TypeError in wsj/s5/steps/libs/nnet3/train

[mthrok]

Hi

Sorry if this is duplicated issue.

It seems that archive_index argument in get_multitask_egs_opts function is expected to be integer,

egs_suffix = ".{0}".format(archive_index) if archive_index > -1 else ""

but compute_preconditioning_matrix is passing a string object.

    multitask_egs_opts = common_train_lib.get_multitask_egs_opts(
        egs_dir,
        egs_prefix="cegs.",
        archive_index="JOB",
        use_multitask_egs=use_multitask_egs)

This causes WSJ example to fail with the following error message.

Traceback (most recent call last):
  File "steps/nnet3/chain/train.py", line 622, in main
    train(args, run_opts)
  File "steps/nnet3/chain/train.py", line 428, in train
    use_multitask_egs=use_multitask_egs)
  File "steps/libs/nnet3/train/chain_objf/acoustic_model.py", line 395, in compute_preconditioning_matrix
    use_multitask_egs=use_multitask_egs)
  File "steps/libs/nnet3/train/common.py", line 74, in get_multitask_egs_opts
    egs_suffix = ".{0}".format(archive_index) if archive_index > -1 else ""
TypeError: '>' not supported between instances of 'str' and 'int'

Changing > to != seems to solve this.

diff --git a/egs/wsj/s5/steps/libs/nnet3/train/common.py b/egs/wsj/s5/steps/libs/nnet3/train/common.py
index d052c78..b8e0322 100644
--- a/egs/wsj/s5/steps/libs/nnet3/train/common.py
+++ b/egs/wsj/s5/steps/libs/nnet3/train/common.py
@@ -71,7 +71,7 @@ def get_multitask_egs_opts(egs_dir, egs_prefix="",
         "valid_diagnostic." for validation.
     """
     multitask_egs_opts = ""
-    egs_suffix = ".{0}".format(archive_index) if archive_index > -1 else ""
+    egs_suffix = ".{0}".format(archive_index) if archive_index != -1 else ""

     if use_multitask_egs:
         output_file_name = ("{egs_dir}/{egs_prefix}output{egs_suffix}.ark"

I am not sure if this is the right thing to do. Could you advice?

[danpovey]

Resolved, thanks.

[mthrok]

Thanks!