Comments on: A noob’s guide to implementing RNN-LSTM using Tensorflow

By: tensorflow - Mínimo RNN ejemplo en tensorflow

tensorflow - Mínimo RNN ejemplo en tensorflow — Fri, 12 Apr 2019 18:19:52 +0000

[…] github.com/aymericdamien/TensorFlow-Examples/blob/master/…. Un LSTM ejemplo con el tutorial: monik.en/…. Recuento de bits: gist.github.com/monikkinom/e97d518fe02a79177b081c028a83ec1c. Uno más: […]

By: N Satya Krishna

N Satya Krishna — Fri, 25 Jan 2019 09:44:07 +0000

In reply to Avijit Thawani. Can you help me? How can we get values of hidden state of LSTM after execution of each batch of input while training?

By: tensorflow - Minimes RNN exemple dans tensorflow

tensorflow - Minimes RNN exemple dans tensorflow — Mon, 17 Dec 2018 23:38:10 +0000

[…] Un LSTM exemple avec la procédure pas à pas: monik.dans/…. Le comte bits: gist.github.com/monikkinom/e97d518fe02a79177b081c028a83ec1c. Un de plus: […]

By: Avijit Thawani

Avijit Thawani — Tue, 25 Sep 2018 05:20:47 +0000

Thanks for the tutorial! I had a conceptual query: Is it an obvious choice to pose this as a classification problem? What if we tried to learn a regression instead, wherein the output to be learnt is the number of 1s in the string? Apart from the fact that the output space will be unbounded in a regression setting, what other problem could we have?
As a regression problem, we could have several times more examples with the same amount of training data, by taking subsequences of the string too? That way, the network must not learn to keep computing and only be tested at its 20th state but instead be forced to devise a way to keep counting at every step. Suggestions?

By: Heather

Heather — Tue, 15 May 2018 01:51:58 +0000

Thank you so much for this very helpful guide! I finally have an LSTM set up and working. 🙂

The only thing that didn’t work for me was the last line, printing a prediction for a particular sample. When I try to include that line I get an error – NameError: name ‘model’ is not defined.

Could you tell me how to solve that error?

By: sumeet

sumeet — Sun, 13 May 2018 23:57:51 +0000

sir , can you suggest a way to get those 20 outputs for each sequence ?

Thanks.

By: sumeet

sumeet — Thu, 03 May 2018 00:32:26 +0000

i am trying to do word recognition with this.
so i will input an array of 26 MFCC at each timestep.
so input dimension will be 26.
Will that work ?

Nice tutorial BTW.
Thanks.

By: Learning to implement LSTMs – Handy Learning Paths

Learning to implement LSTMs – Handy Learning Paths — Tue, 17 Apr 2018 22:04:15 +0000

[…] Easy and fast tutorial to understand a really basic implementation. do it yourself and see it work. […]

By: Bhaskar

Bhaskar — Sat, 07 Apr 2018 10:46:15 +0000

Hi
I would like to suggest using sequence_length argument more liberally so that instead of wasting time doing complex computations:

val = tf.transpose(val, [1, 0, 2])
last = tf.gather(val, int(val.get_shape()[0]) – 1)

to get the last valid output state in each sequence we can simply use the last output state state.h as it will automatically stop computation at the sequence length’s end; if it receives the input argument of of sequence_length otherwise it will compute all the output states till the max sequence length. Please update the code with the same.

By: Vivwek Kumar

Vivwek Kumar — Sat, 07 Apr 2018 01:54:28 +0000

Cannot feed value of shape (1000, 20) for Tensor ‘data_2:0’, which has shape ‘(?, 20, 1)’