Experiments

The experiment lies at the heart of machine learning. Comet has a full set of experiment types depending on different use cases. The experiment classes are:

1. Experiment: constructs a live-streaming, full experiment. This is typically the only class you need. It streams the logging in the background to the cloud so that your code runs as fast as it normally does. The Experiment class supports all of Comet's auto-logging functionality, and full support of the Optimizer.
2. ExistingExperiment: continue an experiment in live-streaming mode. If you wish to continue training, this is probably the class you need. It streams the logging in the background to the cloud so that your code runs as fast as it normally does. The ExistingExperiment class supports all of Comet's auto-logging functionality, and full support of the Optimizer.
3. OfflineExperiment: constructs an offline experiment. It streams the logging in the background to the file system so that your code runs as fast as it normally does. Useful when you don't have an internet connection, or the connection is intermittent. The OfflineExperiment class supports all of Comet's auto-logging functionality.
4. ExistingOfflineExperiment: (combination of above ExistingExperiment and OfflineExperiment) constructs an offline experiment to continue an existing experiment. It streams the logging in the background to the file system so that your code runs as fast as it normally does. Useful when you have an existing experiment, don't have an internet connection, or the connection is intermittent. The ExistingOfflineExperiment class supports all of Comet's auto-logging functionality.
5. APIExperiment and the Python API: constructs, continues, or queries an experiment in non-streaming mode. Useful for quick experiment updates, or for querying and analyzing your experiment data offline.