fetchai/colearn

Colearn is a collection of protocols for running multi-stakeholder machine learning that preserve data privacy

colearn's Issues

Installation with pip fails

Installation fails with pip with the following error

(C:\Users\Karthik\colearn) C:\Users\Karthik\Desktop\Base\colearn>pip install colearn[pytorch]
Collecting colearn[pytorch]
Using cached colearn-0.2.6-py3-none-any.whl (113 kB)
Collecting pydantic~=1.7.0
Using cached pydantic-1.7.4-cp37-cp37m-win_amd64.whl (1.7 MB)
Collecting matplotlib~=3.3.0
Using cached matplotlib-3.3.4-cp37-cp37m-win_amd64.whl (8.5 MB)
Collecting numpy~=1.16.0
Using cached numpy-1.16.6-cp37-cp37m-win_amd64.whl (11.9 MB)
Collecting google-cloud-storage~=1.35.0
Using cached google_cloud_storage-1.35.1-py2.py3-none-any.whl (96 kB)
Collecting torch~=1.7.0; extra == "pytorch"
Using cached torch-1.7.1-cp37-cp37m-win_amd64.whl (184.1 MB)
Collecting opacus~=0.10.0; extra == "pytorch"
Using cached opacus-0.10.1-py3-none-any.whl (84 kB)
Collecting scikit-learn~=0.23.0; extra == "pytorch"
Using cached scikit_learn-0.23.2-cp37-cp37m-win_amd64.whl (6.8 MB)
Collecting Pillow~=8.0.1; extra == "pytorch"
Using cached Pillow-8.0.1-cp37-cp37m-win_amd64.whl (2.1 MB)
Collecting scipy~=1.5.0; extra == "pytorch"
Using cached scipy-1.5.4-cp37-cp37m-win_amd64.whl (31.2 MB)
ERROR: Could not find a version that satisfies the requirement torchvision~=0.8.0; extra == "pytorch" (from colearn[pytorch]) (from versions: 0.1.6, 0.1.7, 0.1.8, 0.1.9, 0.2.0, 0.2.1, 0.2.2, 0.2.2.post2, 0.2.2.post3, 0.3.0, 0.4.1, 0.5.0, 0.9.0,0.9.1, 0.10.0)
ERROR: No matching distribution found for torchvision~=0.8.0; extra == "pytorch" (from colearn[pytorch])

Steps to reproduce the behavior:

1. pip install colearn[pytorch]

• OS:Windows
• Version [e.g. 22]

Was able to resolve this by cloning the repo, updating the setup.py to look for torchvision 0.9.0 instead of 0.8.0 and finally installed via local pip install.

python >3.9

Just running through a few issues

1. I had some installation problems - do we have a docker version of this?
2. tox is suggested for the tests but is not included in the requirements or Pipfile.
3. For the website docs, would be hand to have the "pull this repo" link at the start of the instructions to save a bit of searching.

is this just a simulator?

I cannot find any blockchain-related source code

readme file links are giving 404 errors

Many of your links to documenation on the Readme file do not work.

Class ColearnPlot from utils.plot.py produces empty plots

Hi everyone,

I was running the examples (Keras Mnist & random forest Iris).
In the collective elarning sections in the end, the plot class is being initialized.
in plot.plot_x (x in [votes/results]) it should produce a plot, but it does not.

Code snippet:

###### # Do collective learning
results = Results()
results.data.append(initial_result(all_learner_models))  # Get initial score

plot = ColearnPlot(score_name="accuracy")

for round_index in range(n_rounds):
    results.data.append(
        collective_learning_round(all_learner_models,
                                  vote_threshold, round_index)
    )
    print_results(results)

    # then make an updating graph
    plot.plot_results(results,True)
    plot.plot_votes(results,True)

plot.plot_results(results,block=True)
plot.plot_votes(results, block=True)

print("Colearn Example Finished!")

The output is:

Doing collective learning round
--------------- LATEST ROUND RESULTS -------------
Selected proposer: 0
New model accepted: True

learner id vote test score vote score
0 True 1.000 1.000
1 True 1.000 0.846
2 True 1.000 1.000
3 True 1.000 1.000
4 True 1.000 0.923
5 True 1.000 0.846
6 True 1.000 1.000
7 True 1.000 1.000
8 True 1.000 1.000
9 True 1.000 0.846

Figure size 432x288 with 0 Axes
Doing collective learning round
--------------- LATEST ROUND RESULTS -------------
...

Yet, the class seems to have axes set:

Is this a known issue? And thank you in advance for your support!