Unsupervised Machine Translation implementation(for self-learning)

TODO

• imple add_noise function
• Need 3 neural nets: lang A autoencoder, lang B autoencoder, discriminator a
• Assume A -> B translation
• Three Loss functions: autoencoder(cross-ent for A,B), translation loss(A->B), discrimnator loss
• $ M_t = decB_t(encA_t(x)) $
• Need to jumpstart process by using naive methods for initialization(eg. fastText?)