the use of MultiRNNCell?

[liuyichaosoftware]

doc is in https://www.tensorflow.org/tutorials/recurrent

I found the line
stacked_lstm = tf.contrib.rnn.MultiRNNCell([lstm] * number_of_layers,

I need to use MultiRNNCell

but,I write those lines

a = [tf.nn.rnn_cell.BasicLSTMCell(10)]*3

print id(a[0]), id(a[1])

Its output is 4648063696 4648063696

can MultiRNNCell use the same object BasicLSTMCell as a list for parameter?

[carlthome]

can MultiRNNCell use the same object BasicLSTMCell as a list for parameter?

Yes, you'll get separate trainable weights by repeating the list like you're doing. It's a bit confusing to be honest. You should have asked this here or on StackOverflow though.

[aselle]

@ebrevdo, do you think we should change the docs in some way to make it clearer?

[ebrevdo]

This is indeed confusing and hopefully next week we'll have a push (over a month in the making) that disallows this behavior. It's a bit out of scope of this bug to describe why this behavior is the way it is, but it's been the TF standard for a long time. But after this change, you'll have to write: MultiRNNCell([BasicLSTMCell(10) for _ in range(3)]) otherwise you will get an error when calling dynamic_rnn. With this version of the code, you end up creating 3 separate BasicLSTMCell instances, each of which expects to maintain its own parameters.

…

On Fri, Feb 17, 2017 at 3:04 PM, Andrew Selle ***@***.***> wrote: @ebrevdo <https://github.com/ebrevdo>, do you think we should change the docs in some way to make it clearer? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#7604 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABtim34fMLpVWgZprthmv-Wuv274ysr3ks5rdidxgaJpZM4MD1hN> .

[liuyichaosoftware]

thank you very much ~

[carlthome]

Don't forget to close your issue @liuyichaosoftware. 👍

[tifoit]

Tensorflow 0.9 to 1.0.1 is a big jump. A lot changed.
Details: https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md

In tf 1.0.0, the API has been changed such as:

tf.nn.seq2seq.sequence_loss_by_example
to
tf.contrib.legacy_seq2seq.sequence_loss_by_example

tf.nn.rnn_cell.
to
tf.contrib.rnn.

tf.nn.rnn_cell.MultiRNNCell(
to
tf.contrib.rnn.MultiRNNCell(

...

[lonway]

How to find the API map between the old and new version @tifoit