experiments

Visual Story Telling

1) Check if the learning rate was the cause of not achieving an overfiting
2) Check if pre-trained embedding layers was the cause of not achieving an overfiting
3) Check if the masking layer was the cause of not achieveing an overfitting

4) Try with masking layer in the encoder model
5) Try with 3 layers wit latent dimension of 512
6) Try with 2 layers with latent dimension of 1024

7) Try with 3 layers with latent dimension of 1024
8) Try adding dropout (read about effecient dropout in RNN networks), should we add them to the reccurent layer or the input layer.
9) Try to reverse the image embeddings
10) Shuffle the training samples

#train-
#time: 14:59-15:13
#bleu: 0.5104660537890634
#meteor: 0.17082049527670987
#valid-
#time_bleu:  13:59-14:00
#bleu:  0.4790176691234322
#meteor_time: 15:30-22:00
#meteor: 0.06073113164884557
#test-
#time_bleu: 15:22-15:23
#bleu: 0.4769828372026272