nesl/MultimodalFusedClassifier

This folder contains all the python files for running training, prediction, and evaluation for the reclassification.

To run it, you should probably intialize a virtual environment - I'll set up a packages file for easy install later

Once the required packages are downloaded, you can just run "reclassify.py" to perform training, prediction, and evaluation all in one go.

data_chunker and data_utils contain all the data manipulation and fileIO needed
train_feature_extractors and test_feature_extractors trains each modality to classify the activity, but only saves output from an earlier layer
train_and_predict_fusion_classifier trains the fusion network given the higher level features from before
train_and_predict_feature_generators trains each sensor modality to reproduce the higher level features from the scores outputted from the fusion classifier.


# TODO:
-Still need to add more options to the reclassification (i.e. which activities to train/predict/evaluate on)