This is a set of unofficial Python bindings for the CometML REST API. There are functions for all current endpoints as well as a couple of functions that build on the basic endpoint ones to provide, e.g., simpler output or filtering. I haven't used all of the endpoints myself, so some of the functions haven't been tested at all (e.g. get_html). Documentation is currently limited but the functions provided are generally quite simple.
See Endpoints for the official documentation of the REST API.
pip install cometml_apiNote that there is a different project called comet_ml_api (note the extra underscore); don't accidentally install that one when you're trying to get this one!
Python 2 is not supported. This should work on any OS, but I've only tested on Ubuntu (and a very small amount on Windows).
See this page first for how to generate your REST API key. There are two possible methods of authentication using that key:
~/.comet_rest_key. When you import the api script, it will automatically load a key that it finds in this file.set_api_key function and pass your key in. The key will be saved in a global variable so that the same key is used for all subsequent requests unless you explicitly set a new key.Most of the basic endpoint functions are named get_<endpoint> where <endpoint> is the corresponding endpoint (e.g. get_workspaces to access https://www.comet.ml/api/rest/v1/workspaces). However, there are a few exceptions.
There are two endpoints for metrics: metrics and metrics-raw. I call both of these "raw" because they return data that isn't well-formatted for immediate plotting/analysis. As the metrics endpoint only return the min, max, and most recent data points for a given metric, I call that one a summary, hence get_raw_metric_summaries; the function for metrics-raw is get_raw_metrics. There is also a helper function get_metrics which converts the metrics into a better format for visualization or analysis.
Similarly to metrics, the raw parameters data may not be in the most usable format right away. I thus also call this endpoint get_raw_params and have a helper function get_params which provides a more concise output.
As for params except, though the endpoint is log-other, the functions are get_raw_others and get_others.
The images endpoint doesn't return the images themselves, just the data about them (including the URLs from which the actual images can be downloaded?). I call this endpoint get_image_data, but I haven't tested it.
from cometml_api import api
workspaces = api.get_workspaces()
project_ids = api.get_project_names_and_ids(workspaces[0]) # {name: id}
experiments = api.get_experiments(project_ids.popitem()[1])
api.get_params(experiments[0]["experiment_key"])