search

lfwa / carbontracker

Track and predict the energy consumption and carbon footprint of training deep learning models.
MIT License
374 stars 27 forks source link

Log price corresponding to the energy consumption #38

Open mpariente opened 4 years ago

mpariente commented 4 years ago

Would it be possible to log the price per epoch and per training, along the equivalent number of kms in car? It would be quite informative.

Thanks

lfwa commented 3 years ago

Thanks for the suggestion!

By cost do you mean the cost in terms of the electricity consumed during training? If so I could imagine this to be region specific and we would need some sources (or even better a single central source) from which we can retrieve this data, hopefully in real-time as it might change throughout the day. The other option would be to simple use a reasonable average (and if so what?). Both of these would require some minor rework of the conversion files (https://github.com/lfwa/carbontracker/tree/master/carbontracker/emissions/conversion) for the energy consumption to be available.

There is also the option of converting the CO2eq to the cost in terms of social cost, etc. Again if we have some reasonable average then it can simple be added to co2eq.csv (https://github.com/lfwa/carbontracker/blob/master/carbontracker/emissions/conversion/co2eq.csv).

Feel free to submit a pull request if you have any suggestions on how this should be done. At the moment I cannot dedicate much time to this project, so contributors would help carbontracker improve.

mpariente commented 3 years ago

Thanks for your answer.

What I meant primarily was the cost of the electricity consumed during training, yes.

And yes, simply using an average would seem reasonable to me. Thanks for being open to contributions, I'll see if I can make a PR.