6.3-advanced-usage-of-recurrent-neural-networks: Missing arguments in Keras documentation?

[FriendOfFatBeagle]

Hi Everybody:

I am using Keras 2.0.5. Here are code samples for section 6.3 of book, "Advanced use of recurrent neural network" pertaining to GRU:

`from keras.models import Sequential from keras import layers from keras.optimizers import RMSprop

model = Sequential() model.add(layers.GRU(32, input_shape=(None, float_data.shape[-1]))) model.add(layers.Dense(1))

model.compile(optimizer=RMSprop(), loss='mae') history = model.fit_generator(train_gen, steps_per_epoch=500, epochs=20, validation_data=val_gen, validation_steps=val_steps)`

My question refers to this code: model.add(layers.GRU(32, input_shape=(None, float_data.shape[-1])))

When I looked up documentation for layers.GRU(), at https://keras.io/layers/recurrent/#gru, there is no documentation for argument input_shape. There is no mention of input-shape argument. The same problem also occurs with layers.LSTM(). This makes learning very hard.

I also tried looking at the source code, for class Recurrent(Layer) and class GRU(Recurrent) in recurrent.py for Keras 2.0.5, input_shape does not seems to exists as an argument for these classes!

What did I missed?

Why is argument input_shape not listed as one of the many arguments of LSTM() and GRU() at https://keras.io/layers/recurrent/#lstm and at https://keras.io/layers/recurrent/#gru ?

How do I go about locating specific information on all arguments that are used by these classes?

Thank you.

[birolkuyumcu]

https://keras.io/getting-started/sequential-model-guide/

The model needs to know what input shape it should expect. For this reason, the first layer in a Sequential model (and only the first, because following layers can do automatic shape inference) needs to receive information about its input shape. There are several possible ways to do this:

**Pass an input_shape argument to the first layer.**

....

[FriendOfFatBeagle]

Hi birolkuyumcu:

You did not answer my question. Seems like you do not understand my question.

## Repeat:

My question refers to this code: model.add(layers.GRU(32, input_shape=(None, float_data.shape[-1])))

When I looked up documentation for layers.GRU(), at https://keras.io/layers/recurrent/#gru, there is no documentation for argument input_shape. There is no mention of input-shape argument. The same problem also occurs with layers.LSTM(). This makes learning very hard.

.... ....
See the rest in original post.

Why is argument input_shape not listed as one of the many arguments of LSTM() and GRU() at https://keras.io/layers/recurrent/#lstm and at https://keras.io/layers/recurrent/#gru ?

How do I go about locating specific information on all arguments that are used by these classes if I cannot find them in recurrent.py?

[birolkuyumcu]

https://github.com/keras-team/keras/blob/master/keras/engine/topology.py#L279

 note that 'dtype', 'input_shape' and 'batch_input_shape'  are only applicable to input layers

[FriendOfFatBeagle]

Hi Birol:

Thank you. I have NO idea input_shape argument is declared in file topoloty.py located in folder: `> ~/anaconda3/envs/tensorflow/lib/python3.5/site-packages/keras/engine $ ls

init.py pycache topology.py training.py `

What steps did you take to locate it in topology.py? The only place I knew how to look was in layers folder, like so: > ~/anaconda3/envs/tensorflow/lib/python3.5/site-packages/keras/layers $ ls advanced_activations.py core.py local.py normalization.py recurrent.py convolutional.py embeddings.py merge.py pooling.py wrappers.py convolutional_recurrent.py __init__.py noise.py __pycache__ Addendum: I see there is a reference to importing Layer from engine folder in file init.py in folder, ~/anaconda3/envs/tensorflow/lib/python3.5/site-packages/keras/layers

Is the above folder and init,py is how you locate reference for input-shape argument or is there some other more efficient way?

By the way, input-shape is listed as one of many arguments in https://www.tensorflow.org/api_docs/python/tf/keras/layers/GRU and https://www.tensorflow.org/api_docs/python/tf/keras/layers/LSTM

Inconsistency and missing information in keras documentation is just rampant or I may just do not have adequate knowledge of how to locate appropriate reference. Any instructions on how to efficiently do so would greatly accelerate learning.

Thank you, again.

[birolkuyumcu]

Layer is a abstract base class of all layers so All layers inherited it how to find ? https://github.com/keras-team/keras/blob/master/keras/layers/recurrent.py#L16 then https://github.com/keras-team/keras/blob/master/keras/engine/__init__.py#L6

i think sequential model guide explain input_shape argument maybe explanation about this at about keras layers makes more clear https://keras.io/layers/about-keras-layers/

[FriendOfFatBeagle]

Hi Birol:

Thank you. Below has good explanation (mostly via code samples) on how to specify input_shape for RNN/GRU/LStM:

https://keras.io/getting-started/sequential-model-guide/

[birolkuyumcu]

https://keras.io/layers/recurrent/ RNN ; Base class for recurrent layers.

Input shape 3D tensor with shape (batch_size, timesteps, input_dim).

[FriendOfFatBeagle]

Hi Birol:

Google translate: Çok yardımcı oldun. Cömert zamanınız için teşekkür ederiz. of "You have been very helpful. Thank you for your generous time." Sorry, I do not speak your native tongue.