THE RUN LOGGER

Difficulty: Easy Tags: MLOps

Story

DoorDash's ML platform requires every training run, no exceptions, to log its full configuration and final metrics before the model artifact is allowed into the registry. This is the logger every later MLOps problem in this set assumes already exists.

The Problem

No math here: this is a structured-serialization problem. Given a run id, a set of hyperparameter key/value pairs, and a set of metric key/value pairs, produce a single canonical serialized record.

Input Format

run_id
h
key_1 val_1
...
key_h val_h
m
metric_1 val_1
...
metric_m val_m

Output Format

A single JSON object with keys run_id, hyperparameters (object), metrics (object). Hyperparameter and metric keys are sorted alphabetically within their own object.

Constraints

• 0 <= h, m <= 100
• Time limit: 1.0 second.

Example

Input

run_0042
2
lr 0.001
batch_size 64
1
val_loss 0.231

Output

{
	"run_id": "run_0042",
	"hyperparameters": { "batch_size": "64", "lr": "0.001" },
	"metrics": { "val_loss": "0.231" }
}