Errors when running YOLO notebook while connected to a local runtime

[DaniBodor]

Describe the bug Whenever I try to run the YOLO notebook from a local runtime I an error. As I am not sure what the correct order of events is when running a local runtime (any steps I can skip? Do I connect to the local runtime from the beginning or after having run one more cells?), I have tried a few different things, but they all result in an error. I will describe the things I have tried below.

EDIT: my bad, I have found the correct requirements file, so you can ignore the comment below. However, the problem remains

_Another potential problem is that I am not sure I have loaded the correct requirements. No 'requirements.txt' file is created when I run the dependencies cell from the notebook on a hosted runtime. Instead, I have found this requirements file, but am not sure if this is complete. My best guess is that if I have the correct requirements file and know the order of events, then the errors will be resolved._

To Reproduce

• If I connect to local runtime first and run cell 1.1 the output is a list of 8 'Requirement already satisfied' notes.
• If I then run 1.3 I get this error: SyntaxError: from __future__ imports must occur at the beginning of the file
• If I move all the from __future__ import lines to the top of the code I get a different error: UsageError: Line magic function%tensorflow_versionnot found.
• If I comment out the %tensorflow_version line I get:

  
  ModuleNotFoundError                       Traceback (most recent call last)
  <ipython-input-2-3f2ece8c3582> in <module>
   66 from matplotlib.pyplot import imread
   67 import matplotlib.patches as patches
  ---> 68 import tensorflow as tf
   69 import pandas as pd
   70 import os

ModuleNotFoundError: No module named 'tensorflow'

- __EDITED:__ I haven't tried further steps to fix this, but this does remind me of the following error I always get when running cell 1.1 from a hosted runtime, which might be related

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. tensorflow 2.6.0 requires h5py~=3.1.0, but you have h5py 2.10.0 which is incompatible.

- If I try to skip straight to step 3.1 without running cells and 1 and 2 first, I get this error:

NameError Traceback (most recent call last)

in 16 17 ---> 18 full_model_path = os.path.join(model_path,model_name) 19 if os.path.exists(full_model_path): 20 print(bcolors.WARNING+'Model folder already exists and will be overwritten.'+bcolors.NORMAL) NameError: name 'os' is not defined ```

[guijacquemet]

Hi,

Thanks for reaching out! From the error, it looks like TensorFlow is not installed properly in your environment. If you want to use GPU, make sure to also install the correct version of CUDA.

As the first cell is failing, the OS library is not loaded. This is what is giving you the subsequent error.

I hope this helps cheers Guillaume

[DaniBodor]

Thanks for your reply. Can you maybe explain how I can make sure that TensorFlow and CUDA are properly installed? As far as I can tell I am following all the steps listed in the Running notebooks locally tab on wiki and don't know how to proceed.

I have tried:

• adding tensorflow==1.15.2 to the requirements file (which I found in the requirements_simple file). This was incompatible with gast 0.4.0, so I changed it to gast 0.2.2.
• adding tensorflow==2.6.0 to the requirements file. This was incompatible with h5py 2.10.0 and tensorboard 1.15.0, so I changed these to 3.1.0 and 2.6 (or 2.6.0), respectively. In both cases the local runtime strands in cell 1.3 with the error I wrote above: UsageError: Line magic function '%tensorflow_version' not found.
• For CUDA I have not tried anything, as I have no idea what version I need. Do I understand you correctly that it is required to connect to my local GPU?

Also, regarding my original post, maybe I was unclear about the following:

• The UsageError occurs when running from a local runtime, but not from a hosted runtime.
• Similarly, moving the from __future__ import lines to the top of the cell is only necessary in the local runtime, not the hosted runtime.
• When connected to a hosted runtime, I get this error, but it doesn't have any effect and I can keep running all subsequent cells with no problem.

  ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
  tensorflow 2.6.0 requires h5py~=3.1.0, but you have h5py 2.10.0 which is incompatible.

[DaniBodor]

I have since also tried running RetinaNet and U-Net (3D) in a local environment using the auto-generated requirements.txt file, but have run into similar problems.

[guijacquemet]

Hi, Yes, the notebooks are made to be run in Google Colab and do not work straight out of the box locally.

One thing that needs to be removed (or commented out) is the Google Colab specific functions including:

• %tensorflow_version 1.x

In the YOLO notebook, we use : tensorflow==1.15.2 h5py==2.10

Regarding installing CUDA, check this guide for instance. Just be careful to match the CUDA version to the version of TF you want to use.

Also bear in mind that the requirement files are automatically generated at the moment and may not contain all the libraries required. So you may also need to install a few more manually. We are slowly working on a guide on how to set up the notebook so that they can work locally but this is not yet ready to share.