strided_slice() missing 1 required positional argument: 'strides'

[DanilBaibak]

I'm working with models/tutorials/rnn/ptb/.

After run command python ptb_word_lm.py --data=simple-examples/data/ --model small:

Traceback (most recent call last):
  File "ptb_word_lm.py", line 371, in <module>
    tf.app.run()
  File "/home/danil/miniconda/envs/tf/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 30, in run
    sys.exit(main(sys.argv[:1] + flags_passthrough))
  File "ptb_word_lm.py", line 331, in main
    train_input = PTBInput(config=config, data=train_data, name="TrainInput")
  File "ptb_word_lm.py", line 94, in __init__
    self.input_data, self.targets = reader.ptb_producer(data, batch_size, num_steps, name=name)
  File "/home/danil/projetcs/ml/boots/ptb_word_lm/reader.py", line 116, in ptb_producer
    x = tf.strided_slice(data, [0, i * num_steps], [batch_size, (i + 1) * num_steps])
TypeError: strided_slice() missing 1 required positional argument: 'strides'

Thanks in advance!

[aselle]

What version of TensorFlow are you using? Try the latest release. tf.strided_slice() has had a default added for strides (all_ones of the appropriate size) so it should be optional.

[DanilBaibak]

Currently, I'm using tensorflow 0.12.0rc0.

The parameter strides is optional in the master branch. Maybe you know, when this changes will be released? :)

Thanks!

[aselle]

This will likely not make it into a release until 1.0 since 0.12 is cut.You can work around this on the model by adding a strides to the model's call to strided_slice.

[DanilBaibak]

@aselle, yep, problem can be fixed with adding strides to the model. Thanks for help :)

[LeenaShekhar]

Hello,

I cannot understand what is the fix for this problem. Can somebody please help. Below is the error:

python ptb_word_lm.py --data_path=/home/user/Downloads/simple-examples/data/ --model small
Traceback (most recent call last):
File "ptb_word_lm.py", line 374, in
tf.app.run()
File "/home/user/anaconda3/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 43, in run
sys.exit(main(sys.argv[:1] + flags_passthrough))
File "ptb_word_lm.py", line 334, in main
train_input = PTBInput(config=config, data=train_data, name="TrainInput")
File "ptb_word_lm.py", line 94, in init
data, batch_size, num_steps, name=name)
File "/home/user/anaconda3/lib/python3.5/site-packages/tensorflow/models/tutorials/rnn/ptb/reader.py", line 117, in ptb_producer
[batch_size, (i + 1) * num_steps])
TypeError: strided_slice() missing 1 required positional argument: 'strides'

[DanilBaibak]

Hi @LeenaShekhar,

for fix this issue you can try adjust reader.py like:
x = tf.strided_slice(data, [0, i * num_steps], [batch_size, (i + 1) * num_steps], tf.ones_like([0, i * num_steps]))

[LeenaShekhar]

Thank you so much as that solved the problem, but I am still getting errors. Is it because of TF version? I installed it using the Anaconda installation steps mentioned on the website (see the TF folder path). I want to run the RNN and Seq2Seq model. Is there a preferred installation process? I really appreciate your fast response.

python ptb_word_lm.py --data_path=/home/usr/Downloads/simple-examples/data/ --model small
Traceback (most recent call last):
File "ptb_word_lm.py", line 374, in
tf.app.run()
File "/home/user/anaconda3/lib/python3.5/site-packages/tensorflow/python/platform/app.py", line 43, in run
sys.exit(main(sys.argv[:1] + flags_passthrough))
File "ptb_word_lm.py", line 336, in main
m = PTBModel(is_training=True, config=config, input_=train_input)
File "ptb_word_lm.py", line 119, in init
cell = tf.contrib.rnn.MultiRNNCell(
AttributeError: module 'tensorflow.contrib.rnn' has no attribute 'MultiRNNCell'

[LeenaShekhar]

The code seems to work now. I am able to train the model. I made the following changes:

1. Added the changes pointed by @DanilBaibak in reader.py (2 places).
2. Changed tf.contrib.rnn.MultiRNNCell to tf.nn.rnn_cell.MultiRNNCell.
3. Changed tf.contrib.legacy_seq2seq.sequence_loss_by_example to tf.nn.seq2seq.sequence_loss_by_example.

Though I would still like to know the best way to install TF so that I have the latest version. Thank you.

[hfirouzi]

My code started working after applying changes suggested by @DanilBaibak and @LeenaShekhar and also these two changes in ptb_word_lm.py:

1. tf.contrib.rnn.BasicLSTMCell --> tf.nn.rnn_cell.BasicLSTMCell
2. tf.contrib.rnn.DropoutWrapper --> tf.nn.rnn_cell.DropoutWrapper

I am running the code in windows (python 3.5 and pip "https://www.tensorflow.org/install/install_windows")