sachinsingh3107 / mlia-examples

Python and R Examples

mlia-examples

Python and R code to do miscellaneous Data Mining tasks. Python code started with the Machine Learning in Action book by Peter Harrington, and since then have moved to using scikit-learn.

src/book/

Examples from Peter Harrington's Machine Learning in Action (MLIA).

The code is not a direct copy of whats in the book or in the github site, but does the same thing. I have tried to make the code shorter and (at least to me) more understandable.

Data files are not copied over. You will find the data files (of the same name as in the code) in the book's source code site on GitHub.

src/event_reco/

My solution for the Kaggle Event Recommendation Challenge. Given a training set of users who are interested/not_interested in events, and user and event metadata and some social (user's friends information) metadata, the objective is to order the events in the test set per user, so events with higher probability are recommended first. The approach I have taken is to construct a set of different kinds of recommenders, and construct features from the recommender scores, then build a predictive model (SGD) to predict the value of interested. The distance from the separating hyperplane is the measure of the likelihood of recommendation.

src/salary_pred/

A solution to the Adzuna Salary Prediction Challenge on Kaggle. Given a set of job ads, predict the salary for the job. The solution is incomplete and predicts very poorly. Uses NLTK and Scikit-Learn.

src/bird_strikes

A set of visualizations using R. The data was provided as part of a programming assignment using Tableau in the Coursera Introduction to Data Science course. I wanted to see if I could produce something equivalent using R.

src/network_analysis

Partial/In-progress solutions to Coursera's Social Network Analysis course that I took late 2013. My approach was to solve as many questions as I could without depending on visualization tools such as Gephi or Netlogo (and sometimes my solutions are incorrect) and relying on NetworkX and IGraph.

src/enron_network

Code for my Peer Programming Assignment for the Social Network Analysis course on Coursera. I build a graph of people based on the from and to email addresses in the dataset, and investigate various centrality measures and other indicators of node importance (PageRank and HITS Authority) to see how effective they are at finding the guilty (tested against a list of 12 individuals who were found guilty).

src/recsys_eval

Pandas code to answer questions about a evaluation run for multiple recommenders across multiple user neighborhood sizes.

src/yelp_ufc

Scikit-Learn code to build and evaluate 3 Naive Bayes classifers to predict if a review is useful, funny or cool. Uses data from the Yelp challenge on Kaggle that requires solutions to also predict number of votes that a review will receive in each category. I wanted to figure out if Sentiment Analysis could be treated as a classification problem (rather than use polarity keywords).

src/proc_outliers

Exploratory data analysis of Medicare/Mediclaim data to convert claim codes into a univariate distribution of cluster diameters, and apply univariate outlier detection techniques to find abnormal (possibly fraudulent) claims.

src/mlpas

Programming Assignments from Andrew Ng's Machine Learning Class on Coursera, but not solved using first principles as the course intends. I have used ML toolkits I am familiar with (or became familiar with) such as Numpy, Scipy, Matplotlib, Sklearn, Pybrain, etc, to solve the PAs.

src/mtcrawler

Using ScraPy for building a small focused crawler.

src/hartley

Random Forest Classifier to detect interesting surface features on planetary objects.

src/citrus_pca

Data Analysis of Reflectance Spectrum, Principal Component Analysis (PCA), and Linear Discriminant Analysis (LDA) Classifier to predict Citrus vs Non-Citrus and different varieties of Citrus plants.

src/skin_colors

Simulation of evolution of skin colors using 3 alleles. Based on Quora answer by Pippi Maria Groving.

src/moviequotes

Using Vowpal Wabbit to build a simple binary text classifier to predict memorableness of movie quotes. Dataset from Cornell Movie Quotes Corpus, distributed along with the paper - You Had me at Hello - How Phrasing affects memorability, by Danescu-Niculescu-Mizil, et al.

src/d2v_tagpred

Predicting new movie tags for movies given the plotline and human assigned tags, using Gensim's Doc2Vec implementation. Blog post here.

src/student-alcohol

Predicting student alcohol consumption given 32 student attributes, using Gradient Boosted trees from XGBoost. Blog post here.

src/hyperparam-opt

Random Search vs Grid Search - Blog post here.