<?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 Chichi Okafor on Medium]]></title>
        <description><![CDATA[Stories by Chichi Okafor on Medium]]></description>
        <link>https://medium.com/@jaynechinny?source=rss-11706788e6af------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*rrPWPjOxJx0R_1PVfdP0Rg@2x.jpeg</url>
            <title>Stories by Chichi Okafor on Medium</title>
            <link>https://medium.com/@jaynechinny?source=rss-11706788e6af------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 11:58:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jaynechinny/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[Data Structures and Algorithms in Practice]]></title>
            <link>https://medium.com/@jaynechinny/data-structures-and-algorithms-in-practice-e417153b8382?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/e417153b8382</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[jupyter-notebook]]></category>
            <category><![CDATA[data-structure-algorithm]]></category>
            <category><![CDATA[data-structures]]></category>
            <category><![CDATA[sql]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Fri, 20 Nov 2026 06:26:21 GMT</pubDate>
            <atom:updated>2026-11-20T06:27:01.184Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/540/1*o7XIB6BRCuRf6qe__JEIng@2x.jpeg" /></figure><p>Sometimes when people talk about Data Structures and Algorithms, it can feel like memorizing array tricks and solving coding puzzles for interviews. In Week 11, working through our practice tasks alongside SQL Server (T-SQL) in a Jupyter Notebook helped me see how these ideas can also be useful when working with real data.</p><p>This week, I worked through the algorithmic tasks P1 – P10 and SQL exercises s01 – s04, covering frequency counting, existence checks, prefix sums, running totals, and sliding windows. As an intern still building confidence with SQL and algorithms, some parts took trial and error, but each challenge helped me understand the concepts better.</p><p><strong>What I Learned</strong></p><p>In s01, I worked with frequency counting, similar to using a Counter in Python, to see how often values occur. In s02, I used NOT EXISTS to identify new 2015 UK milk export partners that were not already present in the 2014 export history. This helped me understand how existence checks can be useful when working with related tables.</p><p>For s03, I worked with prefix sums and running totals using SUM(…) OVER and LAG to calculate three-month export totals. By December 2014, the running total reached 898.65 million US$.</p><p>The most challenging task was s04, where I had to find each city’s warmest full seven-day period in 2014 across Beijing, Brasilia, Cape Town, Delhi, London, and Moscow. I combined the city tables with UNION ALL, handled different date columns, calculated seven-day averages using a sliding window, checked that each window contained seven days, and used ROW_NUMBER() to select the warmest period.</p><p>Challenges and Lessons</p><p>Not everything worked on the first attempt. I initially guessed the Week 6 table names and received database errors. Checking INFORMATION_SCHEMA.TABLES showed me the correct names. I also tried nesting window functions, which SQL Server did not allow, so I broke the calculations into separate CTEs. At another point, a failed query left my Jupyter session stuck in a transaction error, and restarting the kernel cleared the problem.</p><p>I also learned to pay closer attention to data types. Using TRY_CONVERT(date, …) helped with text dates, while multiplying by 1.0 or casting values helped avoid problems with integer division. I learned that rounding should generally be left until the end of the calculation.</p><p><strong>Results</strong></p><p>The warmest seven-day periods I calculated were: Beijing – July 17, 29.0°C; Brasilia – October 14, 27.1°C; Cape Town – February 12, 25.9°C; Delhi – August 15, 28.0°C; London – July 15, 22.3°C; and Moscow – July 28, 24.3°C.</p><p><strong>My Main Takeaway</strong></p><p>The biggest lesson from Week 11 was that Data Structures and Algorithms are not only about coding interview questions. Concepts such as frequency counting, prefix sums, and sliding windows can also appear naturally in SQL and data analysis when working with cumulative totals, rolling averages, and trends over time.</p><p>My advice to anyone learning SQL and algorithms is simple: don’t get discouraged when a query fails. Read the full error message, check your table and column names, inspect the data, and break complicated problems into smaller steps. Sometimes the debugging is where the real learning happens.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e417153b8382" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond the Rows: Discovering Trade Patterns Through Python, SQL and Data Visualisation]]></title>
            <link>https://medium.com/@jaynechinny/beyond-the-rows-discovering-trade-patterns-through-python-sql-and-data-visualisation-7a9ff582af4f?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/7a9ff582af4f</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Thu, 29 Oct 2026 12:29:09 GMT</pubDate>
            <atom:updated>2026-10-29T12:29:09.010Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/836/1*9t1EFNFMNP9UFiKivZuwww@2x.jpeg" /></figure><p>From analysing Oman’s dates trade to exploring UK milk imports and exports, how I turned raw datasets into meaningful insights.</p><p>Finding Meaning in Data</p><p>Working with data has taught me that collecting information is only the beginning. The real challenge is understanding what the numbers represent, identifying relationships and communicating findings clearly.</p><p>As I continue my Data Science, Machine Learning and Generative AI internship with DataraFlow, I have been exploring how Python, pandas, SQL and data visualisation can turn large datasets into meaningful comparisons.</p><p>This stage of my learning focused on grouping and aggregating data, analysing international trade, creating visualisations and using SQL to answer analytical questions.</p><p>What made the experience particularly interesting was the opportunity to explore my own dataset alongside the examples provided in the course.</p><p>Understanding Grouping and Aggregation</p><p>One of the key concepts I learnt was split-apply-combine.</p><p>This technique involves dividing data into groups, performing calculations on those groups and combining the results into useful summaries.</p><p>Using pandas, I practised applying groupby() with functions such as sum(), mean(), max() and count().</p><p>For example, grouping international trade data by country helps identify major trading partners, while grouping by month reveals changes in trade activity over time.</p><p>I also explored pivot tables, which allow data to be reorganised for easier comparison.</p><p>These techniques demonstrated how thousands of records can be summarised into smaller tables that answer specific questions.</p><p>Exploring Oman’s Dates Trade</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OTaJt0riHw6hO5juCZXfWA@2x.jpeg" /></figure><p>One of my practical exercises involved obtaining international trade data from the United Nations Comtrade database.</p><p>Rather than limiting myself to the UK’s milk trade example, I selected Oman’s imports and exports of dates in 2023.</p><p>I downloaded the monthly records as a CSV file, loaded them into pandas and examined the dataset’s structure.</p><p>After identifying relevant columns, I renamed fields, separated imports from exports and removed the aggregate World records when comparing individual trading partners.</p><p>This was important because combining World totals with individual country records could result in double-counting.</p><p>I then grouped the data by trading partner and trade flow to identify the countries associated with the highest recorded trade values.</p><p>My analysis showed that Saudi Arabia was the leading source of Oman’s date imports in the selected dataset, while the United Arab Emirates was its leading export destination.</p><p>Other countries, including India and Qatar, also appeared among the major export partners.</p><p>What interested me was how the same commodity could reveal different trading relationships depending on whether I examined imports or exports.</p><p>It also raised further questions about the economic and geographical factors influencing these trade patterns.</p><p>Analysing the UK’s Milk Trade</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hE4U6ar8o9tzRN0j6uoMpA@2x.jpeg" /></figure><p>Another part of the exercise focused on the UK’s international trade in milk and cream.</p><p>The dataset included two commodity classifications: 0401, representing milk and cream that are neither concentrated nor sweetened, and 0402, representing concentrated or sweetened milk and cream.</p><p>Using pandas and SQL, I explored trade values, commodity categories, major trading partners and monthly activity.</p><p>In the 2014 dataset, total exports were approximately US$898.65 million, compared with imports of US$563.88 million.</p><p>Ireland emerged as the leading trading partner in both directions, contributing approximately 46.9% of export value and 30.9% of import value.</p><p>These findings helped me understand why grouping by different variables matters. Overall totals provide one perspective, while country-level and commodity-level comparisons reveal additional patterns.</p><p>Making Data Easier to Understand Through Visualisation</p><p>I also explored data visualisation using Matplotlib and Seaborn.</p><p>The exercises introduced five types of plots: histograms, KDE plots, box or violin plots, regression plots and heatmaps.</p><p>Each serves a different analytical purpose.</p><p>Histograms and KDE plots help examine numerical distributions. Box and violin plots reveal differences in spread and potential outliers. Regression plots help visualise associations between variables, while heatmaps make comparisons across categories easier to interpret.</p><p>Using the UK milk trade data, I created visualisations to examine the available records from different perspectives.</p><p>One important lesson was that visualisation is not simply about producing attractive charts. The choice of chart should depend on the question being investigated.</p><p>A useful visualisation should make information clearer, not more complicated.</p><p>Strengthening My SQL Skills</p><p>SQL formed another important part of the work.</p><p>I practised using GROUP BY to summarise trade data, conditional aggregation to compare imports and exports, and GROUPING SETS to calculate quarterly and annual subtotals.</p><p>I also used window functions to rank trading partners and calculate their percentage contributions to total trade.</p><p>The final exercise involved preparing a tidy table containing the five leading export partners and an Other category for the remaining countries.</p><p>This demonstrated how SQL can prepare data for visualisation without requiring every calculation to be performed separately in pandas.</p><p>After working through the five SQL practice questions, I achieved 5/5.</p><p>However, reaching that result involved more than writing queries. I had to examine errors, review sorting requirements and ensure my results matched the expected structure.</p><p>Challenges and Lessons Learnt</p><p>This experience also came with technical challenges.</p><p>While working in VS Code, I encountered unexpected shutdowns, missing Python modules and database connection errors.</p><p>At one point, I lost notebook changes after VS Code closed unexpectedly, which meant reconstructing parts of my work.</p><p>I also encountered a missing Seaborn package and an SQL practice module that Python could not locate.</p><p>Resolving these issues required checking my file structure, installing the necessary package, correcting filenames and restarting the notebook kernel when necessary.</p><p>Another lesson came from an SQL query that returned the correct values but failed the assessment checker because the months were not arranged chronologically.</p><p>Adding the appropriate ORDER BY clause helped correct the result.</p><p>These experiences reminded me that successful data analysis requires patience, attention to detail and an understanding of the tools involved.</p><p>Going forward, I intend to save my work more frequently, verify my development environment early and make smaller, more deliberate corrections when troubleshooting.</p><p>Final Reflection</p><p>This experience strengthened my understanding of how Python, pandas, SQL and visualisation work together.</p><p>From collecting Oman’s dates trade data to analysing the UK’s milk imports and exports, I learnt how to transform raw information into structured comparisons.</p><p>More importantly, I learnt that data analysis is not only about obtaining answers. It is about asking meaningful questions, checking the reliability of results and recognising the limitations of what the data can tell us.</p><p>As I continue exploring data science and machine learning, I am becoming increasingly interested in how these techniques can be applied to real-world questions.</p><p>Behind every dataset are patterns waiting to be explored, but meaningful insights begin with knowing what to ask.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7a9ff582af4f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring Relationships in Data: Cleaning, Visualisation, Correlation and SQL]]></title>
            <link>https://medium.com/@jaynechinny/exploring-relationships-in-data-cleaning-visualisation-correlation-and-sql-867ca741f151?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/867ca741f151</guid>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-cleaning]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Sat, 24 Oct 2026 02:30:57 GMT</pubDate>
            <atom:updated>2026-10-24T02:30:57.503Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/836/1*eRlXX1_rx2xDzkRF4WnU1Q@2x.jpeg" /></figure><p>Turning raw datasets into meaningful comparisons through Python, pandas, and careful analysis.</p><p>A table of numbers can make a question look easier than it really is.</p><p>We have the data and the tools. Surely, we only need to write a few lines of code and find the answer.</p><p>But what happens when dates are stored as text, measurements are missing, and column names differ from what we expected? What if a calculation produces an answer that the underlying data cannot support?</p><p>These questions shaped my recent work at Datara Flow Academy. Through weather records, World Bank indicators, and SQL exercises, I explored how cleaning, visualisation, and statistical reasoning work together.</p><p>I also discovered how much analysis happens before the final result: inspecting sources, checking assumptions, understanding errors, and deciding whether a comparison is meaningful.</p><p>Understanding the data first</p><p>Using Python and pandas, I began by inspecting rows, columns, data types, and missing values. Methods such as .head(), .info(), and .isna().sum() helped me understand what was available and what needed attention.</p><p>A table’s appearance can be misleading. A column may look numerical while containing text. A date may be readable to a person but unsuitable for chronological calculations.</p><p>In the London weather data, dates needed conversion before I could reliably filter periods and examine changes over time.</p><p>Wind-direction values presented another issue: some contained the unwanted HTML fragment &lt;br /&gt;. Before making numerical comparisons, I needed to remove that text and convert the remaining values.</p><p>The sequence mattered: inspect, clean, convert, and check. Each transformation needed a reason connected to the analysis.</p><p>Handling missing information carefully</p><p>Missing values taught me that one solution does not fit every column.</p><p>Replacing every blank with zero can introduce false information. Removing every incomplete row can also discard useful observations.</p><p>An empty weather-event field may indicate that no event was recorded. A missing wind-gust measurement may mean that the measurement is unavailable. These situations should not automatically receive the same treatment.</p><p>Before making changes, I needed to consider what the column represented and how my decision would affect interpretation.</p><p>This helped me see cleaning as an analytical responsibility. The aim was to make the data usable while preserving its meaning.</p><p>Translating questions into conditions</p><p>Filtering gave me practice turning a question into precise conditions.</p><p>Selecting a date range was relatively straightforward. Identifying northerly winds required more thought because wind direction is circular.</p><p>North sits around both zero and 360 degrees. A condition selecting directions of at least 350 degrees or at most 10 degrees captures observations close to north.</p><p>Without considering how compass directions work, I could have written valid code that selected the wrong records.</p><p>This was a useful reminder that programming and subject knowledge belong together. The computer evaluates the condition I provide; I remain responsible for whether it represents the question.</p><p>Making patterns visible</p><p>I used plots to explore temperature and humidity observations across time.</p><p>Tables provide exact values, while visualisations make changes and unusual observations easier to notice. But a chart needs clear labels, appropriate scales, and correctly ordered dates.</p><p>I also needed to confirm that the selected period matched the task. A polished chart would still be misleading if it displayed the wrong observations.</p><p>Visualisation became a way to examine my work. When a pattern looked unexpected, I could return to the underlying records and investigate.</p><p>That made plotting part of the analysis rather than simply a final presentation step.</p><p>Answering a practical question</p><p>The holiday-weather project connected these techniques to a practical decision: choosing a suitable holiday period.</p><p>Using the London Heathrow 2023 dataset, I compared consecutive fourteen-day periods during summer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u1xhG4dIMDk-G3JmNs7HkA@2x.jpeg" /></figure><p>Looking at an entire period was more useful than identifying one exceptionally hot day. Someone planning a two-week trip needs information about the cooler days within that period too.</p><p>In the data I examined, the warmest fourteen-day summer period was 12 – 25 June, with an average temperature of approximately 20.84°C.</p><p>Rainfall introduced a limitation. The precipitation values in the dataset were all zero. A calculation could therefore make every period appear equally dry.</p><p>That did not justify concluding that London experienced no rainfall throughout the year. Instead, it raised questions about the source or collection process.</p><p>I could report the temperature finding with its context, but the rainfall data did not support a dependable recommendation about the driest period.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jVebn5mTrH9cXosL0WQnaA@2x.jpeg" /></figure><p>This taught me to distinguish a calculated output from a supported conclusion.</p><p>Combining datasets thoughtfully</p><p>The World Bank exercises involved cleaning and combining GDP and life-expectancy tables through a shared country field.</p><p>Before merging, I needed to consider missing measurements and entries representing groups of economies. Mixing regional aggregates with individual countries would weaken the comparison.</p><p>I also examined which observations survived the join.</p><p>An inner join retains matching entries from both tables. A left join preserves entries from the left table and makes unmatched information visible through missing values.</p><p>These operations serve different purposes. One can help prepare comparable observations; the other can help investigate missing matches.</p><p>Checking row counts before and after merging made the consequences clearer. The final table reflected both the available information and the decisions used to combine it.</p><p>Interpreting correlation</p><p>I explored the GDP and life-expectancy relationship using a scatterplot and Spearman’s correlations</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n67VDi1yw0GhRfI9imMKlA@2x.jpeg" /></figure><p>The coefficient in the exercise was approximately 0.50, suggesting a moderate positive association in the analysed data.</p><p>However, the scatterplot showed considerable variation. Countries with similar economic measurements could have different life-expectancy outcomes.</p><p>The coefficient summarised a tendency. It did not explain every country or establish that one variable caused the other.</p><p>Choosing the economic indicator also mattered. Total GDP describes an economy’s size. GDP per capita considers output relative to population, while purchasing-power adjustments address differences in what money can buy.</p><p>This helped me understand that variable selection is part of designing an analysis. The most familiar indicator may not be the best match for the question.</p><p>Developing my own comparison</p><p>For the independent assessment, I chose electricity access and internet use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cOlOS60bJTvhVlJ6U_Jg0A@2x.jpeg" /></figure><p>My question was whether countries with greater access to electricity also tend to have higher internet use.</p><p>I expected a positive association, but that remained a hypothesis to test. Affordability, network coverage, and other conditions could also influence internet adoption.</p><p>I recognised another limitation: electricity access does not necessarily describe supply reliability.</p><p>Planning the comparison required thinking about a shared year, comparable country observations, and missing values.</p><p>It helped me move from following an example towards designing an analysis around a question of my own.</p><p>Applying the concepts in SQL</p><p>The SQL exercises covered cleaning text, converting dates, joining tables, identifying missing matches, and examining seven-day rainfall windows.</p><p>One task involved removing unwanted text from wind-direction values, converting them into integers, and counting observations that met a condition.</p><p>Another required converting the text-based GMT field into a date and selecting observations for 8 – 12 December 2014, inclusive.</p><p>These exercises showed me how the same analytical ideas appear in different tools. Filtering, conversion, and joining still require careful reasoning whether expressed through pandas or SQL.</p><p>I also encountered an “invalid column name” error. The query referred to headers that did not match the database table.</p><p>Guessing alternative names was inefficient. Inspecting the exact database headers was the better next step.</p><p>Understanding what actually failed</p><p>Some challenges came from the working environment.</p><p>VS Code crashes interrupted my progress. Missing modules prevented Python from locating project helpers, while missing environment variables prevented database connections.</p><p>Notebook execution order added another complication. After a kernel restart, code remained visible, but previously created variables were no longer available in memory.</p><p>These errors needed different responses. Syntax problems required code corrections. Import problems required checking project paths. Connection problems required checking setup.</p><p>The habit I want to strengthen is identifying the failing stage before changing anything. Reading the final error message and testing a small example can make the next step clearer.</p><p>I would also inspect headers earlier, organise setup cells more clearly, and check data quality before building recommendations around calculations.</p><p>These exercises have changed the questions I ask while working.</p><p>What does a missing value mean? Which rows disappeared during a join? Does the indicator fit the question? Does the evidence support the conclusion?</p><p>I am learning to make those questions part of my routine. My aim is to produce analyses that another person can understand, check, and reproduce – even when the most useful finding is a limitation in the data.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=867ca741f151" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Working with Python, NumPy, Pandas and SQL: From Data Exploration to Insights]]></title>
            <link>https://medium.com/@jaynechinny/working-with-python-numpy-pandas-and-sql-from-data-exploration-to-insights-c9d4edebd412?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/c9d4edebd412</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[python-pandas]]></category>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[numpy]]></category>
            <category><![CDATA[pandas]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Thu, 15 Oct 2026 14:07:05 GMT</pubDate>
            <atom:updated>2026-10-15T14:07:05.022Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/836/1*-eC-ASi7-ysWfH82WKiiwQ@2x.jpeg" /></figure><p>When I started this assignment, I expected to practise calculations, explore tables, and write queries. I did all of those things, but I also discovered how much attention a data task requires before its results can be trusted.</p><p>A column name can contain an unexpected space. A filter can return nothing because the dataset uses a plural word. A calculation can execute successfully while losing its decimal values. Even a correct query can reveal a pattern that needs further investigation.</p><p>Working with Python, NumPy, Pandas and SQL brought these lessons together. My tasks covered weather observations, customer churn, population data, tuberculosis deaths, milk exports, and loan approvals. Across these different subjects, I kept returning to the same questions: What does the data contain? What exactly am I calculating? What can I reasonably conclude?</p><p>Building a foundation with Python, NumPy and Pandas</p><p>The numerical and data analysis exercises helped me understand how these tools fit together.</p><p>Python provided the programming foundation. NumPy supported work with numerical arrays and calculations, while Pandas allowed me to organise data into DataFrames, inspect columns, count missing values, and calculate summaries.</p><p>As a learner, seeing the output of an operation helped me connect its syntax to its purpose. A DataFrame became more than a table on my screen. It was something I could question, compare, and examine from different directions.</p><p>Working in a Jupyter notebook inside VS Code also taught me about the structure of an analysis. Instructions, code, outputs, and written explanations each had their place. I needed to understand where to enter a solution and how to preserve the supplied setup and checking functions.</p><p>Coming from an English and Literary Studies background, I found something familiar in this process. Close reading matters in programming too. Understanding the wording of a task is often the first step towards choosing the right operation.</p><p>Exploring weather observations from six cities</p><p>One substantial part of the work involved airport weather observations from London, Beijing, Brasilia, Cape Town, Delhi, and Moscow for 2014.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lZcgIw55miMt-15iCLq2uA@2x.jpeg" /></figure><p>Each city’s dataset contained 365 rows and 23 columns. That gave me a consistent starting point, but matching dimensions did not mean every table was equally complete.</p><p>I examined missing values before interpreting the temperature summaries. Columns such as cloud cover, weather events, and maximum gust speed contained different amounts of missing information across the cities.</p><p>This helped me understand why checking completeness belongs near the beginning of an analysis. A summary may look precise while being based on fewer observations than expected.</p><p>For the temperature calculations, missing or nonnumeric entries were excluded rather than replaced with invented measurements. The resulting statistics described the available observations.</p><p>Moscow had the lowest annual average mean temperature among the six datasets, at approximately 5.99°C. Beijing showed the greatest variation in daily mean temperature, with a standard deviation of approximately 10.94°C.</p><p>These findings helped me distinguish between average conditions and variability. Two places can have similar averages while experiencing different ranges of daily temperatures.</p><p>I also needed to describe the scope accurately. These were airport observations for one year. They supported the assignment’s comparisons, but broader conclusions about climate would require more years and additional evidence.</p><p>Inspecting the database before writing queries</p><p>The SQL assessment began with exploring the database schema.</p><p>I used INFORMATION_SCHEMA.COLUMNS to inspect table structures and identify the actual column names. This became useful throughout the assessment because familiar-looking names were not always safe to assume.</p><p>The London weather events column, for example, began with a space. It had to be referenced as [ Events].</p><p>Another issue appeared when filtering milk trade records. The data used “Exports”, while a filter for “Export” returned no rows. There was no syntax error; the condition simply did not match the stored value.</p><p>Checking the distinct values made the problem clear.</p><p>That experience changed how I approached empty results. Before rewriting an entire query, I could inspect one assumption at a time: the table, the column, the spelling, and the filter value.</p><p>It was a practical reminder that understanding the contents of a database is as useful as knowing SQL syntax.</p><p>What missing weather events revealed</p><p>One task asked me to calculate three counts for London:</p><p>*Days with no event recorded.</p><p>* Days whose event description mentioned rain.</p><p>* Days with precipitation above zero but no event recorded.</p><p>The query returned 114, 238, and 6, respectively.</p><p>The six days in the final category were especially interesting. They showed that a missing event description did not necessarily mean that no precipitation occurred.</p><p>This distinction helped me think more carefully about missing data. An empty field tells me something about what was recorded. It does not automatically explain what happened in the real world.</p><p>I used IS NULL to identify missing event values and LIKE ‘%Rain%’ to find descriptions containing rain, including descriptions that combined rain with other events.</p><p>Using CASE expressions inside aggregate functions allowed me to return the three counts together. I could see how conditional aggregation turns several related questions into one readable summary.</p><p>Calculating percentages without losing information</p><p>The customer churn task required the number of customers, the number who churned, and the churn percentage for each contract type.</p><p>The results showed:</p><ul><li>57.6% for Month-to-Month contracts.</li><li>* 53.9% for One Year contracts.</li><li>* 27.1% for Two Year contracts.</li></ul><p>This exercise reinforced the importance of data types.</p><p>In SQL Server, averaging an integer column produces an integer result. Integer division can also remove the fractional part of a calculation. That means a query can run successfully while losing information needed for an accurate rate.</p><p>Using decimal arithmetic before division preserved the fractional values. I could then round the percentage to one decimal place.</p><p>The order mattered. Converting a value after integer arithmetic has already removed its fractional part cannot restore the missing information.</p><p>I began to see why checking a calculation involves more than confirming that the formula looks right. The types of values entering that formula also affect the result.</p><p>Joining sources and examining what does not match</p><p>Another task combined WHO tuberculosis data with World Bank population data.</p><p>I joined the tables on country names and calculated tuberculosis deaths per 100,000 people using the World Bank population figures. The query then returned the five highest rates among the matched countries.</p><p>The following task asked how many WHO countries had no matching country name in the World Bank table. The answer was 28.</p><p>This made the limitations of text-based joins clearer to me. Country names may differ because of spelling, naming conventions, abbreviations, or missing coverage.</p><p>An inner join returns matching records, so an analysis based only on its output can overlook the records that were left behind.</p><p>In a real project, I would review the unmatched countries and use a verified mapping or standard country codes where available. I would also check coverage rather than assume every unmatched name was a spelling problem.</p><p>The join was therefore part of the analytical decision, not just a technical step.</p><p>Grouping, ranking and following values over time</p><p>The later tasks introduced different ways to summarise and organise records.</p><p>For milk exports, I grouped values by partner country and used HAVING to identify partners represented in all twelve months. I excluded the “World” total row so it would not be counted as another partner country.</p><p>For the weather comparison, UNION ALL allowed me to combine the six city tables under consistent city labels before calculating average mean temperatures.</p><p>To find London’s hottest day in each month, I used a common table expression and ROW_NUMBER(). The ranking had to consider maximum temperature first and the earlier date second, so ties followed the task’s instructions.</p><p>The running-total exercise added another perspective. I calculated monthly milk exports and then used a windowed sum to show the accumulated year-to-date value alongside each month.</p><p>The final running total reached US$898,651,935 for the records included in the query.</p><p>These tasks helped me understand the difference between reducing rows into groups and retaining rows while calculating across them. Choosing between aggregation and a window function depends on what the final output needs to show.</p><p>When a correct query produces a suspicious pattern</p><p>The loan approval task gave me the strongest reason to pause and interpret the output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QAD6CHCF5B_ducYQ_L5nvw@2x.jpeg" /></figure><p>Rural applicants all had credit history recorded as zero and an approval rate of 0%. Semiurban and Urban applicants all had credit history recorded as one and approval rates of 100%.</p><p>The query passed the assessment, but those results raised questions.</p><p>Property area and credit history were perfectly aligned in the returned groups. From that summary, I could not separate their individual relationships with approval.</p><p>I would therefore be cautious about presenting the pattern as evidence for a lending decision. Before training a model, I would want the data source and preparation process investigated.</p><p>Possible explanations include synthetic patterns, selection during data preparation, or target leakage. The output suggests reasons to investigate; it does not prove which explanation applies.</p><p>That distinction became one of my main lessons: producing the requested table is one responsibility, and explaining its limitations is another.</p><p>Learning to work through errors patiently</p><p>Some challenges came from the notebook environment itself.</p><p>I had to understand how SQL queries fitted between the existing triple quotes, preserve the answer-checking function, and rerun the supplied setup after restarting the kernel.</p><p>At times, an error looked more complicated than its cause. Breaking it into smaller questions helped: Was this Python syntax? An incorrect identifier? A filter that matched nothing? A function that had not been defined in the current session?</p><p>This approach gave me a more manageable way to troubleshoot. I could correct the relevant part while preserving work that was already correct.</p><p>By the end of the assignment, I understood more clearly what I needed to keep practising.</p><p>Working with these tools showed me how many small decisions sit behind a useful result. My progress came through learning to examine those decisions and becoming more confident about explaining the evidence behind my answers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c9d4edebd412" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Teaching Machines to Understand Nigerian Pidgin: My Experience with Sentiment Analysis and NLP]]></title>
            <link>https://medium.com/@jaynechinny/teaching-machines-to-understand-nigerian-pidgin-my-experience-with-sentiment-analysis-and-nlp-92b647ff6f44?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/92b647ff6f44</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[sentiment-analysis]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Fri, 09 Oct 2026 16:59:42 GMT</pubDate>
            <atom:updated>2026-10-09T16:59:42.254Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*QfpKXMHvkks0YrJLouS5ag@2x.jpeg" /></figure><p>From learning Python to training machine-learning models, my journey into Natural Language Processing has taught me that understanding human language is far more complicated than recognising words.</p><p>When Human Language Meets Artificial Intelligence</p><p>Imagine reading these two expressions:</p><p>“Ginger is good for your health.”</p><p>“Ginger your swagger!”</p><p>The same English word, ginger, appears in both sentences, but the meaning changes significantly.</p><p>In the first sentence, ginger refers to a plant. In the second, it is used in a Nigerian expression associated with encouragement, motivation and confidence.</p><p>A human familiar with Nigerian language and culture can recognise this difference almost immediately.</p><p>But what happens when we ask a machine to interpret these expressions?</p><p>Would it recognise the intended meaning? Would it understand the emotional tone? Or would it rely on patterns learned from standard English and potentially misunderstand the message?</p><p>These questions capture what attracted me to Natural Language Processing (NLP) and, more specifically, sentiment analysis for Nigerian Pidgin.</p><p>During my Data Science, Machine Learning and Generative AI internship at Datara Flow Academy, I began exploring how computers can classify opinions expressed in Nigerian Pidgin.</p><p>What started as a research topic gradually became a practical learning experience involving academic literature, datasets, Python programming, Google Colab, traditional machine-learning algorithms and transformer-based language models.</p><p>This article documents that journey, the experiments I have conducted so far, and the questions that remain unanswered.</p><p>Why Nigerian Pidgin Deserves Attention in NLP</p><p>Nigerian Pidgin is widely used in everyday conversations, entertainment, informal communication and social-media interactions.</p><p>It is expressive, flexible and deeply connected to the social and cultural contexts in which people use it.</p><p>However, Nigerian Pidgin presents challenges for computational language processing.</p><p>Unlike languages with widely adopted standardised writing conventions, Nigerian Pidgin can be written in different ways. A word or expression may have several spelling variations, even when speakers understand the same meaning.</p><p>Another challenge is code-mixing, where speakers combine Nigerian Pidgin with English or other languages within a single expression.</p><p>There are also idioms, abbreviations, humour, sarcasm and context-dependent meanings.</p><p>These features make Nigerian Pidgin particularly interesting for sentiment analysis.</p><p>Sentiment analysis is an NLP task that attempts to identify the emotional orientation expressed in text. In my research, the objective is to classify text into three categories:</p><ul><li>Positive: expressions communicating favourable sentiment.</li><li>Negative: expressions communicating unfavourable sentiment.</li><li>Neutral: expressions without a clearly positive or negative orientation.</li></ul><p>Although this sounds straightforward, human language rarely follows perfectly predictable rules.</p><p>An expression may appear negative when interpreted literally but communicate something different within its cultural context.</p><p>This is why language representation, training data and preprocessing decisions matter.</p><p>From Research Papers to a Research Question</p><p>Before training any model, I spent time reviewing academic literature related to African-language NLP and Nigerian Pidgin sentiment analysis.</p><p>One important resource was NaijaSenti, a human-annotated sentiment dataset covering Hausa, Igbo, Yoruba and Nigerian Pidgin.</p><p>Another was AfriSenti, which expanded sentiment-analysis benchmarking across multiple African languages.</p><p>I also examined research on Nigerian Pidgin semantic enrichment and the limitations of applying English sentiment lexicons directly to Nigerian Pidgin.</p><p>One study particularly interested me because it explored how English-derived words can carry different meanings in Nigerian Pidgin. Its findings reinforced an important lesson: language models need more than vocabulary recognition. They need representations that are sensitive to context.</p><p>Research on orthographic variation raised another question.</p><p>If people write the same Nigerian Pidgin expression in different ways, could standardising those variations improve sentiment classification?</p><p>Or might aggressive text cleaning remove information that a machine-learning model needs?</p><p>These questions helped shape my research topic:</p><p>Sentiment Analysis of Nigerian Pidgin Using Machine Learning and Transformer-Based Models.</p><p>My research focuses on comparing traditional machine-learning approaches with an African-language transformer model, while investigating how different text-preprocessing strategies may affect performance.</p><p>The planned experimental design includes three models and three preprocessing conditions.</p><p>Moving from Theory to Google Colab</p><p>Reading research papers helped me understand what previous researchers had investigated.</p><p>However, implementing a research idea required a different set of skills.</p><p>I had to move from understanding concepts theoretically to working with actual data and Python code.</p><p>Google Colab became my primary environment for the practical experiments.</p><p>It allowed me to organise my notebook, work with Python libraries, inspect data, train models and evaluate their predictions.</p><p>The dataset used in my research is the Nigerian Pidgin component of NaijaSenti, which contains labelled social-media text.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/501/1*yEGfgRWr96poPhauhvKb-g@2x.jpeg" /></figure><p>Its official partitions include 5,121 training examples, 1,281 development examples and 4,154 test examples.</p><p>An important characteristic of the dataset is class imbalance.</p><p>In the training partition, negative and positive examples substantially outnumber neutral examples. This creates a challenge because a model can appear successful by predicting the majority classes while performing poorly on the minority class.</p><p>That observation influenced how I approached evaluation.</p><p>Instead of relying only on accuracy, I selected macro-F1 as the primary performance metric.</p><p>Macro-F1 calculates the F1-score for each sentiment class and gives the classes equal weight when averaging them.</p><p>This makes it particularly useful when the dataset is imbalanced.</p><p>Training My First Machine-Learning Models</p><p>My first practical modelling experiments involved two traditional machine-learning algorithms:</p><p>Logistic Regression and Linear Support Vector Machine (Linear SVM).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/507/1*qeUQb4ZNyWG22uBKMo25DA@2x.jpeg" /></figure><p>Both models used Term Frequency – Inverse Document Frequency, commonly called TF-IDF, to transform text into numerical features.</p><p>Machines cannot directly process sentences in the same way humans read them. Traditional text-classification systems therefore need a method of representing text mathematically.</p><p>TF-IDF assigns numerical importance to terms based on how frequently they appear within a document and across the document collection.</p><p>Once the text is converted into features, a classifier can learn relationships between those features and the sentiment labels.</p><p>After implementing and evaluating the two baseline models, I obtained the following macro-F1 scores:</p><p>Model	Macro-F1</p><p>Logistic Regression	0.4309</p><p>Linear SVM	0.4134</p><p>Logistic Regression achieved the higher score in these initial baseline experiments.</p><p>However, the results also demonstrated that Nigerian Pidgin sentiment classification is not a simple task.</p><p>The models were learning patterns from the available training data, but their performance left substantial room for improvement.</p><p>I began to appreciate the difference between successfully executing machine-learning code and developing a model that performs reliably across different sentiment categories.</p><p>A completed training process does not automatically mean a model understands language effectively.</p><p>Introducing AfriBERTa: A Different Approach</p><p>After experimenting with traditional machine learning, I moved to AfriBERTa Base, a transformer-based language model developed for African-language representation.</p><p>This was one of the most demanding stages of my practical work.</p><p>Unlike TF-IDF-based classifiers, transformer models generate contextual representations of text.</p><p>In simple terms, a transformer considers how words relate to surrounding words rather than relying exclusively on isolated word-frequency features.</p><p>This makes transformer-based approaches particularly interesting for language tasks involving contextual meaning.</p><p>Using Hugging Face Transformers and PyTorch in Google Colab, I fine-tuned AfriBERTa for three-class Nigerian Pidgin sentiment classification.</p><p>The model completed three training epochs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/623/1*TjOCHRSMGeZZT_I1CVeJGA@2x.jpeg" /></figure><p>Its development-set evaluation produced:</p><p>Evaluation metric	Result</p><p>Accuracy	Approximately 69%</p><p>Macro-F1	0.4154</p><p>Evaluated examples	1,011</p><p>At first glance, the accuracy might suggest that the model performed reasonably well.</p><p>But the macro-F1 score told a more complicated story.</p><p>The model struggled with the neutral sentiment category, recording an F1-score of zero for that class in the evaluation.</p><p>This was a particularly important observation.</p><p>A model can achieve relatively high accuracy while failing to recognise a minority class.</p><p>The result highlighted the practical consequences of class imbalance and demonstrated why evaluating each sentiment category matters.</p><p>Although AfriBERTa offers contextual language representations, that capability alone does not guarantee stronger results in every experimental setting.</p><p>The transformer also did not automatically outperform my initial Logistic Regression baseline.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/418/1*X2E0KqPh4bgqtkhxNJ81fA@2x.jpeg" /></figure><p>However, I would not treat the difference between the reported baseline and transformer scores as a definitive model ranking yet. A rigorous comparison requires confirming that the models were evaluated using exactly the same examples and experimental conditions.</p><p>That distinction matters in research: numbers become meaningful comparisons only when the evaluation procedures are comparable.</p><p>What Text Preprocessing Taught Me</p><p>Another important part of my research concerns text preprocessing.</p><p>I proposed three conditions:</p><p>1)Minimal preprocessing, which preserves most of the original language features while applying essential technical cleaning.</p><p>2)Standard social-media preprocessing, which handles selected elements such as whitespace, URLs, mentions and other common social-media artefacts.</p><p>3)Orthographic-aware preprocessing, which aims to account for documented spelling variations in Nigerian Pidgin without imposing inappropriate English-language corrections.</p><p>*The objective is to investigate whether different cleaning strategies influence traditional machine-learning and transformer-based models differently.</p><p>During an initial diagnostic check, however, the three preprocessing approaches produced identical text outputs for the examples examined.</p><p>This was an important reminder that designing different preprocessing functions does not necessarily mean they will produce meaningfully different data.</p><p>Before claiming that one preprocessing strategy improves performance, I need to establish that the strategies actually create the intended differences and evaluate those differences systematically.</p><p>In particular, orthographic-aware preprocessing requires carefully selected and documented spelling mappings.</p><p>It should not simply replace Nigerian Pidgin expressions with standard English words.</p><p>The purpose is to preserve linguistic meaning while investigating whether particular forms of normalisation help the models.</p><p>For me, this became another lesson in experimental design: an experiment must test the difference it claims to test.</p><p>The Challenges Behind the Notebook</p><p>One thing I have learned during this internship is that practical machine learning involves much more than writing code and pressing the Run button.</p><p>There are datasets to inspect, labels to understand, preprocessing decisions to justify, libraries to work with, model configurations to manage and evaluation outputs to interpret.</p><p>Sometimes, the challenge is not understanding what a function does but understanding why a particular result occurred.</p><p>Other times, the code executes successfully, but the outcome raises more questions than it answers.</p><p>Coming into data science from a background outside traditional computer science, I have had to develop my understanding gradually.</p><p>My experience with no-code automation helped me appreciate workflows, structured information and logical processes. Python and machine learning have pushed me further into understanding the computational mechanisms behind those processes.</p><p>I am still learning.</p><p>There are concepts I need to revisit, results I need to investigate and techniques I have not yet mastered.</p><p>But I have also learned that research does not require pretending to know everything.</p><p>It requires curiosity, careful observation, documentation and a willingness to question your own assumptions.</p><p>Publishing My Work on GitHub</p><p>As my experiments progressed, I wanted to organise the work beyond my Google Colab environment.</p><p>I created a public GitHub repository containing my research implementation notebook and project documentation.</p><p>The repository provides a place to document the experimental workflow, explain the dataset and models, and maintain a record of the work as the research develops.</p><p>It is also part of my learning process.</p><p>Reproducibility matters in data science because research findings should be traceable to the data, code, configurations and evaluation procedures used to produce them.</p><p>My repository is still developing, and the planned project structure extends beyond the materials currently uploaded.</p><p>Nevertheless, publishing the notebook and README was an important step towards documenting the research openly.</p><p>GitHub repository:</p><p>Sentiment Analysis of Nigerian Pidgin Using Machine Learning and Transformer-Based Models</p><p>What Comes Next?</p><p>The experiments completed so far represent an initial stage of the research rather than a final answer to the research questions.</p><p>Several important tasks remain.</p><p>I need to establish fully comparable evaluation conditions for the models, investigate the treatment of neutral sentiment, examine the preprocessing procedures more carefully and conduct further error analysis.</p><p>I also want to understand how spelling variation, code-mixing and context-dependent expressions contribute to incorrect predictions.</p><p>The official test partition should remain reserved for final evaluation rather than being used repeatedly during model development.</p><p>These steps are important because the objective is not simply to produce a higher score.</p><p>It is to understand what the models are learning, where they struggle and what the results can reasonably tell us about Nigerian Pidgin sentiment analysis.</p><p>Final Reflection: Learning to Ask Better Questions</p><p>When I began exploring Nigerian Pidgin sentiment analysis, I was interested in whether machines could understand the language people use in everyday Nigerian conversations.</p><p>Working with the dataset and training different models has made that question more interesting.</p><p>I now find myself asking:</p><p>What does it really mean for a machine to understand sentiment?</p><p>How much of a language’s meaning can be captured through statistical patterns?</p><p>What happens when cultural expressions, spelling variations and minority sentiment classes are not adequately represented in training data?</p><p>And how can researchers build more reliable NLP systems for languages that have historically received less computational attention?</p><p>My journey with Datara Flow Academy has shown me that data science is not only about algorithms and technical tools.</p><p>It is also about research questions, evidence, experimentation and intellectual honesty.</p><p>Sometimes, a result that does not meet your expectations teaches you more than one that does.</p><p>For me, Nigerian Pidgin sentiment analysis is more than an internship project. It is an opportunity to connect language, culture and technology while developing practical skills in Python, machine learning and NLP.</p><p>I am not at the end of the research journey. I am learning how to conduct it properly.</p><p>And that is progress worth documenting.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=92b647ff6f44" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Language to Data Science: How I’m Using Code and AI to Understand African Languages]]></title>
            <link>https://medium.com/@jaynechinny/from-language-to-data-science-how-im-using-code-and-ai-to-understand-african-languages-cf2f305c85fd?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/cf2f305c85fd</guid>
            <category><![CDATA[language-model]]></category>
            <category><![CDATA[sentiment-analysis]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Sat, 03 Oct 2026 16:45:22 GMT</pubDate>
            <atom:updated>2026-10-03T16:45:22.795Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*05GKiODQBDZMboz3sFJ-sQ@2x.jpeg" /></figure><p>How studying English, teaching across cultures, and learning to code led me to build AI systems that take language, emotion, and context seriously.</p><p>For a long time, I never imagined that my background in English and Literary Studies would lead me toward data science, artificial intelligence, and machine learning.</p><p>When people hear that someone studied English and Literature, they often think of novels, poetry, grammar, essays, literary criticism, or teaching. Those things are part of my story. But language is much bigger than literature.</p><p>Language is data. It carries culture, emotion, and identity. And increasingly, it’s one of the most important forms of data that AI systems must understand.</p><p>That realization changed how I see my own academic journey.</p><p>From studying language to teaching language</p><p>My academic journey began with English and Literary Studies. I’ve always been interested in how people communicate, how meaning shifts with context, how culture shapes expression, and how the same words can mean different things to different people.</p><p>That interest eventually became practical. Today, I teach English while living in Oman.</p><p>Living here has given me another perspective on language. I’ve come to understand something textbooks don’t always teach: knowing a language is not the same as understanding the people who speak it.</p><p>Arabic is spoken across the Gulf, but it doesn’t sound exactly the same everywhere. Different countries and communities have their own dialects, expressions, pronunciation patterns, and cultural contexts. As someone living in Oman, I began paying closer attention to Arabic – not just as a language to memorize, but as a language connected to the people, environment, and culture around me.</p><p>At the same time, I became interested in learning Turkish. I also speak and write Igbo, my mother tongue.</p><p>So my relationship with language has become increasingly multilingual:</p><p>•	English</p><p>•	Igbo</p><p>•	Arabic and its Gulf varieties</p><p>•	Turkish (in progress)</p><p>I’ve experienced what it means to communicate across different linguistic and cultural environments. And that made me ask a bigger question:</p><p>What happens when technology has to understand all of us?</p><p>A few words from my multilingual life</p><p>To give a sense of the languages that shape my thinking, here are a few everyday words and expressions I move between.</p><p>Nigerian Pidgin</p><p>•	“How far?” – A common greeting, like “How are you?” or “What’s up?”</p><p>•	“No wahala.” – “No problem” or “Everything is fine.”</p><p>•	“E go better.” – “It will get better.” Often used to encourage someone.</p><p>•	“I dey kampe.” – “I’m doing well / I’m strong / I’m managing.”</p><p>These phrases carry more than literal meaning. They carry mood, relationship, and context. The same words can sound warm, sarcastic, or serious depending on how and where they’re used.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*0dutFDfHu-kjxL7Jl7Wc2A@2x.jpeg" /></figure><p>Arabic (Omani / Gulf flavor)</p><p>•	“مرحبا” (marḥaban) – “Hello.”</p><p>•	“كيف حالك؟” (kayf ḥālak? / kayf ḥālik?) – “How are you?” (to a male / to a female)</p><p>•	“ما شاء الله” (mā shāʾ Allāh) – “What God has willed.” Used to express appreciation or admiration, often to avoid envy.</p><p>•	“إن شاء الله” (in shāʾ Allāh) – “If God wills.” Used when talking about future plans.</p><p>Living in Oman, I hear these expressions daily. They’re not just vocabulary; they reflect how people relate to time, fate, and each other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*BixedDYhozUuLIpUNS_Z2w@2x.jpeg" /></figure><p>Turkish</p><p>•	“Merhaba” – “Hello.”</p><p>•	“Nasılsın?” – “How are you?”</p><p>•	“Teşekkür ederim.” – “Thank you.”</p><p>•	“Kolay gelsin.” – Literally “May it come easily.” Said to someone who is working, studying, or going through something.</p><p>Turkish is still a learning journey for me, but even basic phrases have changed how I notice politeness, empathy, and everyday care in conversation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*_SuseeV1DY9Pq7qVaTL6lQ@2x.jpeg" /></figure><p>Igbo (my mother tongue)</p><p>•	“Kèdú?” – “How are you?” / “What’s up?”</p><p>•	“Daalụ.” – “Thank you.”</p><p>•	“Ọ dị mma.” – “It is good / It’s fine.”</p><p>•	“Nnoo.” – “Welcome.”</p><p>Igbo reminds me where I come from. It’s the language of home, family, and many of my earliest memories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*jxHQt9GdSvI4TtOY7VoCdQ@2x.jpeg" /></figure><p>Why these phrases matter for my work</p><p>These examples are not just language trivia. They are a small window into the kind of challenge my research tries to address:</p><p>How can AI systems move beyond literal word meanings and start capturing context, culture, and emotion – especially in languages that have far less data and tooling than English?</p><p>I won’t go into the technical details here, but every phrase above represents a kind of input that a sentiment model might one day need to interpret correctly: not just “positive” or “negative,” but why and in what context.</p><p>In my current work, I focus on one underrepresented African language variety where expressions like these are common, and where existing AI tools often struggle to capture the intended emotion. The goal is not just better accuracy numbers, but technology that understands more of the people it is meant to serve.</p><p>Is language a barrier to technology?</p><p>Artificial intelligence has become extremely powerful. We can ask AI questions, generate images, create videos, write code, analyze documents, translate languages, automate workflows, and build intelligent systems.</p><p>But there is still a fundamental problem: AI does not experience language the way humans do.</p><p>Human communication contains context, slang, culture, sarcasm, spelling variations, code‑switching, and expressions that can’t always be translated word‑for‑word.</p><p>This is especially true for languages and dialects that sit outside the mainstream of global tech. A sentence that looks simple on the surface can carry cultural meanings, humor, or emotions that are invisible to a model trained mostly on high‑resource languages.</p><p>This is where my academic background and my new field begin to connect.</p><p>My English and Literature training taught me to pay attention to meaning and context. My experience with different languages taught me to recognize linguistic variation. And data science is teaching me how to turn those questions into something that can be measured, analyzed, and modeled.</p><p>Why I chose to focus on an underrepresented language</p><p>Early in my data science journey, I realized I didn’t want to work on yet another project in a language that already has endless datasets, benchmarks, and tutorials.</p><p>I wanted to work on something closer to home – something that reflects the way many people actually speak, but that technology still struggles to understand.</p><p>That led me to focus my research on an African language variety that is widely spoken but historically underrepresented in NLP. I won’t go into the technical details here; the key point is that this language sits at the intersection of English, local languages, and everyday digital communication.</p><p>People express joy, frustration, sarcasm, humor, and criticism in this language every day – online, in messages, in comments, in reviews. But if you ask many existing AI systems to interpret that text, they often miss the nuance or get it completely wrong.</p><p>That gap is both a technical challenge and a human one.</p><p>– -</p><p>The problem I care about</p><p>At a high level, the problem I care about is this:</p><p>How can we make AI systems better at understanding sentiment and emotion in languages that have far less data and tooling than English?</p><p>Sentiment analysis – the task of detecting whether text is positive, negative, neutral, or something more complex – is already used for:</p><p>•	Customer feedback analysis</p><p>•	Social media monitoring</p><p>•	Public opinion research</p><p>•	Customer service</p><p>•	Brand analysis</p><p>•	Content moderation</p><p>•	Conversational AI</p><p>•	Information retrieval</p><p>But these systems are not equally good for all languages. When they work best for a small set of high‑resource languages, speakers of other languages end up with worse recommendations, weaker moderation, less accurate assistants, and fewer tailored services.</p><p>That creates a language gap in AI.</p><p>My research explores ways to narrow that gap for one specific language context. I’m comparing different approaches and thinking carefully about data, preprocessing, and evaluation – but I’ll leave the detailed methodology for academic papers and talks.</p><p>For this article, the important part is the why: making AI more inclusive for speakers of underrepresented languages.</p><p>My background in English is actually relevant to machine learning</p><p>One of the most interesting things I’ve discovered during this journey is that my previous field didn’t disappear when I entered technology. It became useful in a different way.</p><p>When I studied literature, I learned to examine texts closely. When I studied English, I learned about language structure, meaning, and communication. When I began teaching English, I had to think about how different learners understand the same concept. When I started learning other languages, I became more aware of how culture influences communication.</p><p>And now, as a data science student, I’m learning how computers can analyze language at scale.</p><p>The questions changed from:</p><p>What does this sentence mean?</p><p>to</p><p>How can a machine learn what this sentence means?</p><p>That’s a very different question – but the two are connected.</p><p>From language to code to models</p><p>Machine learning gives us a way to find patterns in data. In Natural Language Processing (NLP), the data can be text. A model can learn from examples.</p><p>If we provide text that has already been labeled with sentiment categories, the model can learn patterns associated with different emotional tones.</p><p>But real language is much more complicated. People don’t always say “I am angry” or “I love this.” They might use sarcasm, slang, misspellings, code‑mixing, cultural expressions, or indirect ways of communicating emotion.</p><p>That’s why data quality and careful preprocessing matter so much. It’s also why working with underrepresented languages is both difficult and important: you can’t just plug them into existing pipelines and expect good results.</p><p>This is where coding comes in. Writing scripts to clean text, handle spelling variations, prepare datasets, and train models is how I turn linguistic questions into something a computer can work with. The code is not the end goal; it’s the bridge between human language and machine understanding.</p><p>The connection to Generative AI</p><p>My journey has also taken me into Generative AI. I’ve learned that AI isn’t just about generating beautiful images or writing text. One of the most important skills is learning how to communicate with AI effectively.</p><p>That’s where prompting comes in. A prompt isn’t just a question. It can contain context, instructions, constraints, examples, desired output formats, and reasoning requirements.</p><p>The better we understand language, the better we can communicate our intentions to AI systems. This is another point where my background in language becomes valuable.</p><p>I’m interested in understanding how humans communicate with machines and how machines interpret human language. That connects prompting, NLP, machine learning, and generative AI in a very natural way.</p><p>But the problem is bigger than one language</p><p>My research focuses on one specific language context, but the underlying problem is much broader.</p><p>Around the world, people communicate in thousands of languages and dialects. Yet technology cannot develop equally for everyone if AI systems are primarily optimized around a relatively small number of high‑resource languages.</p><p>This can affect:</p><p>a)Access to AI tools</p><p>b)Digital education</p><p>c)Customer service</p><p>d)Online communication</p><p>e)Information retrieval</p><p>f)Sentiment analysis</p><p>g)Translation</p><p>h)Speech technology</p><p>i)Conversational systems</p><p>For me, this is where technology becomes more than technology. It becomes a question of inclusion.</p><p>If AI is going to interact with humanity, then humanity is not going to communicate with AI in only one language.</p><p>Living between languages has changed how I see AI</p><p>Living in Oman has made this even more personal for me. I’ve experienced what it feels like to navigate communication across cultures.</p><p>Sometimes you understand the language but not the expression. Sometimes you understand the words but not the cultural meaning. Sometimes you know what someone is saying but need to understand the context behind it. And sometimes, communication happens without perfect language at all.</p><p>That experience has made me more curious about multilingual AI:</p><p>How does a machine distinguish languages?</p><p>How does it understand dialects?</p><p>How does it handle code‑switching?</p><p>How does it recognize sentiment when spelling is inconsistent?</p><p>How can an AI system understand a culturally specific expression?</p><p>How can we build datasets that represent people accurately?</p><p>These are the kinds of questions that make me excited about research.</p><p>From automation to research</p><p>My journey into technology has also taken me into automation and workflow design. I’ve been learning tools and concepts around workflow automation, data, AI, and systems.</p><p>Automation taught me to think differently: instead of asking only “How do I perform this task?”, I began asking “How can this process be designed so that a system can perform it efficiently?”</p><p>Data science takes that thinking further. Instead of only looking at individual examples, I can work with datasets and ask: “What patterns exist?”</p><p>Machine learning asks: “Can a model learn those patterns?”</p><p>Research asks: “Does the approach actually work, and how can we demonstrate that?”</p><p>Generative AI asks another fascinating question: “How can machines generate useful language, content, and responses from what they have learned?”</p><p>These fields may look separate from the outside. To me, they are increasingly connected.</p><p>I am not leaving language behind. I am taking it into technology.</p><p>That’s probably the biggest lesson I’ve learned from this journey.</p><p>I don’t see my English and Literary Studies degree as something unrelated to data science. I see it as my foundation.</p><p>Language taught me to observe.</p><p>Literature taught me to interpret.</p><p>Teaching taught me to communicate.</p><p>Learning other languages taught me to appreciate linguistic differences.</p><p>Living in another country taught me to pay attention to culture.</p><p>Automation and coding taught me to think in systems and process language at scale.</p><p>Data science is teaching me to analyze patterns.</p><p>Machine learning is teaching me how models learn from those patterns.</p><p>Generative AI is showing me how far human – computer communication can go.</p><p>My research sits right in the middle of all these interests.</p><p>What I hope to contribute</p><p>I’m still learning. I’m not presenting myself as someone who has already solved the problem. That’s the beauty of research.</p><p>You start with a question. You investigate what other researchers have already discovered. You identify what remains unclear. You collect evidence. You experiment. You compare approaches. You learn from your results. And hopefully, you contribute one small piece to a much bigger body of knowledge.</p><p>For me, that contribution begins with one underrepresented African language variety. I want to explore how modern machine learning approaches can be adapted to better capture sentiment and emotion in that context, and how choices around data and preprocessing can affect performance.</p><p>But my curiosity doesn’t stop there. I’m interested in:</p><p>Multilingual AI</p><p>African languages</p><p>How culture influences computational language understanding</p><p>Prompt engineering and human – AI communication</p><p>Automation and intelligent workflows</p><p>What happens when people from different linguistic backgrounds are finally represented properly in the systems being built for the future</p><p>The future of AI must understand more than English</p><p>English gave me my first doorway into the world of language. Igbo reminds me where I come from. Arabic is part of the environment I’m learning to understand in Oman. Turkish represents another language and culture I’ve chosen to explore. And technology has given me a new way to connect all these interests.</p><p>My journey from English and Literature into data science may look unusual. But perhaps it shouldn’t.</p><p>Because data isn’t only numbers.</p><p>•	Data can be words.</p><p>•	Data can be conversations.</p><p>•	Data can be culture.</p><p>•	Data can be emotion.</p><p>•	Data can be stories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*YkbNW5RJ0lJZuS_otjZnHw@2x.jpeg" /></figure><p>And when we teach machines to understand those things, we aren’t simply building better models. We’re trying to build technology that understands more of the people it’s meant to serve.</p><p>That’s why I chose this field. That’s why I chose to focus on an underrepresented African language. And that’s why my journey into data science is, in many ways, still a journey into language.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cf2f305c85fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learning Python: You Don’t Have to Know Everything to Begin]]></title>
            <link>https://medium.com/@jaynechinny/learning-python-you-dont-have-to-know-everything-to-begin-e6130fd299db?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6130fd299db</guid>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 14:28:05 GMT</pubDate>
            <atom:updated>2026-09-29T14:28:05.999Z</atom:updated>
            <content:encoded><![CDATA[<p>When I started learning Python, I knew I had a lot to learn.</p><p>I didn’t know everything. In fact, there were many concepts I was still trying to understand, and there were moments when I had to sit with my code, look at it again and again, and literally crack my brain trying to figure out what I was doing wrong.</p><p>But that is what learning is about.</p><p>I recently completed a series of Python assignments and assessments using VS Code. The exercises covered different areas, including mathematical operations, the datetime module, JSON strings, Python dictionaries, JSON files, error handling, data calculators, pip, and virtual environments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ts2wwvHuHWlGVpVMYqV7zw@2x.jpeg" /></figure><p>Each task introduced me to something different.</p><p>Some concepts came easily, while others required more patience. Sometimes I understood the idea but struggled to apply it in code. Other times, I had to go back, read through my work, rethink my approach, and try again.</p><p>And honestly, I think that is one of the most interesting parts of learning to code.</p><p>You don’t have to be perfect.</p><p>You don’t have to know everything before you start.</p><p>You don’t even have to understand everything immediately.</p><p>What matters is your willingness to keep learning and your zeal to understand.</p><p>Working with JSON and dictionaries, for example, helped me understand how structured data can be represented and manipulated. The datetime module introduced me to working with dates and time. Error handling taught me that programs need to be prepared for things going wrong.</p><p>I also learned about pip and virtual environments – things I had heard about but needed to actually work with before they started making more sense.</p><p>The assignments weren’t just about completing tasks and moving on to the next one. They challenged me to think.</p><p>There were times I had to pause and ask myself:</p><p>“Why is this happening?”</p><p>“What am I missing?”</p><p>“How can I make this work?”</p><p>And sometimes, after finally figuring something out, I would look at the solution and think, “Oh… so that’s what I was missing!”</p><p>Those little moments are part of the journey.</p><p>Learning Python is teaching me that coding isn’t about being extremely perfect. It is about practice, patience, curiosity, and persistence.</p><p>Nobody starts knowing everything.</p><p>You learn. You make mistakes. You get confused. You ask questions. You try again. Eventually, something that once looked complicated begins to make sense.</p><p>That is the beauty of learning.</p><p>I am still at the beginning of my journey, and I know there is so much more to learn – data analysis, statistics, automation, machine learning, and AI.</p><p>But I am enjoying the process.</p><p>For every concept I understand today, there are several more waiting for me tomorrow.</p><p>And that’s okay.</p><p>I don’t need to know everything today.</p><p>I just need to keep showing up, keep practicing, keep asking questions, and keep trying my best to understand.</p><p>Learning is a journey, and I am enjoying mine – one line of Python at a time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6130fd299db" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Python Core Concepts and Mastery: OOP, Inheritance, Iterators, Polymorphism, and Scope Explained]]></title>
            <link>https://medium.com/@jaynechinny/python-core-concepts-and-mastery-oop-inheritance-iterators-polymorphism-and-scope-explained-9ce53fbb724c?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ce53fbb724c</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[oop-concepts]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Sat, 19 Sep 2026 07:17:11 GMT</pubDate>
            <atom:updated>2026-09-19T07:17:11.006Z</atom:updated>
            <content:encoded><![CDATA[<p>Introduction</p><p>Python is one of the most beginner-friendly programming languages, but mastering its core concepts separates novice developers from true practitioners. Whether you’re building automation workflows, data pipelines, or complex applications, understanding Object-Oriented Programming (OOP), Inheritance, Iterators, Polymorphism, and Scope is essential.</p><p>In this article, I’ll walk you through these five foundational concepts with practical examples and real-world code. By the end, you’ll have a solid foundation to write cleaner, more efficient, and more maintainable Python code.</p><ol><li>Object-Oriented Programming (OOP)</li></ol><p>OOP is a programming paradigm that organizes code into reusable objects. Instead of writing scattered functions, you bundle data and methods together in classes.</p><p>Key Components:</p><p>A class is a blueprint for creating objects. An object is an instance of a class. Methods are functions inside a class. Attributes are data stored in a class.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R_Gndd58_Z3t4EgE9fVZvg@2x.jpeg" /></figure><p>Why it matters: OOP makes your code modular, reusable, and easier to debug. Instead of writing the same logic multiple times, you create a class once and use it everywhere. This approach keeps your code organized and maintainable as your projects grow larger.</p><p>When you use OOP, you think about what data an object should hold and what actions it should perform. This makes your code more intuitive and easier for other developers to understand.</p><p>2. Inheritance</p><p>Inheritance allows you to create a new class based on an existing class. The new class (child) inherits all properties and methods from the parent class. This means you don’t have to rewrite code that already exists in the parent class.</p><p>Benefits:</p><p>Code reusability is the main advantage – you write functionality once and use it in multiple classes. Logical hierarchy helps organize your code better. Reduced redundancy means less duplicate code in your project.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Brer91cjc8GvckrKOABcKw@2x.jpeg" /></figure><p>A child class can use all the methods and attributes of its parent class. It can also add its own new methods or override existing ones with different implementations. This flexibility makes inheritance a powerful tool for building scalable applications.</p><p>When you have a BankAccount class with basic functionality, you can create a SavingsAccount class that inherits from it and adds specific savings account features like interest calculation without rewriting the basic banking functionality.</p><p>3. Iterators</p><p>Iterators allow you to loop through objects one element at a time. To create an iterator, you need to implement two methods. The first method returns the iterator object itself. The second method returns the next value from the sequence. When there are no more items, the iterator raises StopIteration to signal the end.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_JyYQHNKOXhIU5hrG1rybw@2x.jpeg" /></figure><p>How it works:</p><p>Iterators are useful for working with large datasets or custom sequences. Instead of loading all data into memory at once, iterators produce one item at a time, making your code more efficient.</p><p>When you use a for loop, Python is actually using an iterator behind the scenes. Custom iterators let you define exactly how your objects should behave when looped through.</p><p>Iterators enable seamless iteration through any object you create. They’re particularly powerful when working with infinite sequences or when you want to generate values on-the-fly rather than storing them all in memory. This makes them essential for handling large data and complex looping scenarios.</p><p>4. Polymorphism</p><p>Polymorphism means “many forms.” It allows objects of different classes to be used interchangeably, especially when they have methods with the same name but different implementations.</p><p>Key Idea: Different objects respond differently to the same method call. This is powerful because you can write code that works with any object that has a particular method, without needing to know the exact class of that object.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lDaxRmrH6L9-Y1QdbMMnqQ@2x.jpeg" /></figure><p>Polymorphism makes your code flexible and maintainable. You can write functions that work with multiple types of objects without checking which type each object is. This reduces code duplication and makes your programs easier to extend with new classes in the future.</p><p>For instance, if you have different animal classes, each with a speak() method, you can write a single function that makes any animal speak without needing different functions for dogs, cats, and birds.</p><p>5. Scope</p><p>Scope determines where a variable can be accessed in your code. Python has three main scopes that control variable visibility and modification.</p><p>Local Scope: Variables created inside a function are local to that function only. They cannot be accessed outside the function. Each time you call the function, new local variables are created.</p><p>Global Scope: Variables created outside any function are global. They can be accessed from anywhere in your program, including inside functions. However, to modify a global variable from inside a function, you must use the global keyword. Without it, Python treats the assignment as creating a new local variable.</p><p>Nonlocal Scope: In nested functions, the nonlocal keyword lets you modify variables from the enclosing function’s scope. This is useful when you have functions inside functions and need to access or modify variables from the outer function’s scope.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d2mjdt0XiiOLa3mL9mEMgg@2x.jpeg" /></figure><p>Why it matters: Understanding scope prevents variable conflicts and unintended behavior changes. When you understand how Python resolves variable names, you can write cleaner code and avoid bugs caused by accidentally modifying the wrong variable.</p><p>Bringing It All Together</p><p>These five concepts form the foundation of Python programming. OOP gives your code structure and organization by grouping related data and functions. Inheritance lets you reuse code efficiently without duplication. Iterators make looping through objects seamless and memory-efficient. Polymorphism creates flexible, reusable code that works with multiple types. Scope helps you manage variable access properly and prevent conflicts.</p><p>When you combine these concepts, you write code that is maintainable, scalable, reusable, and professional-grade. These patterns are used in real-world applications everywhere, from web frameworks to data science libraries.</p><p>Key Takeaways</p><p>Use classes to bundle related data and methods together. Leverage inheritance to avoid code duplication and establish logical hierarchies. Create iterators for custom looping behavior and efficient data handling. Implement polymorphism for flexible code that works with multiple types. Understand scope to avoid variable conflicts and write cleaner code.</p><p>Master these concepts, and you’ll be well on your way to becoming a Python expert. Start applying them in your projects today and watch your code quality improve dramatically.</p><p>Ready to level up your Python skills? Share your learning journey in the comments below!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9ce53fbb724c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Started Learning Python, and It Changed How I See Automation]]></title>
            <link>https://medium.com/@jaynechinny/i-started-learning-python-and-it-changed-how-i-see-automation-358a5babadfe?source=rss-11706788e6af------2</link>
            <guid isPermaLink="false">https://medium.com/p/358a5babadfe</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Chichi Okafor]]></dc:creator>
            <pubDate>Wed, 02 Sep 2026 17:48:58 GMT</pubDate>
            <atom:updated>2026-09-02T20:27:10.537Z</atom:updated>
            <content:encoded><![CDATA[<p>A week ago, I was looking at Python from the outside.</p><p>Today, I am beginning to understand what it means to actually write code.</p><p>It may sound like a small difference, but for me, it isn’t.</p><p>My background has mostly been around automation and digital workflows. I’ve spent the past few years working with no-code tools, creating systems, experimenting with automations, and finding ways to make repetitive processes easier.</p><p>I became comfortable with the idea of telling a system:</p><p>When this happens, do this. Then do that. If this condition is met, continue here.</p><p>That way of thinking became natural to me.</p><p>But I eventually became curious about what happens when the visual blocks disappear.</p><p>What if I want to create the logic myself?</p><p>What if the solution I have in mind doesn’t fit neatly into an automation platform?</p><p>What if I want to work directly with data?</p><p>That curiosity is what pushed me toward Python.</p><p>Starting From Zero</p><p>I didn’t approach Python expecting to become an expert in a week.</p><p>My goal was much simpler.</p><p>I wanted to stop being intimidated by code.</p><p>I wanted to understand the language enough to look at a Python program and have an idea of what was happening.</p><p>Most importantly, I wanted to experience what it feels like to solve a problem by writing the instructions myself.</p><p>So I started with the fundamentals.</p><p>I learned about different types of data and how Python stores and works with information. I practiced variables, conditions, loops, functions, and different ways of working with collections of data.</p><p>At first, some of these concepts felt very basic.</p><p>But I quickly realized that “basic” doesn’t mean “unimportant.”</p><p>Almost every little program I wrote required me to combine several of these ideas.</p><p>And that was when things started becoming interesting.</p><p>My First Programs Were Simple – and That’s Exactly What I Needed</p><p>I didn’t start by building an AI application.</p><p>I started with problems small enough that I could understand the entire solution.</p><p>One of my first exercises involved student grades.</p><p>I created a program that takes marks from five subjects, works out the average, and assigns a grade.</p><p>It sounds simple.</p><p>And it is.</p><p>But when you’re learning programming, even a simple problem makes you think differently.</p><p>I had to decide where the marks would be stored, how the average would be calculated, and how Python would determine which grade to return.</p><p>Suddenly, something that could be done manually became a set of instructions that a computer could follow.</p><p>That was one of those small moments where programming started to make sense to me.</p><p>Then I Built a Calculator</p><p>My next project was a basic calculator.</p><p>Two numbers go in.</p><p>An operation is selected.</p><p>A result comes out.</p><p>Again, very simple.</p><p>But this time, I was practicing something that I know will become increasingly important as I continue learning: breaking a program into smaller responsibilities.</p><p>Instead of putting every operation into one place, I created functions for the different calculations.</p><p>It helped me understand why functions matter.</p><p>A function isn’t just something you learn because it’s on a Python syllabus.</p><p>It’s a way of saying:</p><p>“This piece of work has a purpose. I want to be able to use it when I need it.”</p><p>That way of organizing logic is something I can already see myself using in larger projects.</p><p>My Favourite One Was the Guessing Game</p><p>The guessing game was probably the project that felt most like an actual little application.</p><p>Python secretly chooses a number between 1 and 20.</p><p>The player tries to guess it.</p><p>If the guess is too low, Python says so.</p><p>If it’s too high, Python says so.</p><p>And when the correct number is found, the game ends.</p><p>To make it work, I had to bring several things together.</p><p>I used Python’s random module to generate the number.</p><p>I used a loop to keep the game going.</p><p>I used conditions to compare the guesses.</p><p>And I practiced exception handling so that invalid input wouldn’t immediately bring everything down.</p><p>What I liked about this project was seeing the code respond to what was happening.</p><p>It stopped feeling like isolated exercises.</p><p>It started feeling like I was giving a computer a set of instructions and watching those instructions turn into behaviour.</p><p>Then Came the Errors</p><p>Of course, learning Python wasn’t just about seeing things work.</p><p>Some of my biggest lessons came from seeing things fail.</p><p>One of the first problems I encountered involved input().</p><p>I was working in a notebook environment, and interactive input didn’t behave the way I expected. At one point, the notebook could sit there waiting instead of responding normally.</p><p>Rather than spending the entire session trying to force it to work, I used predefined values to test my code and focus on the actual logic.</p><p>That experience reminded me that programming doesn’t happen in a vacuum.</p><p>The environment matters too.</p><p>Then there were indentation errors.</p><p>Python has a very particular relationship with indentation, and I discovered that a small spacing mistake can be enough to stop a program from working.</p><p>At first, I found those errors frustrating.</p><p>Eventually, I started understanding what Python was trying to tell me.</p><p>And then there was another kind of error I found even more interesting:</p><p>the code worked, but the answer was wrong.</p><p>Those mistakes forced me to slow down.</p><p>I had to test different situations, follow my own logic, and figure out where my thinking had gone wrong.</p><p>That may have been one of the most valuable parts of the week.</p><p>I’m beginning to understand that programming isn’t just about making code run.</p><p>It’s about making sure the code does what you intended it to do.</p><p>Something Clicked Between Automation and Programming</p><p>This is probably the most interesting thing I’ve taken away from the experience.</p><p>Learning Python didn’t make my previous automation experience feel irrelevant.</p><p>It made me understand it differently.</p><p>When I build an automation workflow, I’m already thinking about logic.</p><p>Something happens.</p><p>A condition is checked.</p><p>An action takes place.</p><p>Another step follows.</p><p>Python gives me the ability to express much more of that logic myself.</p><p>That’s exciting.</p><p>I don’t see myself abandoning no-code tools.</p><p>In fact, I think learning Python will make me better at using them.</p><p>There are situations where a no-code solution is the fastest and most practical option.</p><p>There are other situations where I may need custom code.</p><p>And there will probably be situations where the best solution is a combination of both.</p><p>That’s the direction I’m interested in exploring.</p><p>My Biggest Surprise Wasn’t Python Syntax</p><p>Before I started, I thought the biggest challenge would be remembering Python commands.</p><p>It wasn’t.</p><p>The bigger challenge was learning to think through a problem before trying to write the solution.</p><p>When something doesn’t work, I can’t just stare at the code and hope the answer appears.</p><p>I have to ask questions.</p><p>What was I expecting?</p><p>What actually happened?</p><p>Which part of my logic controls that behaviour?</p><p>What happens if I give it a different input?</p><p>That process is slowly changing how I approach problems.</p><p>And I think that’s where the real value of learning to code begins.</p><p>I’m Not Trying to Rush This</p><p>Technology moves quickly.</p><p>There is always another framework, another AI tool, another programming language, another course, and another trend.</p><p>It would be easy to look at everything happening in AI and machine learning and feel like I should skip straight to the exciting part.</p><p>But I’m choosing not to.</p><p>Right now, I want to understand the foundation.</p><p>I want variables to make sense.</p><p>I want loops to feel natural.</p><p>I want functions to become second nature.</p><p>I want to become comfortable reading and debugging code.</p><p>Because eventually, I want to work with much more complicated systems.</p><p>And I don’t want to build that future on a foundation I never properly understood.</p><p>Where I’m Going Next</p><p>Python is only one part of the larger journey I’m on.</p><p>My next step is working more seriously with data, particularly using Pandas.</p><p>I’m interested in moving from small programs to actual datasets.</p><p>I want to learn how to take messy information, organize it, clean it, analyze it, and find something meaningful inside it.</p><p>From there, I want to continue toward machine learning and generative AI.</p><p>But I also want to keep automation at the centre of what I do.</p><p>I’m interested in the space where these things overlap:</p><p>Automation + Python + Data + AI.</p><p>I don’t know exactly what the final destination looks like yet.</p><p>And honestly, I like that.</p><p>For now, I’m learning one concept at a time, writing one program at a time, making mistakes, fixing them, and gradually becoming more comfortable with code.</p><p>One Week Isn’t Much. But It’s a Beginning.</p><p>I’m still a beginner.</p><p>I have a lot to learn.</p><p>But I can already feel the difference between knowing that Python exists and actually beginning to understand how it works.</p><p>A week ago, code was something I mostly interacted with through the tools built by other people.</p><p>Now I’m starting to write some of that logic myself.</p><p>That’s a small step.</p><p>But every bigger journey begins with one.</p><p>And this is mine.</p><p>If you’re learning something new right now, I’d love to know:</p><p>What are you learning, and what made you decide to start?</p><p>I’m looking forward to sharing what happens when I take the next step into data analysis with Pandas.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=358a5babadfe" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>