Extract all examples/use cases into seperate repositories and clean up clutter about agentmet4fof

[BjoernLudwigPTB]

@bangxiangyong since I need some input here, could we maybe meet tomorrow for a short while to keep going with this?

[bangxiangyong]

sorry for the late reply, just saw this and the previous one comment awhile ago.

Yes, i think it is quite tricky, since in all the examples i have used BNN it seems. I think we can replace BNN with some other models in sklearn for the examples and still illustrate the same points. Thats a valid suggestion to have agentMET4FOF_ml, but not at this point, as for now, i'd try to keep ML tied in as much as possible with the agents.

it'll be busy for me tomorrow, can we meet on Friday noon++ instead?

[BjoernLudwigPTB]

TODO for @bangxiangyong

• Rewrite the examples.ML_EXPERIMENTS tutorial to get rid of ml_uncertainty and pytorch together with it and put it in agentMET4FOF_tutorials

• deal with the examples.custom_dashboard.py which should become a separate tutorial and move out of examples.

• separate develop.ML_Experiment.py and all its dependencies from dashboard.py into a separate repository, which will then be incorporated as an extension into the agents' documentation on ReadTheDocs like we showed it is possbile in the draft PR #95 (never to be merged) and installable with pip install agentMET4FOF[ml] or simply

  pip install agentMET4FOF agentMET4FOF_ml_extension

[BjoernLudwigPTB]

As discussed during an extended session on April 27th, we will conduct the following measures to achieve a cleaner core repo.

Seperate repos

The repos we will create and the according subfolders of agentMET4FOF are

• ZEMA_BNN (new repo you can find here)
• ZEMA_EMC (new repo you can find here)
• ML (new repo you can find here)
• SENSORS (new repo you can find here)

Improved Documentation

We want to add documentation pages for the following examples:

• examples.demo.run_dashboard.py with simpler (maybe the present) modules for agents and streams to drop dependency on ZEMA_EMC

We removed the file already. The former content was:

# Agent modules
from agentMET4FOF.agents import AgentDashboard
from agentMET4FOF import agents
from agentMET4FOF import streams

# This example shows how we can run the dashboard separately on a different IP
# The dashboard will need to connect to an existing up and running AgentNetwork,
# which is running on another process or script. Otherwise there will be error messages.

modules = [agents, streams]


def run_dashboard():
    AgentDashboard(
        dashboard_modules=modules,
        dashboard_update_interval=3,
        agent_ip_addr="127.0.0.1",
        agent_port="3333",
    )


if __name__ == "__main__":
    run_dashboard()

Tutorials

We want to transfer these modules into tutorial notebooks and/or scripts:

• examples.custom_dashboard.py

Remove

We want to remove:

• examples.demo.demo_agent_network.py
• examples.demo.demo_agents.py
• ml_uncertainty aligned with the ML_EXPERIMENTS tutorial which should in the end not contain references to ml_uncertainty anymore

Extract into extension package

From the following contents of the subfolder develop we want to create a separate package to be installed along agentMET4FOF if desired.

• ML_EXPERIMENTS.ml_experiment_complex.py
• ML_EXPERIMENTS.ml_experiment_multi.py
• ML_EXPERIMENTS.ml_experiment_simple.py
• develop.ML_Experiment.py goes into a separate Plugin-like package

[BjoernLudwigPTB]

Seperate repos

The repos we will create and the according subfolders of agentMET4FOF are

• ZEMA_BNN (new repo you can find here)
• ZEMA_EMC (new repo you can find here)

@bangxiangyong seemingly we made a mistake here, since ml_uncertainty is heavily used in ML_EXPERIMENTS, which we want to transform into a tutorial inside agentMET4FOF. This would mean, we cannot extract ml_uncertainty and it would mean, we will need pytorch as a dependency of agentMET4FOF which we should try to avoid, because it imposes difficulties for us during the packaging and for the users during the installation.

We just created a branch and PR #76 to at least transfer ml_uncertainty into ML_EXPERIMENTS but should decide how to continue with that shortly, because of the heavyweight dependency pytorch. Maybe in this case it would be an idea to create a separate extendension repository like agentMET4FOF_ml, like other bigger projects do it (e.g. dash with dash_cytoscape, dash_canvas, ...)

What would you suggest?

Edit: This is actually resolved by the following conversation and the mentioned PR is closed already.

[BjoernLudwigPTB]

sorry for the late reply, just saw this and the previous one comment awhile ago.

Yes, i think it is quite tricky, since in all the examples i have used BNN it seems. I think we can replace BNN with some other models in sklearn for the examples and still illustrate the same points. Thats a valid suggestion to have agentMET4FOF_ml, but not at this point, as for now, i'd try to keep ML tied in as much as possible with the agents.

it'll be busy for me tomorrow, can we meet on Friday noon++ instead?

Well, if we can drop scikit-multiflow and pytorch as dependencies, there is no real reason to split up something. The suggestion was meant to allow for keeping pytorch inside of the agents, but if you see a way, to just get rid of that in the core, like replace ml_uncertainty in ML_EXPERIMENTS by scikit-learn-internals, that would already solve the issue.

[BjoernLudwigPTB]

TODO for @BjoernLudwigPTB

• Move develop.ML_Experiment.py to agentMET4FOF now

@bangxiangyong Hey Xiang, we agreed on me moving develop.ML_Experiment.py into agentMET4FOF but I did not recognize then, that this means merging develop.evaluator.py and develop.datastream.py into agents.py. Since this produces name clashs and it is actually your implementation this merging should definitely done by you. If I can assist you further, please let me know, but I cannot do this on my own, so I transfer this "ticket" to you.

• Move develop.ML_Experiment.py into agentMET4FOF and merge develop.evaluator.py and develop.datastream.py into agents.py now
• check how much effort it would be to include the documentation of the newly to create package develop.ML_Experiment.py into the agentMET4FOF docs on rtd

[bangxiangyong]

The examples.custom_dashboard.py has been moved into a set of tutorials for plotting:
f7fa5dc