MystikHub / berkeley-time-series

Time series prediction using python and scikit learn

Berkeley Earth Climate Data Time Series Analysis

Time series prediction using python and scikit learn

Dependencies

This project uses Python and requires the following packages from pip:

• matplotlib
• numpy
• pandas
• sklearn

Preparing the feature sets

Download the data from Berkeley Earth and extract the "GlobalLandTemperaturesByCity.csv" file into the repository's directory. Then, use python3 make-feature-sets.py to export the various different feature set files into a feature-sets directory.

Running the machine learning models

You can then configure and run whichever code you want for each model (i.e. cross-validation, training, evaluation, etc.) in each file. These are generally configured by a string near the top of each model's file.

Here's what each file does:

• lasso.py Cross-validates the lasso regression penalty and trains and evaluates a lasso regression model against an "average of features" baseline predictor
• linear.py Trains and evaluates a linear regression model against an "average of features" baseline predictor
• load_feature_sets.py Contains some useful functions used by the tree models for loading the feature data generated by make_feature_sets.py
• make_feature_sets.py Reads the entire Berkeley Earth data set and writes each country and city's feature set files into a feature-sets directory
• ridge.py Cross-validates the ridge regression penalty and trains and evaluates a ridge regression model against an "average of features" baseline predictor