Metric format

HyperOptimizer collects metrics from stdout lines that match the hpo.metrics.<key>=<json> format.

Format

hpo.metrics.<key>=<json>

Name
hpo.metrics.
Type
prefix
Description
Required prefix. The collector ignores lines without it.
Name
<key>
Type
string
Description
Metric name, such as sharpe, loss, or max_drawdown.
Name
<json>
Type
JSON value
Description
JSON-serializable value produced by your program.

Valid values

Use JSON numbers, strings, booleans, arrays, or objects. Numeric objective metrics are the most useful for optimization.

hpo.metrics.loss=0.182
hpo.metrics.accuracy=0.94
hpo.metrics.valid=true
hpo.metrics.notes="completed"

Examples

import json

def emit_metric(key, value):
    print(f"hpo.metrics.{key}={json.dumps(value, default=str)}")

emit_metric("objective", 0.84)
emit_metric("duration_seconds", 42.1)

stdoutcollected

hpo.metrics.objective=0.84

hpo.metrics.duration_seconds=42.1

hpo.metrics.status="completed"

Common mistakes

Missing prefix

metrics.loss=0.2 will not be collected. Use hpo.metrics.loss=0.2.

Invalid JSON

hpo.metrics.loss=nan is not valid JSON. Convert special values before printing.

Wrong objective name

If the dashboard expects sharpe, printing sharp will not satisfy the objective.

Only writing files

Writing metrics to a JSON file is not enough for collection. Print them to stdout.