Ray Import Failed

Describe the bug I tried to re-use a model that I have used multiple times however, today I get the error that ray cant be imported. I tried re-running my experiment and I get the same error. Exact error codes are, Experiment:

!ludwig experiment \
 --dataset CompanyAndIndust.csv\
 --config_file model_definition.yaml

NumExpr defaulting to 2 threads.
import ray failed with exception: No module named 'ray'
usage: ludwig experiment [options]
ludwig experiment: error: one of the arguments -c/--config -cs/--config_str is required

Predict:

from ludwig.api import LudwigModel
model = LudwigModel.load("/path/")
predictions, _ = model.predict(dft.rename(columns={'company_name': 'name'} ))

Import ray failed with exception: No module named 'ray'
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/conv.py:299: UserWarning: Using padding='same' with even kernel lengths and odd dilation may require a zero-padded copy of the input be created (Triggered internally at  ../aten/src/ATen/native/Convolution.cpp:744.)
  self.padding, self.dilation, self.groups)

To Reproduce Steps to reproduce the behavior:

1 installed the tools + libraries !pip install tensorflow-gpu !pip install ray !pip install ludwig

import pandas as pd from datetime import datetime as dt import numpy as np from google.colab import drive from google.colab import files drive.mount(‘drive’)

2 I run predict and I get the error

Environment (please complete the following information):

• OS: Mac running Python on Google Colab
• Version 12.2,
• Python version 3.7.13
• Ludwig version 0.5

Thank in advance guys

[tgaddair]

Hey @MarijnQ, is it possible to upgrade to Python 3.7 or newer? We are officially dropping support for Python 3.6 in Ludwig 0.5, so it may be that some dependency packages are not supported in 3.6.

[tgaddair]

Hey @MarijnQ, it looks like your notebook is GPU-capable. Another thing you can check is GPU visibility with nvidia-smi.

Apologies, I sent you the wrong torch command earlier. Can you actually run torch.cuda.is_available() and share the output?