Replicate the RL paper parameters and to run

Question

Replicate the RL paper parameters and to run

Closed this issue 5 years ago · 4 comments

ayush2051 commented 5 years ago

@rpoli40 Were u able to run it and replicate as given in the RL paper. I tried it but ran into an error
ValueError: Dimensions must be equal, but are 2269 and 2446 for 'train_loss/mul_4' (op: 'Mul') with input shapes: [?,?,2269], [?,?,2446].

I kept the advanced encoding & context_as_set FLAG as TRUE. Am I doing it right or do I need to change any other parameters in the flags file.

Answer 1 · 2019-07-30T13:02:15.000Z

Hi, could you please confirm exactly what command you ran to trigger this error?

Answer 2 · 2019-08-01T15:23:10.000Z

Yeahh Tom I ran the script as ./train.sh --advanced_condition_encoding --nocontext_as_set --model_type RL-MALUUBA with all the flags as set earlier to defaults. Am I doing it right or do I need to change any other flags?

Answer 3 · 2019-09-15T16:14:35.000Z

@ayush2051 I just tried running with those flags and it worked OK for me. Did you find the problem?

Answer 4 · 2019-10-11T08:51:56.000Z

@tomhosking thanks a lot. Sorry couldn't reply, it worked finally.