CS221: Emergency Call Analysis on San Francisco Fire Department dataset

Downloading raw data and generate datasets

This is optional step, since processed data are already part of source tree

1. Run download_datasets.py from root folder.
2. Then run make_ds/sf-fire-build-ds.py

Vanilla RNN

Run vanilla_rnn/rnn_predict_week.py. This will just do "creative" prediction for one week on test data. To train again constants at the top of the scripts needs to be changed: 'rebuild_artifacts' set to True.

LSTM RNN

Run lstm_rnn/sf-fire-lstm.py to execute RNN and LSTM time series prediction. The script runs training, performs test error validation and draws target vs prediction graph. Optional arguments:

-n - number of neurons (default=100)

-l - number of layers (deafult=3)

-m - type of network, rnn or lstm (default rnn)

Clustering

Run clustering by running clustering/location_clustering.py.

Optional parameter '-n' - number of clusters (default = 10)

Example: python location_clustering.py -n 15

Classification

For classification, just go to classification folder and run python classification.py
The ML model will be developed based on small portion of data