README.md

Master-Project

Requirements

1. Python3
2. Tensorflow
3. Numpy
4. Tesseract
5. Datasets
6. gensim The steps for training CNN by learning embeddings from scratch:

#Train: python train.py

Running Pruning Techniques: For pruning we need to restore the checkpoint, which is saved in runs folder

#For running pruning by calculating variance

python prunebyvariance.py --prune

#For making new computational graph for fine tuning and pruning iteratively

python prunebyvariance --make_graph

#For running pruning by calculating absolute weight

python prunebyweight.py--prune

#For making new computational graph for fine tuning and pruning iteratively

python prunebyweight.py--make_graph

#For running pruning by calculating L2norms

python prunebyl2norm.py--prune

#For making new computational graph for fine tuning and pruning iteratively

python prunebyl2norm.py --make_graph

#For running pruning by resetting negative filter weights as zero

python negativefilters.py--prune

#For making new computational graph for fine tuning and pruning iteratively

python negativefilters.py --make_graph

The steps for training CNN by using fasttext:

Download fasttext pretrained word2vec file.

Generate fasttext_vocab_en.dat, fasttext_embedding_en.py

python util_fasttext.py

#Train

python train.py --pre_trained

For running pruning techniques for the model same procedure has to be followed as above.

References :

1. https://github.com/dennybritz/cnn-text-classification-tfy