DataraFlow Data Science training now, and this week was less about new syntax and more about actually breaking things and having to figure out why. Week 3 covered modules, dates, math, JSON, pip, try/except, virtual environments, basic file handling, and a first taste of SQL. Here’s a breakdown of what I learned in each area, and the specific mistakes I made and fixed along the way.</p><h3>1. Modules: import module vs. from module import function</h3><h3>What I Learned</h3><ul><li>Used import math, import datetime, import json, and import random to bring in Python&#39;s standard library tools for different tasks.</li><li>Accessed functions and constants using dot notation: math.sqrt(144), math.factorial(6), datetime.date.today().</li><li>Learned that import module brings in the whole module as one object, so everything inside it stays attached to that module&#39;s name.</li><li>Learned that from module import function pulls a single function out directly, so it can be called on its own without the module prefix.</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> I assumed import math and from math import sqrt gave me the same ability to just call sqrt(25) directly, without thinking about where each import style actually places the function.</li><li><strong>The Fix:</strong> I worked through both side by side and saw exactly why one needs the prefix and the other doesn’t:</li></ul><pre>import math<br>print(math.sqrt(25))     # works — sqrt lives inside math</pre><pre>print(sqrt(25))           # NameError — sqrt was never pulled into my namespace</pre><pre>from math import sqrt<br>print(sqrt(25))           # works — sqrt was imported directly, no prefix needed</pre><pre>print(math.sqrt(25))      # NameError — math itself was never imported this time</pre><p>The difference is about <em>which namespace</em> the function lands in, not just a style preference.</p><h3>2. Dates: Working with the datetime Module</h3><h3>What I Learned</h3><ul><li>Used datetime.datetime.now() to get the current date and time, and strftime() with format codes like %Y, %m, %B, and %d to display specific parts.</li><li>Used datetime.date.today() and datetime.date(year, month, day) to build specific date objects.</li><li>Learned that subtracting two date objects returns a timedelta object, and that .days extracts the actual number of days from it.</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> In my birthday countdown program, I subtracted in the wrong direction:</li></ul><pre>today_date = datetime.date.today()<br>birthday = datetime.date(year, month, day)</pre><pre>age_diff = today_date - birthday<br>print(age_diff.days)      # -92 — negative, for a birthday still months away</pre><ul><li><strong>The Fix:</strong> I traced through what the subtraction actually meant: today_date - birthday calculates &quot;days <em>since</em> the birthday,&quot; not &quot;days <em>until</em> it.&quot; Flipping the order fixed it:</li></ul><pre>age_diff = birthday - today_date<br>print(age_diff.days)      # 92 — positive, exactly what I wanted</pre><pre>if age_diff.days &lt; 0:<br>    # birthday already passed this year — roll it to next year<br>    birthday = datetime.date(year + 1, month, day)</pre><h3>3. Math: The math Module and Its Limits</h3><h3>What I Learned</h3><ul><li>Used math.sqrt(), math.factorial(), and math.pi for square roots, factorials, and a high-precision value of pi (accurate to about 15 decimal places).</li><li>Used math.floor() (always rounds down) and math.ceil() (always rounds up).</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> I assumed the math module would have some built-in way to round a number to a specific number of decimal places — something like math.round(x, 2). It doesn&#39;t exist:</li></ul><pre>import math<br>math.round(5.768, 2)      # AttributeError — math has no round() function</pre><ul><li><strong>The Fix:</strong> round() is a separate Python builtin, not part of math at all. To round 5.768 to two decimal places using only math functions, I built it manually:</li></ul><pre>number = 5.768<br>rounded = math.floor(number * 100 + 0.5) / 100<br>print(rounded)             # 5.77</pre><p>Multiplying by 100 shifts the digits I care about into whole-number territory, adding 0.5 makes flooring behave like rounding instead of truncating, and dividing by 100 shifts it back.</p><h3>4. JSON: Serializing and Storing Structured Data</h3><h3>What I Learned</h3><ul><li>Used json.dumps() and json.loads() to convert Python objects to and from JSON <strong>strings</strong>.</li><li>Used json.dump() and json.load() to write to and read from JSON <strong>files</strong> directly.</li><li>Learned that both Python list and tuple convert to a JSON array when serialized.</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> Storing a list of student records to a file went through two broken attempts. First, looping and writing dictionaries directly:</li></ul><pre>with open(&quot;students.json&quot;, &quot;w&quot;) as f:<br>    for record in students:<br>        f.write(record)     # TypeError — write() only accepts strings, not dicts</pre><p>Then, calling json.dump() with a variable that was never defined, sitting outside the with block entirely:</p><pre>with open(&quot;students.json&quot;, &quot;w&quot;) as f:<br>    f.write(record)</pre><pre>json.dump(students, file_json)   # NameError — file_json doesn&#39;t exist,<br>                                  # and the file is already closed anyway</pre><ul><li><strong>The Fix:</strong> Moved json.dump() inside the with block and used f — the actual open file object with creates — instead of an undefined variable:</li></ul><pre>with open(&quot;students.json&quot;, &quot;w&quot;) as f:<br>    json.dump(students, f)</pre><pre>with open(&quot;students.json&quot;, &quot;r&quot;) as f:<br>    data = json.load(f)<br>    print(data)</pre><h3>5. PIP, Try/Except, and Virtual Environments</h3><h3>What I Learned</h3><ul><li>Used pip install package_name to install third-party packages, and pip list to confirm what&#39;s installed.</li><li>Used try/except blocks with specific exception types — ValueError, ZeroDivisionError — to handle bad input and division errors separately, rather than catching everything the same way.</li><li>Created and activated a virtual environment with python -m venv &lt;name&gt; and the matching activation command, to keep a project&#39;s packages isolated from my global Python setup.</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> In one exercise, I tried to validate integer input by checking the type <em>after</em> calling input() — which always returns a string, so this check failed every single time, even for genuinely valid input:</li></ul><pre>user_input = input(&quot;Type in a number: &quot;)</pre><pre>if not type(user_input) is int:   # always True — input() is ALWAYS a str<br>    raise Exception(&quot;Only integers allowed&quot;)</pre><ul><li><strong>The Fix:</strong> Converted directly with int(input(...)) inside a try block and let Python&#39;s own conversion raise the error for me:</li></ul><pre>try:<br>    user_input = int(input(&quot;Type in a number: &quot;))<br>    print(f&quot;You entered {user_input}&quot;)<br>except ValueError:<br>    print(&quot;Only integers are allowed&quot;)</pre><ul><li><strong>A second error, with virtual environments:</strong> I created an environment with python -m venv firstproject, but almost activated it with a generic command copied from an example:</li></ul><pre>python -m venv firstproject<br>venv\Scripts\activate      # fails — there is no folder called &quot;venv&quot;</pre><p>The fix was matching the activation path to the actual folder name I chose:</p><pre>firstproject\Scripts\activate</pre><h3>6. Read &amp; Write Files: Handling File Modes Correctly</h3><h3>What I Learned</h3><ul><li>Used with open(filename, mode) as f: to open files safely, so they close automatically even if an error happens inside the block.</li><li>Learned the difference between file modes: &quot;r&quot; requires the file to already exist, &quot;w&quot; overwrites the file&#39;s entire contents the moment it opens, and &quot;a&quot; appends to the end instead.</li></ul><h3>The Error &amp; Fix</h3><ul><li><strong>The Error:</strong> In an early file-handling exercise, I tried to write to a file opened with no mode specified — which defaults to read-only:</li></ul><pre>f = open(&quot;demofile.txt&quot;)   # defaults to &quot;r&quot; — read-only<br>f.write(&quot;lorum ipsum&quot;)     # io.UnsupportedOperation — can&#39;t write in read mode</pre><ul><li><strong>The Fix:</strong> Opened it explicitly in write mode instead:</li></ul><pre>with open(&quot;demofile.txt&quot;, &quot;w&quot;) as f:<br>    f.write(&quot;lorum ipsum&quot;)   # works — &quot;w&quot; mode allows writing</pre><p>I learned to always ask “does this file need to already exist, get overwritten, or get appended to?” before choosing a mode, rather than assuming the default would work for whatever I wanted to do.</p><h3>7. Introduction to SQL: Querying Structured Data</h3><h3>What I Learned</h3><ul><li>Learned that full table names in a database are typically written as schema.table, and that column names containing spaces need to be wrapped in square brackets, like [Max TemperatureC].</li><li>Learned that GROUP BY produces one summary row per category — for example, grouping readings by month to calculate a monthly average instead of one row per individual reading.</li><li>Learned that table aliases (FROM games AS g JOIN players AS p) exist purely to avoid retyping long table names throughout a query, not to change what the query returns.</li></ul><pre>SELECT LEFT(GMT, 7) AS month, AVG([Max TemperatureC]) AS avg_temp<br>FROM weather.london_readings<br>GROUP BY LEFT(GMT, 7)</pre><h3>Hands-On Practice: Putting It All Together</h3><p>To reinforce all of this, I worked through a full practice set covering:</p><ul><li><strong>JSON File Handling:</strong> Storing a list of student records to a file and reading it back with json.dump() and json.load().</li><li><strong>A Math Quiz:</strong> Generating 5 random arithmetic questions with random.randint() and random.choice(), comparing rounded answers to avoid floating-point mismatches, and tracking a running score.</li><li><strong>An Error Handling System:</strong> Dividing 100 by user input, with separate except blocks for ZeroDivisionError and ValueError.</li><li><strong>A Date Calculator:</strong> Calculating days remaining until a birthday, correctly handling the case where the birthday has already passed this year by rolling it forward.</li><li><strong>PIP and Virtual Environments:</strong> Documenting the full sequence of creating an environment, activating it, installing numpy, and verifying it with pip list.</li></ul><p>Working through these made it clear that the hard part was never the syntax, it was reasoning through <em>why</em> a piece of code produced the wrong (or no) result, and fixing the actual logic rather than guessing at a patch.</p><h3>Conclusion</h3><p>This practice set turned a handful of concepts I could technically describe (modules, dates, JSON, error handling, virtual environments) into things I have actually broken and fixed myself. The negative day count, the undefined variable, the type check that could never succeed: each of those forced me to slow down and reason about what the code was really doing, instead of assuming it worked because it looked right. This is the part I am carrying forward into the next stage of this journey.</p><p>Are you learning to code too? 👩🏽‍💻 I’d genuinely like to know: <strong>what’s a bug 🐛 you fixed recently that taught you more than getting it right the first time would have? 💡</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=475a851e30d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Using Classes Blindly to Truly Understanding Them: My OOP Learning Journey]]></title>
            <link>https://medium.com/@aderafat/from-using-classes-blindly-to-truly-understanding-them-my-oop-learning-journey-cc3025d36381?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc3025d36381</guid>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Sun, 20 Sep 2026 17:09:21 GMT</pubDate>
            <atom:updated>2026-09-20T17:09:21.232Z</atom:updated>
            <content:encoded><![CDATA[<p>How turning off Copilot, and reasoning through problems myself, finally made object-oriented programming click — as a data analyst learning to code the hard way.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ILyhSMn-3B3HVHZ4" /></figure><p>I’ve used classes to solve problems in Python before, but this task cemented my understanding of the concepts behind them. Now, when I see a question relating to classes, I can relate to it and work through it with confidence.</p><p>It all started as part of DataraFlow’s curriculum — the second week’s task, the assignments, and the overall project: a Library Management System. I began the week by reading through DataraFlow’s course content, then cemented it with a 2+ hour Python course on YouTube teaching classes.</p><p>One of our tasks was to work through Python classes and object-oriented programming concepts.</p><p>The take-home task covered several areas, including:</p><p>* Classes and objects<br>* `__init__()` and `__str__()`<br>* Inheritance<br>* Parent, child, base, and derived classes<br>* Method overriding<br>* `super()`<br>* Iterators and the iterator protocol<br>* `__iter__()` and `__next__()`<br>* `StopIteration`<br>* Polymorphism<br>* Abstraction<br>* `ABC` and `<a href="http://twitter.com/abstractmethod">@abstractmethod</a>`<br>* Variable scope<br>* `global` and `nonlocal`</p><p>Looking at that list now, I can say, *“Okay, let’s solve it.”*</p><p>But that wasn’t how I started.</p><p>I started with a little fear</p><h4>Watching and reading wasn’t the same as understanding</h4><p>I did what most people do first: I found a 2+ hour Python OOP course on YouTube and watched the whole thing. I read the content given to us, but the second I opened my own blank VS Code file and tried to write a class from scratch, I froze. Watching someone else solve a problem and being able to solve it yourself are two completely different skills, and I hadn’t built the second one yet.</p><h4>The turning point: turning off Copilot</h4><p>The first real shift happened when I disabled GitHub Copilot in VS Code. It sounds small, but it changed everything. Before that, I’d start typing class Animal: and Copilot would auto-suggest the rest of the structure before I&#39;d even had to think about it. I was &quot;writing&quot; code, but I wasn&#39;t <em>learning</em> it — I was just accepting suggestions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XYqbR7gZsWEjf_9mngOymA.png" /><figcaption><strong>I deliberately turned off Copilot so I could practise solving the problems myself.</strong></figcaption></figure><p>Once I turned it off, every line had to come from my own head. It was slower. It was more frustrating. I made more mistakes. But those mistakes are exactly what taught me — debugging my own broken __init__ method or fixing a subclass that wasn&#39;t overriding a method correctly stuck with me in a way that an auto-completed answer never could have.</p><h4>Using AI as a sparring partner, not a solution machine</h4><p>Here’s where I found the balance that actually worked: instead of asking AI tools like Claude or ChatGPT to solve my assignments for me, I used them differently. When I hit a question I found genuinely hard — say, a task on iterators or scope — I’d ask the AI to generate a <em>similar</em> question, with different names and a different scenario, but testing the exact same concept.</p><p>Then I’d solve <em>that</em> one myself first.</p><p>This turned out to be one of the best decisions I made. If I understood the AI’s version but got stuck, I knew exactly which piece of the concept I was missing. If I could solve the AI’s version confidently, I’d go back to my original assignment and suddenly it wasn’t intimidating anymore — it was just the same shape of problem wearing a different name. Practicing on a parallel problem before returning to the real one gave me a kind of rehearsal space, without ever letting an AI just hand me the finished answer.</p><h4>Where it clicked</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/967/1*a-nIezjhiG1v1g8cSiljnA.png" /><figcaption>The concepts that once intimidated me are now problems I can calmly break down and attempt.</figcaption></figure><p>The real “aha” moment came while building a Library Management System from scratch — books, members, borrowing rules, private and protected attributes, custom iterators, all of it. I had to combine everything at once: Book and EBook/PrintedBook subclasses overriding a shared method, StudentMember and TeacherMember enforcing different borrowing limits through polymorphism, a private __library_name that could only be touched through getter/setter methods, and a custom __iter__ so I could loop over books and borrowed items directly.</p><p>The first version I wrote had bugs — a borrowing limit that didn’t actually stop anything, a check that referenced the wrong class’s limit, a missing attribute that would’ve crashed the program the moment someone searched for available books. But finding and fixing those bugs myself, by actually reading my own logic line by line, was the moment OOP stopped being abstract vocabulary and started being a tool I could reach for.</p><h4>From scared to calm</h4><p>I started this having genuine anxiety around classes, polymorphism, and abstraction. Now, when I see a problem that touches these concepts, I don’t panic, I get calm. I already have a mental model for how to approach it: identify the base class, figure out what varies between subclasses, decide what needs to stay protected or private, and write the method that lets each object handle its own version of “doing the thing.” That shift — from dread to quiet confidence — is the real outcome of this whole process, more than any single line of code.</p><h4>I am not done learning</h4><p>I don’t want to pretend that I have mastered Python because I completed one challenging task. I haven’t. There are still concepts I need to practise.There are still problems that will make me stop and think.There will definitely be bugs that make me question my life choices for a few minutes. 😄But something has changed. I have moved from:</p><blockquote><strong><em>“I don’t know how to solve this.”</em></strong></blockquote><p>to:</p><blockquote><strong><em>“I think I understand what this problem is testing. Let me work through it.”</em></strong></blockquote><p>That shift matters. Especially because I am building my career in <strong>data analytics, data science and AI</strong>. I am not learning Python just so I can write Python code. I am learning Python because I want to use technology and data to <strong>understand problems, find patterns and build solutions</strong>.And that requires more than memorising syntax. It requires the confidence to approach unfamiliar problems.</p><h3>A thumbs up to DataraFlow 👍</h3><p>I started this part of my journey with some concepts that genuinely intimidated me. Now I can encounter a problem involving polymorphism, abstraction, inheritance, iterators or scope and think:</p><blockquote><strong><em>“Okay. I know how to approach this.”</em></strong></blockquote><p>I may not always get it right on the first attempt.But I am willing to attempt it. I am willing to make mistakes. I am willing to ask questions. And I am willing to try again.</p><p>That is why I want to give a big thumbs up to <strong>DataraFlow</strong> for creating a challenging environment where learning means more than simply watching lessons and moving on. The challenges are pushing me to think. And I can see myself growing.</p><p><strong>I am growing.</strong></p><p>And perhaps that is the most important lesson I am taking away from this experience:</p><blockquote><strong><em>Confidence doesn’t always come before the challenge. Sometimes, the challenge is what builds the confidence.</em></strong></blockquote><h3>Your turn</h3><p><strong>What is one technical concept that used to intimidate you, but today, when you see it in a problem, you can calmly say:</strong></p><blockquote><strong><em>“I know how to approach this.”</em></strong></blockquote><p>I’d love to hear about it in the comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc3025d36381" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Change Hits Hard, Great Leaders Rise: My Leadership Lessons from WTF Soft Skill Training]]></title>
            <link>https://medium.com/@aderafat/when-change-hits-hard-great-leaders-rise-my-leadership-lessons-from-wtf-soft-skill-training-4fe57a27fcad?source=rss-f2026239515------2</link>
            <guid isPermaLink="false">https://medium.com/p/4fe57a27fcad</guid>
            <category><![CDATA[leading-through-growth]]></category>
            <category><![CDATA[leading-through-change]]></category>
            <dc:creator><![CDATA[rafat adebanjo]]></dc:creator>
            <pubDate>Fri, 21 Nov 2025 16:21:12 GMT</pubDate>
            <atom:updated>2025-11-21T16:21:12.815Z</atom:updated>
            <content:encoded><![CDATA[<p>Today, I had the privilege of attending a deeply insightful leadership training titled <strong>“Leading Teams Through Change, Growth, and Accountability”</strong> led by <strong>Jane Odonwodo</strong>, Founder of Workflow Consult.</p><p>✨ It was one of those sessions that quietly but powerfully shift how you see yourself as a leader.</p><figure><img alt="The 2 Faces of Change" src="https://cdn-images-1.medium.com/max/1024/1*sohL9ElaCxcdlPGLiS68jA.png" /></figure><h3>🌗 The Two Faces of Change</h3><p>Every change comes with <strong>risks</strong> and <strong>opportunities</strong>.</p><p><strong>🌱 Opportunities</strong></p><ul><li>Innovation and fresh thinking</li><li>Better team alignment</li><li>Leadership development</li><li>Stronger culture and cohesion</li><li>Skill growth and resilience</li><li>Greater collaboration and cohesion</li></ul><p><strong>⚠️ Risks</strong></p><ul><li>Confusion and misalignment</li><li>Resistance and low morale</li><li>Communication breakdown</li><li>Loss of productivity</li></ul><blockquote><strong><em>“If change is well-managed, the opportunities that emerge are enormous.”</em></strong><em> — Jane Odonwodo</em></blockquote><p>Change forces us to think constructively and discover strengths we didn’t know we had.</p><figure><img alt="Managing Change and Guiding Teams Through Transitions" src="https://cdn-images-1.medium.com/max/1024/1*Ao4ExKGmZnB7Dm9GBXs8sg.png" /></figure><h3>🧭 How Leaders Guide Teams Through Change</h3><p>People resist change not because they are stubborn, but because they don’t understand it.<br> Jane introduced the <strong>4 W’s — plus two important additions</strong> every leader must communicate clearly.</p><p><strong>1️⃣ Why is it changing?</strong><br> People commit when they understand the reason.<br> At one call center, agents were spending too long on calls — costing millions. Explaining <em>why</em> call-time needed to reduce changed everything.</p><p><strong>2️⃣ What is changing?</strong><br> Clarity removes fear. Even an admin officer told the company was “going digital” needs to understand what the change <em>actually</em> means for their job.</p><p><strong>3️⃣ When is it happening?</strong><br> People hate being kept in the dark. Clear timing reduces anxiety.</p><p><strong>4️⃣ What does it mean for them?</strong><br> Help them see:</p><ul><li>How the change benefits their role</li><li>How it strengthens the company</li><li>What support they will receive</li></ul><p>Many workers fear losing their jobs during digital transformation. A leader’s job is to <strong>explain how the change strengthens, not threatens, their work.</strong></p><p><strong>5️⃣ What support is available?</strong><br> Teams want to know their leaders are <strong>walking through the change with them.</strong></p><p><strong>6️⃣ Insights &amp; feedback</strong><br> Listening closes the loop and makes change smoother.</p><blockquote><em>Leadership isn’t distance — it’s presence.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YX3y7quSgdfVuVgusRi_TQ.png" /></figure><blockquote>“When expectations are unclear, performance becomes subjective and unfair.”</blockquote><p>Here is the 3-step framework we discussed to move from subjective opinions to clear benchmarks:</p><p><strong>1. Define the Target 🎯</strong> <strong>What does success look like?</strong> Before you judge the output, you have to align on the input. If the team doesn’t know exactly what they are aiming for, they can’t hit the bullseye.</p><p><strong>2. The Evaluation Method 📏</strong> <strong>How is it measured?</strong> This is about proof. How do we know we arrived? Move away from “I feel like you did well” to “Here is the data that shows you did well.”</p><p><strong>3. The “What Ifs” ⚖️</strong> <strong>What are the consequences?</strong> We need to be transparent about what happens when expectations aren’t met. Accountability isn’t scary when the rules are known in advance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*HL7MBmvgpc7VMc_Rsi7qKA.png" /></figure><p><strong>Recognition is a Strategy, Not an Afterthought</strong></p><p>Here are the 5 pillars of effective recognition that drive real culture:</p><p><strong>1. Be Timely ⏰</strong> Praise delayed is impact lost. Don’t wait for the quarterly review to tell someone they did a great job. The closer the recognition is to the action, the more powerful the reinforcement.</p><p><strong>2. Be Specific 🎯</strong> Generic “good job” compliments don’t change behavior. We need to be specific. Celebrate the exact behavior you want to see repeated. It shows you are paying attention.</p><p><strong>3. Make it Public 📢</strong> When appropriate, recognize wins publicly. This doesn’t just reward the individual; it reinforces the culture and sets the standard for the rest of the team.</p><p><strong>4. Be Consistent 🔄</strong> Recognition shouldn’t fluctuate with the company’s revenue. It must be consistent. Motivation is needed most when times are tough, not just when business is booming.</p><p><strong>5. Make it Count ✨</strong> Ultimately, motivation comes down to human needs. People stay motivated when they feel <strong>Seen, Safe, and Significant.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0uRaFrAnywI4qrE0hq1i6A.png" /></figure><p>The biggest takeaway? You don’t have to be a superhero with a magic crystal ball. 🔮</p><p>Here’s the golden nugget from the session:</p><blockquote>“Leading teams through CHANGE isn’t about having all the ANSWERS, it’s about giving your people the CLARITY, CONFIDENCE, and ACCOUNTABILITY they need to move forward with you.”</blockquote><p>Instead of pretending to know everything, focus on these three pillars:</p><ul><li><strong>Transparency:</strong> Be open about what you know and what you don’t. 🗣️</li><li><strong>Trust:</strong> Empower your team to take ownership. 🤝</li><li><strong>Fairness:</strong> Reward effort and outcomes consistently. ⚖️</li></ul><p>When you lead this way, your team won’t just survive transitions — they’ll grow through them. 🌱</p><p>What’s your biggest challenge when leading through change? Let’s discuss! 👇</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4fe57a27fcad" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>