Training ended after 3 epochs even though 5 epochs is given as parameter about deepnl HOT 5 CLOSED

Hi @attardi

Is this expected behavior ?

from deepnl.

It is called early stopping: if there is no progress in an epoch, the train stops.
In your case you had 48600 corrections skipped, i.e. examples that where already correctly classified and were not used.

from deepnl.

Thank you. Is there any randomness involved while training. with same training data and same parameters, do we always get same model ? If not, how to get same model with rest remains the same ?

from deepnl.

If you use the same initial embeddings and the same training data, you always get the same model, since the network is initialized with random values with a fixed seed.
The seed is set in dl-ner.py. You can comment it out if you want to get different results in different runs.

On 29/ago/2015, at 11:26, kiran-surya [email protected] wrote:

Thank you. Is there any randomness involved while training. with same training data and same parameters, do we always get same model ? If not, how to get same model with rest remains the same ?

—
Reply to this email directly or view it on GitHub #10 (comment).

from deepnl.

got it.

from deepnl.