Compute the carbon footprint of our pipeline for each baseline

[hosseinfani]

@Sharjeeliv
When you're done with the installation of seera, as we briefly talked about carbon footprint of models, start this task. We want to know what is the carbon footprint of each baseline (topic modeling by graph embedding)

Here is the helpful link to find the library that calculates carbon footprint of a model:
https://twitter.com/ZetaVector/status/1547916747507871744?s=20&t=MZWEJDzLVUSx0hOKpQC5-Q

[Sharjeeliv]

For this task should I submit the code or the results? In the meantime, I've installed the package and I'm working on implementing it.

Also, the project is still not available on SharePoint for me, I've contacted Dr. Brunet and he has granted me an extension on the contract for the time being.

[hosseinfani]

@Sharjeeliv
You can send pull request (pr) when you're sure that your code is ready to be merged to the codebase.

I already submitted the project and we're waiting for Dr. Kobti to confirm the project. You can follow up from Dr. Kobti as well.

[Sharjeeliv]

Update: Reviewing git and GitHub

[Sharjeeliv]

Update

• Implemented carbon footprint for each of the topic modelling LDAs (was it this or each embedding method?)
• Also added in a coloured log for the result (to make it more visible)
• Followed up with Dr. Kobti and Dr. Brunet to get the contract done
• Reviewed how to do pull requests and use git

To-Do

• Setting up workflow using (1) and submit pr

Notes

• Attached is the log file (sadly it's only coloured in the terminal) Log.txt

(1) https://learntocodetogether.com/create-your-first-pull-request/

[hosseinfani]

@Sharjeeliv
Thank you. I had a look at the log file. It shows you could successfully run the pipeline. Awesome.
However, I was not able to see the carbon footprint result. Looking forward to your pr. thanks.

[Sharjeeliv]

Hi Sir,
I've been sick this week so I haven't been able to wrap up and send pr, I will do this as soon as I feel I can (hopefully in 1-2 days).

[hosseinfani]

@Sharjeeliv
Sorry to hear that. Wish you fast and full recovery.
Take your time and rest.

[Sharjeeliv]

Update

• In the above log file if you search for emissions it will show you the carbon footprint.
• CodeCarbon also generates a file with more details after execution.
• Pulled new changes and am in the process of figuring out how to handle merge conflicts.
• Once I've successfully merged it, I will open a pr.

[Sharjeeliv]

Update

• Everything has been merged

• Bitermplus is not installing

• Different parts of the pipeline are not working anymore (log.txt)

• Trying to resolve the issues to bring it back to a working state

• After deleting output and rerunning: Log.txt

• Ok looking at other issues I just realized not every combination was working originally

[hosseinfani]

@Sharjeeliv
You only need to test your code on a running combination. Leave the bugs in other combinations to @soroush-ziaeinejad

[Sharjeeliv]

I've opened a pr for this feature. The pipeline is not working for me, but since the feature has been done for some time and successfully tested on earlier versions, it seemed best to open the pr for review.

[hosseinfani]

Hi @Sharjeeliv
Thank you.
There are some issues with your pr:

1. Please only include the files or changes that are necessary. I reviewed your pr and seems only requirement.txt and TopicModeling.py is necessary, am I right?

2. the carbon tracer only targets the topic modeling layer. However, we need to trace the whole pipeline. So, you should put it in the main.py here:

SEERa/src/main.py

Line 162 in 4e032be

def run(tml_baselines, gel_baselines, run_desc):

Let me know if you need more info on this.

[Sharjeeliv]

Update

• I updated the codeCarbon based on the meeting, it is attached below
• I have contacted Soroush about the errors, so once this is resolved everything should be done

def run(tml_baselines, gel_baselines, run_desc):
    for t in tml_baselines:
        for g in gel_baselines:
            tracker = EmissionsTracker()  # We want to reset the tracker on each iteration to get the emission of each combination
            tracker.start()
            try:
                cmn.logger.info(f'Running pipeline for {t} and {g} ....')
                baseline = f'{run_desc}/{t}.{g}'
                with open('ParamsTemplate.py') as f:
                    params_str = f.read()
                new_params_str = params_str.replace('@baseline', baseline).replace('@tml_method', t).replace(
                    '@gel_method', g)
                with open('Params.py', 'w') as f:
                    f.write(new_params_str)
                importlib.reload(Params)
                main()
            except:
                cmn.logger.info(traceback.format_exc())
            finally:
                emissions: float = tracker.stop()
                cmn.logger.info(f'Pipeline Emissions for {t} - {g}: {emissions}')
                cmn.logger.info('\n\n\n')

[hosseinfani]

@Sharjeeliv
any update? can we have the log of the carbon for all combinations on toy sets?

[Sharjeeliv]

Yep, this was tested to work for all Gensim combinations, it does not work with Mallet (it has issues running bat file on macOS, I have not found a solution yet). Soroush recommended that I make it work for Gensim, and he'll test it out for Mallet.

[hosseinfani]

@Sharjeeliv thanks.
@soroush-ziaeinejad would you please verify this task and close this issue. Then, assign a new task to Sharjeel. Thanks.

[soroush-ziaeinejad]

@Sharjeeliv Thank you for pr.
@hosseinfani Sure, I'll do it tomorrow.

[soroush-ziaeinejad]

@hosseinfani,
Done.

Thanks, @Sharjeeliv