`mode="floret"` not supported argument for `floret.train_supervised(...)`?

[narayanacharya6]

Training method floret.train_supervised(...) breaks when mode="floret" is used as a kwarg. Is this expected behavior or does the mode argument only make sense for train_unsupervised(...)?

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[102], line 1
----> 1 model = floret.train_supervised(
      2     input=FASTTEXT_TRAIN_DOCS_PATH,
      3     mode="floret",
      4     lr=0.1,
      5     epoch=10,
      6     dim=300,
      7     wordNgrams=2,
      8     ws=20,
      9     pretrainedVectors="wiki-news-300d-1M-subword.vec"
     10 )

File ~/.local/lib/python3.8/site-packages/floret/floret.py:548, in train_supervised(*kargs, **kwargs)
    534 supervised_default.update({
    535     'lr': 0.1,
    536     'minCount': 1,
   (...)
    540     'model': "supervised"
    541 })
    543 arg_names = ['input', 'lr', 'dim', 'ws', 'epoch', 'minCount',
    544              'minCountLabel', 'minn', 'maxn', 'neg', 'wordNgrams', 'loss', 'bucket',
    545              'thread', 'lrUpdateRate', 't', 'label', 'verbose', 'pretrainedVectors',
    546              'seed', 'autotuneValidationFile', 'autotuneMetric',
    547              'autotunePredictions', 'autotuneDuration', 'autotuneModelSize']
--> 548 args, manually_set_args = read_args(kargs, kwargs, arg_names,
    549                                     supervised_default)
    550 a = _build_args(args, manually_set_args)
    551 ft = _floret(args=a)

File ~/.local/lib/python3.8/site-packages/floret/floret.py:507, in read_args(arg_list, arg_dict, arg_names, default_values)
    505     arg_name = param_map[arg_name]
    506 if arg_name not in arg_names:
--> 507     raise TypeError("unexpected keyword argument '%s'" % arg_name)
    508 if arg_name in ret:
    509     raise TypeError("multiple values for argument '%s'" % arg_name)

TypeError: unexpected keyword argument 'mode'

[adrianeboyd]

Sorry for the late reply!

We've mainly developed/tested this for unsupervised training, but as far as I know it should work in supervised mode, too.

It looks like I didn't add it to the python wrapper, which is why you're seeing this error. If you haven't already, if you want to try it out I think it should work from the command-line instead with floret supervised -mode floret ...