Kind Stdout File TensorFlow Event Prometheus Custom None
Metrics file TensorFlow Events Directory
HTTP Port HTTP Path
Scheme HTTP HTTPS Host Name (optional)
Define how Katib should collect the metrics from each trial, such as the accuracy and loss metrics. Katib collects the metrics from an arbitrary file, where the Training Container should log them. You can also check an example YAML . Katib collects the metrics from the operating system’s default output location (standard output). Katib collects the metrics from a directory path containing a tf.Event. You can also check the TFJob MNIST with Summaries example. You must define your own container that will be responsible for gathering the metrics. You can also check the custom metrics collector example.