domino-experiment-setup git:20261001.b8e8054A

Immutable. This exact content never changes and is served at /api/v1/blob/10c5bac1853aa777.

---
name: domino-experiment-setup
description: Set up MLflow experiment tracking in a Domino project by generating an experiment_setup.py helper with deployment-unique experiment names, Domino context tags, and framework auto-logging, plus an example training script. Use when the user asks to add or set up experiment tracking, MLflow logging, or autologging for a traditional ML project in Domino.
---

# Set Up Domino Experiment Tracking

Add MLflow experiment tracking to a traditional ML project. For LLM or agent tracing, use the
`domino-trace-setup` skill instead.

## Inputs

1. **Experiment base name**: ask if not given. Default to the project name.
2. **ML frameworks**: detect them from `requirements.txt`, `pyproject.toml`, `environment.yml`,
   or imports in the code. Confirm the list with the user.

## Steps

1. Copy `assets/experiment_setup.py` from this skill's folder into the project. It provides:
   - `setup_experiment(base_name)`: sets an experiment named `<base>-<project>-<username>`.
   - `log_domino_context()`: tags the run with the Domino user, project, run ID, and hardware tier.
   - `setup_autolog()`: enables MLflow autologging for whichever supported frameworks are installed.
2. Report which frameworks were detected and will be autologged, for example:

   ```
   Detected frameworks:
   ✅ scikit-learn (requirements.txt)
   ✅ xgboost (requirements.txt)
   ❌ tensorflow (not found)
   ```

3. If the project trains large models, add the multipart-upload settings below.
4. Create an example training script, or wire the helpers into the user's existing script.

## Example training script

```python
# train.py
from experiment_setup import setup_experiment, log_domino_context, setup_autolog
import mlflow
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_iris

experiment_name = setup_experiment("iris-classifier")
setup_autolog()

iris = load_iris()
X_train, X_test, y_train, y_test = train_test_split(
    iris.data, iris.target, test_size=0.2, random_state=42
)

with mlflow.start_run(run_name="random-forest-v1"):
    log_domino_context()
    mlflow.log_param("custom_param", "value")

    model = RandomForestClassifier(n_estimators=100, max_depth=5)
    model.fit(X_train, y_train)

    test_accuracy = model.score(X_test, y_test)
    mlflow.log_metric("test_accuracy", test_accuracy)
    print(f"Test accuracy: {test_accuracy:.4f}")
    print(f"Run ID: {mlflow.active_run().info.run_id}")
```

## Important rules

- **Experiment names must be unique across the entire Domino deployment**, not just the project.
  Always keep the username and project suffix that `setup_experiment` adds.
- For large artifacts (LLMs, deep learning checkpoints), enable multipart upload:

  ```python
  os.environ['MLFLOW_ENABLE_PROXY_MULTIPART_UPLOAD'] = "true"
  os.environ['MLFLOW_MULTIPART_UPLOAD_CHUNK_SIZE'] = "104857600"  # 100MB
  ```

For run comparison, the model registry, and deeper MLflow usage, use the
`domino-experiment-tracking` skill.