readme.md

This is a reverse dictionary application using the Webster's unabridged dictionary based on

1. LSTM
2. RNN (Recursive Neural Networks)

Installation:

1. Python 3 Dependencies: Tensorflow

2. Syntaxnet

3. Docker (optional; for running syntaxnet)

Preparing data

1. Make a folder named data in the root directory
2. Place the websters dictionary from this link: https://raw.githubusercontent.com/matthewreagan/WebstersEnglishDictionary/master/WebstersEnglishDictionary.txt into the data folder
3. Rename the file as websters_def.txt
4. From data_preprocess run make_data.py

For LSTM: To randomise and augment the data run make_random.py in the data_preprocess folder

For RNN: To make the parsed trees run make_parse_tree.py (config.py in the root directory contains the input and output file locations)

Training

1. LSTM In ml folder run train_log_loss.py

2. RNN In ml folder run train_recursive_nn.py

3. RNN 2 In ml folder run train_recursive_nn_2.py

config.py contains the parameters and the model and data file paths

Inference/Test

1. LSTM In test folder run test_end2end.py

2. RNN In ml folder run test_parsed.py

3. RNN 2 In ml folder run test_parsed_2.py

TODO

In model_recursive_nn_2.py in getWeightage take the product of the outputs instead of maxpool