search

tensorflow / skflow

Simplified interface for TensorFlow (mimicking Scikit Learn) for Deep Learning
Apache License 2.0
3.18k stars 439 forks source link

How can one control the number of epochs? #138

Closed evahlis closed 8 years ago

evahlis commented 8 years ago

In this example: https://github.com/tensorflow/skflow/blob/35e12d74537ffbf7a27a2136337b1c52c922ae96/examples/boston.py

The number of training steps is set to 5000. The documentation of DNN Regressor states that "steps" specifies the number of steps to run over data. Does that mean that the above example uses 5000 training epochs?

More generally, what is the intended way to control number of epochs?

Thanks.

evahlis commented 8 years ago

Reading through the code, the steps argument seems to set the number of mini-batches to train on. So if there are n samples and the batch size is b then number of epochs will be s*b/n