search

Paperspace / gradient-utils

Python utils and helpers library for Gradient
MIT License
3 stars 5 forks source link

feat(metrics): Tensorboard support. #43

Closed colin-welch closed 3 years ago

colin-welch commented 3 years ago

This PR adds a functional sync_tensorboard option to the metrics.init() call, and changes the semantics of that call to leverage with. This implementation parses raw numeric values from Tensorboard events and logs them as metrics in Gradient. Currently, tensor-encoded numeric data is not supported.

linear[bot] commented 3 years ago
CR2-83 Consume tensorboard events and log them as custom metrics

* On new event in the consumer, the event will be parsed and written using our custom metrics library * This could be potentially messy, looking at wandb’s implementation of it * We can use `prometheus_client`’s `Gauge` metrics or the single-line `metrics.log()` function if it is complete