ray_proj/setup_dagshub_example.py

import os
import time
from git import Repo, GitCommandError

import mlflow
import dagshub

from ray import air, tune
from ray.air import session
from ray.air.integrations.dagshub import setup_dagshub, upload_artifacts

def evaluation_fn(step, width, height):
    return (0.1 + width * step / 100) ** (-1) + height * 0.1

def train_function_dagshub(config):
    # Hyperparameters
    width, height = config["width"], config["height"]
    setup_dagshub(config=config)
    
    for step in range(config.get("steps", 100)):
        # Iterative training function - can be any arbitrary training procedure
        intermediate_score = evaluation_fn(step, width, height)
        # Log the metrics to mlflow
        mlflow.log_metrics(dict(mean_loss=intermediate_score), step=step)
        # Feed the score back to Tune.
        session.report({"iterations": step, "mean_loss": intermediate_score})
        time.sleep(0.1)

def tune_with_setup(save_artifact=False):
    DAGSHUB_REPO = "timho102003/raytest_0607_v12"
    dagshub.init(repo_name=DAGSHUB_REPO.split(os.sep)[1], repo_owner=DAGSHUB_REPO.split(os.sep)[0])

    # Set the experiment, or create a new one if does not exist yet.
    mlflow.set_tracking_uri(os.getenv("MLFLOW_TRACKING_URI", None))
    mlflow.set_experiment(experiment_name="mixin_example")

    tuner = tune.Tuner(
        train_function_dagshub,
        tune_config=tune.TuneConfig(
            num_samples=5
        ),
        run_config=air.RunConfig(
            name="dagshub",
        ),
        param_space={
            "width": tune.randint(10, 100),
            "height": tune.randint(0, 100),
            "steps": 5,
            "dagshub": {
                "experiment_name": "mixin_example",
                "dagshub_repository": DAGSHUB_REPO,
                "log_mlflow_only": False,
            },
        },
    )
    results = tuner.fit()
    if save_artifact:
        print("Save Artifact!")
        upload_artifacts(results=results, repo_name=DAGSHUB_REPO.split(os.sep)[1], repo_owner=DAGSHUB_REPO.split(os.sep)[0])

tune_with_setup(save_artifact=True)