Playing with Data

This repository features a collection of Jupyter notebooks exploring data science concepts.

• best_practices.ipynb (nbviewer) demonstrate best practices and model evaluation metrics. Notebook topics include:
  • pipelining
  • learning curves
  • validation curves
  • confusion matrix
  • receiver operator characteristics (ROC and AUC)
• digits with bokeh.ipynb (nbviewer) demonstrates using Bokeh to visualize the mNIST digits data.
• sci-data_with_datashader.ipynb (nbviewer) demonstrates using Datashader to accurately plot tens-of-thousands to millions of data points.
• mnist_shallow_learning.ipynb (nbviewer) is [the beginnings of] a machine learning model (not using deep learning) to classify handwritten digits using either supervised or unsupervised learning.
• mnist_deep_learning.ipynb (nbviewer) is [the beginnings of] a reproduction of Mason Victor's SciPy 2016 lightning talk demonstrating a deep learning model (using theano together with keras) to classify handwritten digits.

Subdirectories within this repo contain the following:

• hands_on_machine_learning/ contains [the beginnings of] a collection of notebooks reproducing/exploring the concepts presentid in Aurélien Géron's book Hands-On Machine Learning with Scikit-Learn and TensorFlow.
• bayes_problems/ contains a series of sample Bayes' theorem problems and solutions.
• playlist_predictor/ contains [the skeleton of] a project to use supervised machine learning to build a playlist of songs based on the consistentency of seed songs.