logs21: Steps to make small RL work

Log Type	Detail
1: What specific output am I working on right now?	See if this small set of RL is working.
2: Thinking out loud - hypotheses about the current problem - what to work on next - how can I verify	- Reward 1.0 when len == 8 or len == 0 otherwise reward -1.0
3: A record of currently ongoing runs along with a short reminder of what question each run is supposed to answer	Run1 & Run2
4: Results of runs and conclusion	Run1 Eventually converge to produce len == 1
5: Next steps
6: mega.nz

Run 1

sampled lengths=[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] 0 sampled lengths=[0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0] 0 sampled lengths=[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] 0 sampled lengths=[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] 0 sampled lengths=[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] 0

{'machine': 'client2', 'batch_size': 16, 'num_units': 256, 'num_layers': 2, 'vocab_size': 34, 'embedding_size': 40, 'learning_rate': 0.1, 'learning_rate_decay': 0.99, 'use_attention': True, 'encoder_length': 8, 'decoder_length': 8, 'max_gradient_norm': 5.0, 'beam_width': 2, 'num_train_steps': 5000, 'model_path': 'model/tweet_small'}