*Character level language generation using GANs without pre-training-----------------

*The project is divided into the following files:------------------------------------

parameters.py -- All parameters of the project are defined here. Most of them are defined in argument form, so they can be passed in when calling a script.
train_model.py -- Implements a simple Curriculum Learning and Variable length learning.
text_generation.py -- Code to generate text from a trained model.
evaluate_result.py -- Evaluate the results store by text_generation.py
define_objevtive_function.py -- Define the objective function for the improved Wasserstein GAN training. Based on https://arxiv.org/abs/1704.00028, by Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron Courville
model.py -- Generator and Discriminator definitions.
train_a_sequence.py -- Main training code. train_model calls this for every sequence length.
helper_methods.py -- Save/load pickled files, load dataset wrapper code etc.
language_helpers.py -- Taken from https://github.com/igul222/improved_wgan_training/blob/master/language_helpers.py


*The following packages are required:------------------------------------------

1. Python 2.7
2. Tensorflow >= 1.1
3. Scipy
4. Matplotlib


*To run----------------------------

1. First download the sanitized NIPS dataset from : https://www.dropbox.com/s/h23e2381adwndt0/nips.zip?dl=0
2. Place it in the dataset folder. Have a look at params.py for exact location(-DATA_DIR)
3. Change any params in parameters.py
4. Call train_model.py with appropriate parameters.(If different from parameters.py)
5. The model will be save in logs/[model name according to current time]/seq-[sequence length]/
6. To generate text, Call generate_text.py -CKPT_PATH=[model_path from above]/ckp -BATCH_SIZE=[length of text you want to generate]
7. Your results will be save in output/samples.txt

-----------------------------------------------------------------------------