Reinfrocement Learning with full_model

[Ulitochka]

Hello.

If I use full_model.th in reinforcment learning script: python reinforce.py
--cuda
--alice_model_file full_model.th
--bob_model_file full_model.th
--output_model_file rl_model.th
--context_file data/negotiate/selfplay.txt
--temperature 0.5
--verbose
--log_file rnn_rl.log
--sv_train_freq 4
--nepoch 4
--selection_model_file selection_model.th
--rl_lr 0.00001
--rl_clip 0.0001
--sep_sel

I have this:

Traceback (most recent call last):
File "/home/.../end-to-end-negotiator/src/reinforce.py", line 169, in
main()
File "/home/.../end-to-end-negotiator/src/reinforce.py", line 163, in main
reinforce.run()
File "/home/.../end-to-end-negotiator/src/reinforce.py", line 51, in run
self.dialog.run(ctxs, self.logger)
File "/home/.../end-to-end-negotiator/src/dialog.py", line 171, in run
out = writer.write(max_words=words_left)
File "/home/.../end-to-end-negotiator/src/agent.py", line 1661, in write
_, lat_h, log_q_z = self.model.forward_prediction(self.cnt, self.mem_h, sample=self.train)
File "/home/.../end-to-end-negotiator/src/models/latent_clustering_model.py", line 541, in forward_prediction
z = q_z.multinomial().detach()
TypeError: multinomial() missing 1 required positional arguments: "num_samples"

torch==1.0.0