How to use InputLayer/ How to interact TensorFlow and TensorLayer

[llan-ml]

The input of each layer (e.g., Dense layer and Conv layer) except "Input layer" must be a Layer instance.
I think this behavior is not so transparent to TensorFlow.

In the case that I want to use an intermediate tf.Tensor as input for a Layer instance,
is there any existing method of TensorLayer to implement this function?

[zsdonghao]

Hi, you can feed your tf.Tensor into tl.layers.InputLayer and then feed the layer into Dense and Conv layer.

If you look into the source code, the InputLayer just creat a Layer.outputs , more details can be found on the documentation : http://tensorlayer.readthedocs.io/en/latest/modules/layers.html#understand-layer

[llan-ml]

@zsdonghao thx, the solution is the same as mine.
But this solution causes that newly created layers cannot track components,which are located before tl.layers.InputLayer, using methods such as layer.all_params and layer.all_drop.

[zsdonghao]

@llan-ml Yes, indeed. Layer cannot track the parameters outside TensorLayer, I suggest you to usetrain_params = layer.all_params + [you parameters] if you want to update the parameters outside TensorLayer .

For layers.all_drop, if you control the keep probabilities by placeholder, then you will need to feed the values into feed_dict, but in this case, to keep everything simple, I suggest you to create different graph for training and evaluating :
e.g. https://github.com/zsdonghao/tensorlayer/blob/master/tutorial_ptb_lstm_state_is_tuple.py