Issues about generation from test dataset

[zhenli]

Hello! What do I do if I want to generate responses from a test dataset using a trained model? Thanks!

[Qian]

If you want to evaluate the response using predefined metrics such as F1, you could use eval_f1.py to achieve it;
If you want to obtain all generated responses, you could modify the option(line43-line49) in eval_f1.py as:

  parser.set_params(
      model='agents.transmitter.transmitter:TransformerAgent',
      model_file=model_name,
      gpu=1,
      batchsize=1,
      beam_size=2,
      report_freq=1.0
  )

And the Text and Prediction will be printed as following:

TEXT:  <t1start> that is great , me too ! i'm married and my husband and i've 2 children . <t1end>
PREDICTION:  i ' m a nurse and i love to spend time with my family 

And you could extract them from the log.

[zhenli]

Thanks!