Optional dependencies not working properly

[ColdTeapot273K]

• Issue: For some reason both pip install tensor-sensor pip install tensor-sensor[torch] attempt to install Tensorflow too.

• Environment:

  • win10 latest (10.10.2020)
  • conda 4.8.3 virtual env
  • pytorch 1.6.0 installed via conda (the official way)
  • no tensorflow

• Workaround: pip install tensor-sensor --no-deps pip install graphviz

[parrt]

@noklam any ideas?

[noklam]

@parrt I can install with the dependency I needed frome the source dircetory. @parrt I guess pip install tensor-sensor[torch] will only works after you push it to the pypi package?

Here are the steps that I can reproduce my environment without unnecessary dependencies.

conda create -n ts_new python==3.7.5
git clone https://github.com/parrt/tensor-sensor.git
cd tensor-sensor
pip install -e . # Editable install from source, it won't install either torch or tensorflow

[noklam]

I guess you have to run this command? Saw it from your README python setup.py sdist upload

I don't have any experience publishing a package before

[drscotthawley]

I'd like to jump in on this! I just tried pip install tensor-sensor and found that it was trying to install PyTorch:

$ pip install tensor-sensor
Collecting tensor-sensor
  Downloading tensor-sensor-0.1b1.tar.gz (20 kB)
Collecting graphviz>=0.14.1
  Downloading graphviz-0.14.2-py2.py3-none-any.whl (18 kB)
Requirement already satisfied: numpy in /home/shawley/anaconda3/envs/spnet/lib/python3.7/site-packages (from tensor-sensor) (1.19.0)
Collecting torch
  Downloading torch-1.6.0-cp37-cp37m-manylinux1_x86_64.whl (748.8 MB)
     |▉                               | 19.1 MB 3.4 MB/s eta 0:03:36^C
^C
ERROR: Operation cancelled by user

I've already got PyTorch installed in a different environment, so I only wanted just tensor-sensor. From my reading of the documentation I thought that was all I should be getting.

[noklam]

@drscotthawley can you try this?

git clone https://github.com/parrt/tensor-sensor.git cd tensor-sensor pip install -e .

[drscotthawley]

@noklam That installed fine with no trying-to-install torch! Thanks.

Although I note that upon execution there's still a 'torch' dependency in the code :-( :

$ cat test.py

import numpy as np

n = 200                          # number of instances
d = 764                          # number of instance features
n_neurons = 100                  # how many neurons in this layer?

W = np.random.rand(d,n_neurons)  # Ooops! Should be (n_neurons,d) <=======
b = np.random.rand(n_neurons,1)
X = np.random.rand(n,d)          # fake input matrix with n rows of d-dimensions

import tsensor
with tsensor.clarify():
    Y = W @ X.T + b
#Y = W @ X.T + b                  # pass all X instances through layer

$ python test.py
Traceback (most recent call last):
  File "test.py", line 11, in <module>
    import tsensor
  File "/home/shawley/Downloads/tensor-sensor/tsensor/__init__.py", line 29, in <module>
    import tsensor.viz
  File "/home/shawley/Downloads/tensor-sensor/tsensor/viz.py", line 39, in <module>
    import tsensor.analysis
  File "/home/shawley/Downloads/tensor-sensor/tsensor/analysis.py", line 27, in <module>
    import torch
ModuleNotFoundError: No module named 'torch'

[noklam]

@drscotthawley If you don't have torch installed, you should install with pip install -e .[torch]

pip install tensor-sensor[torch] won't work until @parrt update the pypi package.

currently the pypi package is behind the master branch

[drscotthawley]

@noklam Sorry, huh? My comment is about how I don't want to install torch or the torch dependency in this environment. I'm not installing tensor-sensor[torch], only tensor-sensor. Nor am I wanting to install the [all] option.

Running pip install -e .[torch] just starts installing torch again:

~/Downloads/tensor-sensor$ pip install -e .[torch]
Obtaining file:///home/shawley/Downloads/tensor-sensor
Requirement already satisfied: graphviz>=0.14.1 in /home/shawley/anaconda3/envs/spnet/lib/python3.7/site-packages (from tensor-sensor==0.1b2) (0.14.2)
Requirement already satisfied: numpy in /home/shawley/anaconda3/envs/spnet/lib/python3.7/site-packages (from tensor-sensor==0.1b2) (1.19.0)
Requirement already satisfied: IPython in /home/shawley/anaconda3/envs/spnet/lib/python3.7/site-packages (from tensor-sensor==0.1b2) (7.18.1)
Requirement already satisfied: matplotlib in /home/shawley/anaconda3/envs/spnet/lib/python3.7/site-packages (from tensor-sensor==0.1b2) (3.2.2)
Collecting torch
  Downloading torch-1.6.0-cp37-cp37m-manylinux1_x86_64.whl (748.8 MB)
     |▍                               | 8.3 MB 1.4 MB/s eta 0:08:38^C
^C
ERROR: Operation cancelled by user

The environment I'm in is actually set up for Keras. I could do the tensorflow version I suppose if that's required. But the docs didn't seem to indicate the that torch or tensorflow were required.

[noklam]

@drscotthawley Ah, sorry I misunderstood your question. I have opened a separate issue #9 about this, the original issue is only about the installation. this seems to related to the source code always tried to import torch first, I think we could get rid of that dependency with a separate PR later.

[parrt]

Ooops. Sorry. I'm an idiot. forgot to push new pip stuff. Done just now. 0.1b3