tensorboard colab tensorflow._api.v1.io.gfile' has no attribute 'get_filesystem

user_4360433 0
Viewed 8506

I am trying to use tensorboard on colab. I manage to make it work, but not for all commands. add_graph and add_scalar works, but when I tried to run add_embedding I am getting the following error:

AttributeError: module 'tensorflow._api.v1.io.gfile' has no attribute 'get_filesystem'

This is the relevant code (I think);

import os
from torch.utils.tensorboard import SummaryWriter
writer = SummaryWriter(log_dir ="logs" )

images, labels = select_n_random(trainset.data, trainset.targets)
images = torch.from_numpy(images)
labels = torch.from_numpy(np.array(labels))
class_labels = [classes[lab] for lab in labels]

# log embeddings
features = images.reshape((-1,32*32*3))
writer.add_embedding(features,metadata=class_labels) #, label_img=images.unsqueeze(1))

The complete error is:

/tensorflow-1.15.0/python3.6/tensorflow_core/python/util/module_wrapper.py in __getattr__(self, name)
    191   def __getattr__(self, name):
    192     try:
--> 193       attr = getattr(self._tfmw_wrapped_module, name)
    194     except AttributeError:
    195       if not self._tfmw_public_apis:

AttributeError: module 'tensorflow._api.v1.io.gfile' has no attribute 'get_filesystem'

using

  • tensorflow-1.15.0 (tried to install 2.0 but had different issues)
  • Python 3.6.9
  • torch 1.4.0
  • tensorboard 2.1.1 (tried also with 1.15.0 but same issue)

I also tried using the "magic" command:

%load_ext tensorboard
%tensorboard --logdir logs

But I wasn't able to make it work that way (other issues).

Any suggestions how can I make it work?

pytorch tensorboard
3 Answers

For me, this fixed the problem:

import tensorflow as tf
import tensorboard as tb
tf.io.gfile = tb.compat.tensorflow_stub.io.gfile
user_13347610

Uninstall the tensorflow. Do not install tensorflow with torch in the same environment. If you install tensorflow, tensorboard may try to use the api of tensorflow firstly.

Then you may encounter this problem: 'LocalFileSystem' object has no attribute 'makedirs'.
There is a solution - https://github.com/pytorch/pytorch/issues/34028

  • Uninstalled tensorflow
  • Reinstalled Tensorboard
    PS: Restart Kernel and tensorboard.

tensorboard 2.2.0 and torch 1.14.0 is worked for me.

user_86453000

Solution 1:

It does slow down program execution by a huge factor.

import tensorflow as tf
import tensorboard as tb
tf.io.gfile = tb.compat.tensorflow_stub.io.gfile

Solution 2:

If you created a virtualenv to use pytorch, you have to make sure that you only installed pytorch and tensorboard, but not tensorflow, as it has conflicts with the others. Removing tensorflow from the vritualenv containing pytorch and tensorboard solves the problem.

If you did not create a virtualenv and installed everything in the same spot, either

  1. you do not need tensorflow and should remove it
  2. you need tensorflow as well as pytorch and should actually use virtual environments to manage your python dependencies (good practice anyway)

Uninstalling tensorflow will remove the %tensorboard magic, so you need to keep it if you're using tensorboard inline in a jupyter cell.

So, solution 1 made more sense in my case.

user_169419250
Related