Closed HiuNaoki closed 8 years ago
Currently it's not very simple at all.
The easiest way, would be to dump numpy arrays from the pre-trained model (using get_tensor_value
method on estimator) and then use initializer
for your variables in your target model.
I'll make it easier, but sklearn doesn't have anything remote to API for this, so will need to think carefully to have something that would be useable in years to come.
Here is suggested API: https://github.com/tensorflow/tensorflow/issues/2141 and will track this functionality there.
i want to using autoencoder for pre-training in skflow....
please help me..