Metrics Overview

This guide describes the different types of evaluation metrics and how you can view them.

Experiments and Deployments on Gradient can record metrics which are available both in realtime or after they are finished running. Gradient will display these metrics in the web UI and they can also be queried or streamed in the CLI.

We log three different kinds of metrics: hardware metrics, framework metrics, and custom user metrics.

Note: Framework and custom metrics are only available in Gradient Enterprise. Contact Sales for inquiries!

Hardware metrics

All Gradient workloads like Experiments and Deployments monitor and track CPU, Memory, and Network. If the machine is equipped with a GPU, this will be tracked as well.

System Metrics showing CPU and Memory Usage

Framework metrics

For example, accuracy and mean squared errors are two common metrics for classification and regression, respectively.

If your deployment uses TF Serving, some metrics such astensorflow:core:direct_session_runs, tensorflow:cc:saved_model:load_attempt_count etc. will be logged automatically.

User metrics

You can log custom user metrics from inside of a experiment or deployment using the Python CLI utils. Its based on prometheus Python Client. Here's a trivial example:

from gradient_utils.metrics import
logger = MetricsLogger(grouping_key={'ProjectA': 'SomeLabel'})
logger.add_gauge("Gauge")
logger.add_counter("Counter")
while datetime.now() <= endAt:
randNum = randint(1, 100)
logger["Gauge"] = 5
logger["Gauge"].set(randNum)
logger["Counter"].inc()
logger.push_metrics()