<?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 Hemanth Kakarla on Medium]]></title>
        <description><![CDATA[Stories by Hemanth Kakarla on Medium]]></description>
        <link>https://medium.com/@hemanthkakarla?source=rss-163bf74094b9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*HNsoi0weUDv-Ub1-</url>
            <title>Stories by Hemanth Kakarla on Medium</title>
            <link>https://medium.com/@hemanthkakarla?source=rss-163bf74094b9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 14:11:37 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@hemanthkakarla/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[Record Linkage or Multivariate Normalization using Graph clustering]]></title>
            <link>https://hemanthkakarla.medium.com/record-linkage-or-multivariate-normalization-using-graph-clustering-e6773b30eab7?source=rss-163bf74094b9------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6773b30eab7</guid>
            <category><![CDATA[similarity]]></category>
            <category><![CDATA[deduplication]]></category>
            <category><![CDATA[clustering-algorithm]]></category>
            <category><![CDATA[edit-distance]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Hemanth Kakarla]]></dc:creator>
            <pubDate>Wed, 10 Mar 2021 11:34:30 GMT</pubDate>
            <atom:updated>2021-03-10T11:34:30.479Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Abstract</strong></p><p>Deduplicating the database has been a critical task as part of Data cleansing and integration. We focus on the efficient way to join or find similar records in the database leveraging the other entities or attributes in the data table. We propose various filtering methods and also graph clustering techniques to reduce the impurities in the formed clusters and hence saves the computation time. We demonstrate experimentally that the new algorithm outperforms alternative methods on large-scale real datasets under a wide range of parameter settings.</p><h3><strong>Introduction</strong>:</h3><p>Traditional similarity between two entities is calculated using any single variate or attribute, recent development in computer science introduces to multivariate similarity joins or using multiple attributes of two objects while calculating the similarity between them.</p><p>For instance, we may join two products (respectively, customers) if the similarity between their part descriptions (respectively, customer names and addresses) is high or above the specified threshold. This problem of joining similar data has been studied in the context of record linkage, of identifying approximate duplicate entities in databases.</p><p>There are variety of methods considered such edit distance, jaccard similarity, cosine similarity, generalized edit distance similarity. However, no single string similarity function is known to be the overall best similarity function, and the choice usually depends on the application domain.</p><p>Example: characteristics of a effective similarity function for matching products based on their part names where error are usually spelling errors would be different from those matching street addresses because even small differences in the street numbers such as “<em>148th Ave</em>” and “<em>147th Ave</em>” are crucial.</p><h3><strong>Problem Definition and background:</strong></h3><p>Most of the database deduplication algorithms run forever long unless there is a grouping in the dataset, <strong>Table A</strong> has <strong>m</strong> rows the brute force way to deduplicate the records would be <strong><em>mxm</em></strong> similarity calculations which is worse case scenario and for a very large database this will bottle neck the resources. Using filtering techniques, we try to reduce the cartesian product or comparisons of the algorithm.</p><h3><strong>Clustering or deduplication</strong>:</h3><p>In a multivariate setting we consider multiple columns in the deduplication process and identify a golden record or form a cluster based on the requirement of the domain.</p><p>Ex: for the Vendor master data management as business case we need to identify different forms of the naming conventions of the vendors given by the data steward and project them as single entity. In this typical example we name different names of the companies to single owner.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/472/1*MGU1OqW3PxjEzyQi4Ig9Aw.png" /><figcaption>Sample Company data with address as attributes</figcaption></figure><p>With the data of different forms of company names the algorithm has to perform deduplication at the primary column level by using the supporting columns as a multivariate approach.</p><p>To perform deduplication the comparison has to happen between the records of the current database, i.e. with the above example 10 records has to be compared with the same 10 records Cartesian product of the set. 10X10 atmost 90 comparisons are required to get the best candidate pairs out the table.</p><p>This article also suggests a optimized way of filtering the unnecessary comparisons with a strategic methods used according the Use cases.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*dH12lW7sEjN15vnEajQHAg.png" /><figcaption>Cartesian product or all possible candidate pairs of the data.</figcaption></figure><p>The above table is the replication of cartesian product of the candidate pairs. If a direct edit distance is calculated on the give table with multiple variables. The sim_* represents the similarity score of the columns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*hvXlT8ay-WrMisAyWY2ZNw.png" /><figcaption>Similarity Score calculated for all the pairs using Jaro-Winkler edit distance.</figcaption></figure><p>Setting weightage to the data columns from the domain knowledge certain weights can be provided to the columns so that the overall summation of the similarity score can be calculated and threshold can be set.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*1Pl8D_v2vCPVXqi46hPIsQ.png" /><figcaption>Given a weights of Name-50%, Country-25%,Address-25%</figcaption></figure><p>In the above table ‘overall_sim’ is the weighted overall similarity score of the candidate pair.</p><p>When a threshold is set in the overall_sim the rows with met the threshold can assumed as similar candidate pairs satisfied our requirement. There are multiple ways to form a cluster out the similarity table and give a normalized name to the formed cluster. Using Graph clustering technique use the threshold as weighted edges in the graph to form the clusters.</p><h3><strong>Graph Techniques in clustering</strong>:</h3><p>Graph clustering refers to clustering of data in the form of graphs. Two distinct forms of clustering can be performed on graph data. Vertex clustering seeks to cluster the nodes of the graph into groups of densely connected regions based on either edge weights or edge distances. The second form of graph clustering treats the graphs as the objects to be clustered and clusters these objects on the basis of similarity. The second approach is often encountered in the context of structured data.</p><p><strong>Graph-Based clustering uses the proximity graph</strong>:</p><p>– Start with the proximity matrix</p><p>– Consider each point as a node in a graph</p><p>– Each edge between two nodes has a weight which is the proximity between the two points</p><p>– Initially the proximity graph is fully connected</p><p>The formed can be pruned based on the outliers using the weights which represent the similarity between the nodes in the graph.</p><p>Sparsification techniques keep the connections to the most similar (nearest) neighbors of a point while breaking the connections to less similar points.</p><p>– The nearest neighbors of a point tend to belong to the same class as the point itself.</p><p>– This reduces the impact of noise and outliers and sharpens the distinction between clusters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*d7rPGhhxkcVsuUFcqEFfuQ.png" /><figcaption>Clustering the data and using edge cut.</figcaption></figure><p><strong>Reweighted Shared Nearest Neighbor Graph:</strong></p><p>Is the widely used technique for similarity graphs for pruning the graph, Removes edges from the SNN with weight less than τ and recalculated to the total weight of the cluster including and excluding the current less weighted node to the cluster.</p><p>Normalization or deduplication:</p><p>When the clusters are formed represent similar names grouped together, so giving a normalized name to represent the cluster from the names within the cluster.</p><h3><strong>Performance statistics:</strong></h3><p>Using Intel Core Processor (Broadwell) 64-bit, 8 cores, CPU MHz: ~2200 below are the run time for calculating the candidate pairs.</p><p>below is the overall time taken for execution of process <strong>without using filter approach</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/383/1*I0sPbkSe5YWNI-oqbsBfJg.png" /><figcaption>Total computation time without using filtering methods</figcaption></figure><p>Below is the overall time taken for execution of process <strong>with filter approach.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/380/1*Dk6r5PPjeoVMb9p33pOh1A.png" /><figcaption>Total computation time with filter approach</figcaption></figure><h3><strong>Filtering Methods:</strong></h3><p>There are certain methods like prefix filtering, subset filtering, q-gram filtering, cosine filtering and content based mismatched filtering. This will be explaining prefix filtering, cosine filtering and q-gram filtering.</p><p><strong>Prefix filtering</strong>:</p><p>A prefix-based filtering is proposed to quickly filter out candidate pairs that are guaranteed not to meet the (α) threshold. Prefix filtering helps in the finding the similar company names.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/381/1*UcSgvpTkGISs7gAgBH7Dww.png" /><figcaption>Using Prefix filter with l=4</figcaption></figure><p>This example If prefix length is considered as 4 edits then of the entities {Microsoft, micron, Macrosoft} could form a set of candidate pairs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/429/1*E2qL2jf_ZUrigLEsWqiBKg.png" /><figcaption>Total Computation time to filter the data.</figcaption></figure><p><strong>Cosine Filtering</strong>:</p><p>Using Cosine for filtering the two similar vectors are most widely used method we leverage this in the current context to modify the data to Dense sparse matrices and compare the vectors.</p><p>Drawback of using cosine filtering is a very large sparse matrix will be formed in case we are using a average number of records in the database some of the existing system architectures doesn’t support very large matrices. Using a SparsedottopN algorithm will efficiently remove the vectors which doesn’t fall in the threshold (α) and reduce the sparsity of the matrix size.</p><p><strong>Q-gram Filtering:</strong></p><p>This is the efficient solution for a high level edit distance filtering where the computations are reduced drastically compared to levenshtein edit distance algorithm. The method essentially relaxes the edit distance constraint to a weaker count constraint on the number of matching q-grams. Two q-grams match if they have the same token and their locations are within the edit distance threshold.</p><p>The pair has to share at least of provided threshold q-grams. ‘q’ is denoted as the sliding window of the string.</p><p>Let q=2 and threshold =3</p><p>T= ‘abccabcc’, S=’abbcabbc’</p><p>[ab,bc,cc,ca,ab,bc,cc][ab,bb,bc,ca,ab,bb,bc]</p><p>consider one string T, and it has only two out of seven q-grams that are not matched i.e.</p><p><strong>count filtering</strong> gives 5 matching q-grams. These will become the candidate pairs and ultimately verified by edit distance.</p><p>Using <strong>location filtering</strong> gives 3 matching q-grams these will also form a candidate pairs and can be verified by edit distance</p><p>q-gram filtering is mimic of the Jaccard similarity calculation but we leverage the location or index of the string for better computational accuracy.</p><h3><strong>Summary on experimental results:</strong></h3><p>- The execution time for the whole set of process is very high for traditional methods using to find string similarity, If the Data is too large to be processed prefer to run them in batch wise. Non of the existing algorithm didn’t perform well.</p><p>- As mentioned in the experiments the filtering method might not always reduce the complexity in time. If the filtered cluster is biased and has large amount of records to be processed, this computation might effect the performs of the algorithm.</p><p>- Different techniques of algorithm can be used as per the problem needs</p><p>If a short range of texts to be processed like company names or product names ‘prefix filtering’ and Jaro winkler.</p><p>If a text with high similarity is expected like the address comparisons then the single edit might show more penalty on the similarity score. Using Levinshtein edit distance is much preferable.</p><p>References:</p><ol><li>Eliminating fuzzy duplicates in data warehouses.(R. Ananthakrishna, S. Chaudhuri, and V. Ganti.)</li><li>A Primitive Operator for Similarity Joins in Data Cleaning. SSJoin by Microsoft Research(Surajit Chaudhuri, Venkatesh Ganti, Raghav Kaushik)</li><li>Querying multiple features in relational databases. (D. Chatziantoniou and K. Ross)</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6773b30eab7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Complete topics Overview of Data science and Analytics.]]></title>
            <link>https://hemanthkakarla.medium.com/complete-topics-overview-of-data-science-and-analytics-cc8720b728b2?source=rss-163bf74094b9------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc8720b728b2</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[datascience-certification]]></category>
            <category><![CDATA[course]]></category>
            <category><![CDATA[statistics]]></category>
            <dc:creator><![CDATA[Hemanth Kakarla]]></dc:creator>
            <pubDate>Mon, 30 Sep 2019 19:42:56 GMT</pubDate>
            <atom:updated>2019-09-30T20:01:32.348Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Complete Overview of Data science and Analytics</strong>.</h3><p>Data Science and Analytics is a multidisciplinary field that requires a range of skills in statistics, mathematics, predictive modeling and business analysis. A Data scientist professional should feel at ease to build the algorithms necessary, work with various data sources (often in disparate forms) and provide the solution.</p><p>Below is the brief intro of topics in Statistics, Analytics, Machine Learning and Visualization.</p><p>To master Data science it is suggested to dig in details of each topic.</p><blockquote><strong>Descriptive Statistics</strong>:</blockquote><ul><li><em>Data exploration (histograms, bar chart, box plot, line graph, scatter plot)</em></li><li><em>Qualitative and Quantitative Data</em></li><li><em>Measure of Central Tendency (Mean, Median and Mode)</em></li><li><em>Measure of Positions (Quartiles, Deciles, Percentiles and Quantiles)</em></li><li><em>Measure of Dispersion (Range, Median, Absolute deviation about median</em></li><li><em>Variance and Standard deviation), Anscombe’s quartet</em></li><li><em>Other Measures: Quartile and Percentile, Interquartile Range</em></li></ul><blockquote><strong>Inferential Statistics</strong>: Understanding the data attributes and distributions</blockquote><ul><li><em>Sample Vs population</em></li><li><em>Sample techniques (simple, stratified, clustered, random)</em></li><li><em>Sampling Distributions, Parameter Estimation</em></li><li><em>Test of Hypothesis (Concept of Hypothesis testing, Null Hypothesis and Alternative Hypothesis)</em></li><li><em>Cross Tabulations (Contingency table and their use, Chi-Square test, Fisher’s exact test)</em></li><li><em>One Sample t test (Concept, Assumptions, Hypothesis, Verification of assumptions, Performing the test and interpretation of results)</em></li><li><em>Independent Samples t-test</em></li><li><em>Paired Samples t-test</em></li><li><em>One-way ANOVA (Post hoc tests: Fisher’s LSD, Tukey’s HSD).</em></li><li><em>z-test and F-test</em></li></ul><blockquote><strong>Initial Data Analysis</strong>:</blockquote><ul><li><em>Relationship between attributes:</em></li><li><strong><em>Covariance</em></strong></li><li><strong><em>Correlation Coefficient</em></strong></li><li><strong><em>Chi Square</em></strong></li><li><em>Measure of Distribution (Skewness and Kurtosis), </em><strong><em>Box</em></strong><em> and </em><strong><em>Whisker Plot</em></strong><em> (Box Plot and its parts, Using Box Plots to compare distribution) and other statistical graphs.</em></li></ul><blockquote><strong>Probability</strong></blockquote><ul><li><em>Probability (Joint, marginal and conditional probabilities)</em></li><li><em>Probability distributions (Continuous and Discrete)</em></li><li><em>Density Functions and Cumulative functions</em></li></ul><blockquote><strong>Data Pre-Processing, Preparation, Transformation, Quality: </strong>Most of the time is spend on data pre-processing and cleansing steps.</blockquote><ul><li><em>Data Munging, Wrangling</em></li><li><em>Data imputation</em></li><li><em>Data Transformation (minmax, log transform, z-score transform etc.,).</em></li><li><em>Binning, Classing and Standardization.</em></li><li><em>Outlier/Noise&amp; Anomalies.</em></li><li><strong><em>Handling Text Data and Pre-processing:</em></strong></li></ul><ol><li><em>Bag-of-words</em></li><li><em>Regular Expressions</em></li><li><em>Sentence Splitting and Tokenization</em></li><li><em>Punctuations and Stop words, Incorrect spellings</em></li><li><em>Properties of words and Word cloud</em></li><li><em>Lemmatization and Term-Document TxD computation</em></li></ol><blockquote><strong>Predictive analytics: </strong>predict trends and behavioral patterns<strong> </strong>and classification.</blockquote><p><strong>Linear Regression</strong></p><ul><li><em>Regression basics: Relationship between attributes using Covariance and Correlation</em></li><li><em>Relationship between multiple variables: Regression (Linear, Multivariate) in prediction.</em></li><li><em>Residual Analysis</em></li><li><em>Identifying significant features, feature reduction using AIC, multi-collinearity</em></li><li><em>Non-normality and Heteroscedasticity</em></li><li><em>Hypothesis testing of Regression Model</em></li><li><em>Confidence intervals of Slope</em></li><li><em>R-square and goodness of fit</em></li><li><em>Influential Observations — Leverage</em></li></ul><p><strong>Multiple Linear Regression</strong></p><ul><li><em>Polynomial Regression</em></li><li><em>Regularization methods</em></li><li><em>Lasso, Ridge and Elastic nets</em></li><li><em>Categorical Variables in Regression</em></li></ul><p><strong>Classification Non-Linear Regression</strong></p><ul><li><em>Logit function and interpretation</em></li><li><em>Logistic regression</em></li><li><em>Types of error measures</em><strong><em> (ROCR)</em></strong></li><li><em>Model performance</em></li><li><em>Model validation</em></li><li><em>Model interpretation</em></li><li><em>Accuracy</em></li><li><em>Precision</em></li><li><em>Recall</em></li><li><em>F1 score</em></li><li><em>Confusion matrix</em></li></ul><p><strong>Forecasting models</strong></p><ul><li><em>Trend analysis</em></li><li><em>Cyclical and Seasonal analysis</em></li><li><em>Smoothing</em></li><li><em>Moving averages</em></li><li><em>Box-Jenkins</em></li><li><em>Holt-winters</em></li><li><em>Auto-correlation</em></li><li><em>ARIMA</em></li></ul><blockquote><strong>Model and Data modification techniques</strong>: Pre or Post model creation these techniques can be used to improve the performance of the model.</blockquote><ul><li><em>Validation Techniques (Cross-Validations)</em></li><li><em>Feature Reduction/Dimensionality reduction</em></li><li><em>Principal components analysis (Eigen values, Eigen vectors, Orthogonality)</em></li><li><em>L1 and L2 Regularization</em></li><li><em>Gradient Descent</em></li></ul><blockquote><strong>Clustering</strong></blockquote><ul><li><em>Different clustering methods (Distance, Density, Hierarchical, Spectral)</em></li><li><em>Hierarchical Clustering</em></li><li><em>Divisive and Agglomerative Clustering</em></li><li><em>DBSCAN</em></li><li><em>K-means Clustering</em></li></ul><blockquote><strong>Classification: </strong>Classification is the mostly widely used in business problems.</blockquote><p><strong>Naïve Bayes Classifier:</strong></p><ul><li><em>Model Assumptions, Probability estimation</em></li><li><em>Required data processing</em></li><li><em>M-estimates, Feature selection: Mutual information</em></li></ul><p><strong>K-Nearest Neighbors:</strong></p><ul><li><em>Computational geometry; Voronoi Diagrams; Delaunay Triangulations</em></li><li><em>K-Nearest Neighbor algorithm; Wilson editing and triangulations</em></li><li><em>Aspects to consider while designing K-Nearest Neighbor</em></li></ul><p><strong>Support Vector Machines:</strong></p><ul><li><em>Linear learning machines and Kernel space, Making Kernels and working in feature space</em></li><li><em>SVM for classification and regression problems.</em></li></ul><p><strong>Decision Trees:</strong></p><ul><li><em>Gini index</em></li><li><em>Entropy or Information gain</em></li><li><em>Chi-square</em></li></ul><p><strong>Ensembles methods:</strong></p><ul><li><em>Bagging &amp; boosting and its impact on bias and variance</em></li><li><em>Random forest</em></li><li><em>Gradient Boosting Machines and XGBoost</em></li></ul><blockquote><strong>Data Visualization:</strong> Visualizing and Communicating clearly and effectively about the patterns we find in data is a key skill.</blockquote><ul><li><em>Maps, Charts, Graphs</em></li><li><em>Story telling</em></li><li><em>Identifying the opportunities</em></li><li><em>Tableau, Microsoft Excel, QlikView, Power BI.</em></li></ul><p><strong>Next:</strong></p><ul><li>Topics in Neural networks and deep learning.</li><li>Data science project overview and data process flow.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc8720b728b2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Course structure for Data Science Freshers and Beginners.]]></title>
            <link>https://medium.com/datasciencekickstartcareer/course-structure-for-data-science-freshers-and-beginners-6555941a305e?source=rss-163bf74094b9------2</link>
            <guid isPermaLink="false">https://medium.com/p/6555941a305e</guid>
            <category><![CDATA[beginner]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[course]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Hemanth Kakarla]]></dc:creator>
            <pubDate>Fri, 27 Sep 2019 19:03:41 GMT</pubDate>
            <atom:updated>2019-09-27T19:03:41.558Z</atom:updated>
            <content:encoded><![CDATA[<p>If you are looking for a decent start in data science. Then this page will provide you the course structure and resource links. Learn the basic concepts to get into the data science world.</p><blockquote><strong>Prologue</strong>: Knowledge sharing should be free, there are more than enough resources available on internet to learn data science for <strong>Free, </strong>The below links and resources shared might require you to sign up for courses which are free.</blockquote><p><strong>Machine learning course by </strong><a href="https://www.coursera.org/instructor/andrewng"><strong>Andrew N</strong></a><strong>G:</strong></p><p>Note: This is the best course over the internet giving intuition about machine learning and linear algebra and also covers all the areas in depth and also might require your efforts in assignment submission.</p><p><a href="https://www.coursera.org/learn/machine-learning">Course Link</a> or (<a href="https://www.coursera.org/learn/machine-learning">https://www.coursera.org/learn/machine-learning</a>)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r6BZezyURZaGZoSEaTaq7A.png" /></figure><p>Details: The above course link covers all the major concepts of Machine learning and hands on with assignments.</p><p>The course is hands on with <strong>Octave/Matlab</strong> because the understanding of mathematical notations would be easier in Octave/Matlab than <strong>Python/R</strong> at this point.</p><p>Topics covered:</p><ol><li>Linear Algebra: Matrices &amp; Vectors and operations.</li><li>Linear Regression: Model Representation, Cost function, Gradient Descent.</li><li>Linear Regression Multiple variables: Gradient Descent for multiple features, features for polynomial regression.</li><li>Logistic Regression: Classification, Hypothesis representation, Decision boundary, Cost function and Gradient Descent, Advanced optimization, Multi-Class classification.</li><li>Regularization: understanding over-fitting, Cost function, Regularized linear function, Regularized logistic function.</li><li>Intro Neural Networks: Neural Nets intuition and model representation.</li></ol><p><strong>Other readings:</strong></p><p><strong>Google </strong>also provides a self learning guide with videos and also the course material with examples in brief.</p><p><strong>Link</strong>: <a href="https://developers.google.com/machine-learning/crash-course/ml-intro">https://developers.google.com/machine-learning/crash-course/ml-intro</a></p><p>Google resource will be really helpful for revision and clarification.</p><p>The above 2 resources will provide a very good foundation for Machine learning and Data science.</p><p><strong>Myths about Data science:</strong></p><ol><li>Data science don’t require programming or coding skills. most of the code is already written into modules, just need to implement using APIs and understand the algorithm.</li><li>Need not be a pro in stats and math. Basic math and linear algebra which is thought in school will suffice.</li><li>Doesn’t require high configuration machines. Most of the datasets for practice are small enough and will be easy to understand the patterns.</li><li>Need not be computer science graduate. Data science is not domain specific, it is across all the domains.</li><li>Need not spend lot of money to learn Data science. Most of the best tutorials are free and open to internet.</li></ol><p>The above are only myths and having them is only a add-on.</p><p><strong>Useful resources links:</strong></p><p>If you are new to programming, I would suggest to learn basic programming language like <strong>C/Java/Python</strong> and Object oriented concepts which will make your life easier to understand the already built modules.</p><p>Google provides a free python course contains videos and material which covers most of the python concepts like:</p><p>Data types, Strings, Lists, sortings, Dicts, Files, Regex, Utils.</p><p><strong>Link</strong>: <a href="https://developers.google.com/edu/python/">https://developers.google.com/edu/python/</a></p><blockquote>This article contains only the resources links and intro to data science, Will publish another article which will contain all the list of concepts in Machine learning for a quick overview.</blockquote><p><strong>Next</strong>:</p><ol><li>Real time analytics and Datasets.</li><li>Hands on with datasets on Kaggle.</li><li>Deeper dive to Machine learning.</li><li>Neural Networks and Deep learning.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6555941a305e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datasciencekickstartcareer/course-structure-for-data-science-freshers-and-beginners-6555941a305e">Course structure for Data Science Freshers and Beginners.</a> was originally published in <a href="https://medium.com/datasciencekickstartcareer">datasciencekickstartcareer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>