Rnn testable doctests #2147
Rnn testable doctests #2147
Conversation
|
You are owner of some files modified in this pull request. |
| >>> input = tf.keras.Input((100, 100)) | ||
| >>> output = tf.keras.layers.Layer(input) | ||
| <... Note: Values used for timesteps and input_dim have been used purely |
There was a problem hiding this comment.
Is this warning? I think we can at least print output.
There was a problem hiding this comment.
I'll include the output of output but it's only going to be displayed as a Layer object. I think the example tries to depict the context in which the PeepholeLSTMCells can be utilized, it doesn't necessarily demonstrate the full process. That's why I included a short note stating that the values used for input have no significance.
There was a problem hiding this comment.
Oh I see. Can we change the functional API to
>>> output, state = rnn(input)
>>> output
>>> stateThis is much more informative.
There was a problem hiding this comment.
@WindQAQ Sorry for the late reply I somehow missed this comment. When I try
>>> keras.layers.RNN(peephole_lstm_cells,return_state=True)
>>> layer.statesit displays ([None, None], [None, None]). I'm a little confused on how to view the state of the layer.
|
@Harsh188 We should canonicalize the example. See the following colab for more details. Note that different cell may have different lengths of state, so when https://colab.research.google.com/drive/1Ndgy5iseSrcBm6tjQGdnLyMhbszvdAM8?usp=sharing |
|
@WindQAQ Thank you for providing the colab example! I'll get to work on this and implement them for the other rnn cells as well. |
|
Friendly ping @WindQAQ :) |
* added doctests layer_norm_simple * added doctests peephole_lstm_cell.py * minor changes * including output for example * Updated examples to be descriptive and standardized
5 participants
Description
Adding testable docstrings to
rnn.Part of #2066
Changes made to:
esn_cell.pylayer_norm_simple_rnn_cell.pypeephole_lstm_cell.pyType of change
Checklist:
How Has This Been Tested?
Using pytest