Fast Text unsupervised model loss with Python API

问题

Is there any way to get the model loss for the unsupervised training of models using Fast Text with the python API? At the moment I am doing the training using the C++ model and loading it using the Python API.

For e.g., I first run the following code to tweak hyper parameters

./fasttext skipgram \
-input /data/cleaned.txt \
-output /models/cleaned-model \
-epoch 12000 \
-minCount 2 \
-ws 3

The command-line interface gives an estimate of the loss like so:

Progress: 100.0% words/sec/thread:  103006 lr:  0.000000 loss:  1.803622 ETA:   0h 0m

However, doing the same using the Python API:

import fastText
model = fastText.train_unsupervised('/data/cleaned.txt', 
                                    epoch=12000, 
                                    minCount=2, 
                                    ws=3)

This trains but does not output the loss? I checked increasing the verbosity level verbosity=3 in the training function options but nothing happens. Is this a missing feature or something I am missing?

回答1:

If you run the python script from the shell, it prints the desired output.

Maybe you're using Jupyter Notebook. In this case, currently there's not a simple way to see the output in the notebook.

来源：https://stackoverflow.com/questions/54773706/fast-text-unsupervised-model-loss-with-python-api