---
title: "Setup and Quickstart"
excerpt: "Learn how to setup MLflow and run a quickstart"
tags: ['mlops', 'mlflow', 'python', 'machine-learning']
---

## Local setup 

**Define the dependencies in a requirements.txt file:**

```txt title="requirements.txt"
pandas==2.3.3
numpy==2.2.6
scikit-learn==1.7.2
joblib==1.5.2
pyyaml==6.0.3
python-dotenv==1.2.3
mlflow==3.17.1
```

**Install the packages:**

```bash
# create project venv
uv venv

# install only packages listed in requirements.txt (plus their transitive deps)
uv pip install -r requirements.txt

# optional: install pip so MLflow can pin it in conda.yaml when logging models
uv pip install pip

# activate (optional if you use `uv run`)
source .venv/bin/activate
```

<Guide title="Quickstart" numbered headingLevel={2} stepHeadingLevel={3}>
<GuideStep title="Step 1. Define baseline ML code">

- [MLflow Quickstart](https://mlflow.org/docs/latest/ml/getting-started/quickstart/)

```python title="mlflow_base.py"
import argparse
import mlflow
import pandas as pd
import numpy as np
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.linear_model import ElasticNet

# Load the Iris dataset
X, y = datasets.load_iris(return_X_y=True)

# Split the data into training and test sets
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)
parser = argparse.ArgumentParser()
parser.add_argument("--alpha", type=float, required=False, default=0.7)
parser.add_argument("--l1_ratio", type=float, required=False, default=0.7)
args = parser.parse_args()

# Define the model hyperparameters
params = {
    "random_state": 8888,
    "alpha": args.alpha,
    "l1_ratio": args.l1_ratio,
}


# evaluation function
def eval_metrics(actual, pred):
    rmse = np.sqrt(mean_squared_error(actual, pred))
    mae = mean_absolute_error(actual, pred)
    r2 = r2_score(actual, pred)
    return rmse, mae, r2
```
</GuideStep>


<GuideStep title="Step 2. Setup MLflow">

```python title="mlflow_base.py" showLineNumbers {2-3}
# mlflow mendatory configs
mlflow.set_tracking_uri("sqlite:///mlflow.db")
mlflow.set_experiment("my-first-experiment")
```
Here, we Set the tracking URI to a local database URI (e.g., `sqlite:///mlflow.db`). This is recommended option for quickstart and local development.
And set the experiment name to `my-first-experiment`.

- [https://mlflow.org/docs/latest/ml/getting-started/running-notebooks/](https://mlflow.org/docs/latest/ml/getting-started/running-notebooks/)


</GuideStep>

<GuideStep title="Step 3 - Train a model with MLflow Autologging">

In this step, we train the model on the training data loaded in the previous step, and log the model and its metadata to MLflow. The easiest way to do this is to using MLflow's `Autologging` feature.


```python title="mlflow_base.py" showLineNumbers {2}
# Enable autologging for scikit-learn
mlflow.sklearn.autolog()  # this will log the dataset, model, params, metrics, etc.

# Just train the model normally
lr = ElasticNet(**params)
lr.fit(X_train, y_train)

# Evaluate the model
y_pred = lr.predict(X_test)
(rmse, mae, r2) = eval_metrics(y_test, y_pred)

print("Elasticnet model (alpha={:f}, l1_ratio={:f}):".format(args.alpha, args.l1_ratio))
print("  RMSE: %s" % rmse)
print("  MAE: %s" % mae)
print("  R2: %s" % r2)
```


With just one line of additional code `mlflow.sklearn.autolog()`, now you get the best of both worlds: you can focus on training the model, and MLflow will take care of the rest:

- Saving the trained model.
- Recording the model's performance metrics during training, such as accuracy, precision, AUC curve.
- Logging hyperparameter values used to train the model.
- Track metadata such as input data format, user, timestamp, etc.

To learn more about autologging and supported libraries, see the [Autologging documentation](https://mlflow.org/docs/latest/ml/tracking/autolog/).



Full code:


```python title="mlflow_base.py"
import argparse
import mlflow
import pandas as pd
import numpy as np
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.linear_model import ElasticNet

# Load the Iris dataset
X, y = datasets.load_iris(return_X_y=True)

# Split the data into training and test sets
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)
parser = argparse.ArgumentParser()
parser.add_argument("--alpha", type=float, required=False, default=0.7)
parser.add_argument("--l1_ratio", type=float, required=False, default=0.7)
args = parser.parse_args()

# Define the model hyperparameters
params = {
    "random_state": 8888,
    "alpha": args.alpha,
    "l1_ratio": args.l1_ratio,
}


# evaluation function
def eval_metrics(actual, pred):
    rmse = np.sqrt(mean_squared_error(actual, pred))
    mae = mean_absolute_error(actual, pred)
    r2 = r2_score(actual, pred)
    return rmse, mae, r2


# mlflow mendatory configs
mlflow.set_tracking_uri("sqlite:///mlflow.db")
mlflow.set_experiment("my-first-experiment")
# Enable autologging for scikit-learn
mlflow.sklearn.autolog()  # this will log the dataset, model, params, metrics, etc.

# Just train the model normally
lr = ElasticNet(**params)
lr.fit(X_train, y_train)

# Evaluate the model
y_pred = lr.predict(X_test)
(rmse, mae, r2) = eval_metrics(y_test, y_pred)

print("Elasticnet model (alpha={:f}, l1_ratio={:f}):".format(args.alpha, args.l1_ratio))
print("  RMSE: %s" % rmse)
print("  MAE: %s" % mae)
print("  R2: %s" % r2)
```

<GuideStep title="Step 4. Run the training and open the MLflow UI">

```bash 
uv run python mlflow_base.py
# Run multiple times with different arguments to the script. e.g. --alpha 0.01 --l1_ratio 0.01
uv run python mlflow_base.py --alpha 0.01 --l1_ratio 0.01
uv run python mlflow_base.py --alpha 0.02 --l1_ratio 0.02
```

Open the MLflow UI:

```bash
uv run mlflow ui
```

</GuideStep>

</GuideStep>

<GuideStep title="Step 5. Explore the runs in the MLflow UI">

Open http://127.0.0.1:5000 and you should see the runs.


Here are some screenshots of the MLflow UI:

<Figure
    src="/learn/mlops/mlflow/mlflow1.png"
    alt="Vist: Model Training -> Experiments --> my-first-experiment"
    caption="Vist: Model Training -> Experiments --> my-first-experiment"
    width="960"

/>

<Figure
    src="/learn/mlops/mlflow/mlflow2.png"
    alt="Vist: Model Training -> Experiments --> my-first-experiment --> Runs"
    caption="Vist: Model Training -> Experiments --> my-first-experiment --> Runs"
    width="960"

/>

<Figure
    src="/learn/mlops/mlflow/mlflow3.png"
    alt="Add more columns to the runs table"
    caption="Add more columns to the runs table"
    width="960"

/>
<Figure
    src="/learn/mlops/mlflow/mlflow4.png"
    alt="Vist: Model Training -> Experiments --> my-first-experiment --> Runs --> Run Details"
    caption="Vist: Model Training -> Experiments --> my-first-experiment --> Runs --> Run Details"
    width="960"

/>
<Figure
    src="/learn/mlops/mlflow/mlflow5.png"
    alt="Chart View of the runs"
    caption="Chart View of the runs"
    width="960"

/>
<Figure
    src="/learn/mlops/mlflow/mlflow6.png"
    alt="Select multiple runs and compare them"
    caption="Select multiple runs and compare them"
    width="960"

/>

<Figure
    src="/learn/mlops/mlflow/mlflow7.png"
    alt="Multiple runs comparison"
    caption="Multiple runs comparison"
    width="960"

/>


</GuideStep>


</Guide>