Full Model Training¶

Welcome to the notebook that will guide you in using MLflow to register and track the training of a logistic regression model with the Titanic dataset. In this tutorial, we will explore how MLflow, an open-source platform for the machine learning lifecycle, can facilitate experiment tracking, metric monitoring, and model management in a reproducible environment.

The Titanic dataset is a classic in the data science community, containing detailed information about passengers, such as their age, gender, ticket class, and whether they survived the disaster. We will use this information to build a logistic regression model capable of predicting the survival probability of a passenger based on their features.

Throughout this notebook, we will not only focus on training the logistic regression model but also learn how to leverage the capabilities of MLflow. We will start by exploring the Titanic dataset, performing analysis and visualizations to better understand the distribution of variables and patterns present.

Next, we will move on to the data preprocessing stage, where we will clean and transform the data to make it suitable for modeling. During this process, we will use MLflow to log the preprocessing steps, allowing us to have a complete and organized record of the transformations applied to the data.

Then, we will dive into training the logistic regression using the training dataset. Here, MLflow will play a crucial role in logging training details, including the hyperparameters used, performance metrics, and other relevant aspects of the model. This will enable us to have a comprehensive overview of the process and facilitate comparison between different experiments and configurations.

Once we have trained our model, we will use MLflow to register the model on the platform. This will allow us to have a complete record of the trained models, including the details of each model, such as the hyperparameters used, performance metrics, and associated source code. Additionally, MLflow will enable us to export the model in a standard format, making it easier to deploy in different environments.

In summary, this notebook will provide you with the opportunity to work with MLflow and discover how this powerful tool can simplify and enhance the training process of a logistic regression model. Throughout the tutorial, you will learn to log experiments, metrics, and model details, allowing you to have a comprehensive and reproducible tracking of the entire process. So get ready to delve into Titanic data analysis while exploring the capabilities of MLflow in logistic regression training!

In [1]:

from datetime import datetime

from sklearn.preprocessing import LabelEncoder  
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
import seaborn as sns
import pandas as pd
import mlflow

from mlops_course.settings import config

from datetime import datetime from sklearn.preprocessing import LabelEncoder from sklearn.model_selection import train_test_split from sklearn.linear_model import LogisticRegression import seaborn as sns import pandas as pd import mlflow from mlops_course.settings import config

Get the Data¶

We will load the dataset from a CSV file and display the first few rows to get an understanding of the data.

Read the data¶

In [2]:

DATASET_URL = "https://raw.githubusercontent.com/datasciencedojo/datasets/master/titanic.csv"


# Load dataset
dataframe = pd.read_csv(DATASET_URL)
dataframe.head()

DATASET_URL = "https://raw.githubusercontent.com/datasciencedojo/datasets/master/titanic.csv" # Load dataset dataframe = pd.read_csv(DATASET_URL) dataframe.head()

Out[2]:

	PassengerId	Survived	Pclass	Name	Sex	Age	SibSp	Parch	Ticket	Fare	Cabin	Embarked
0	1	0	3	Braund, Mr. Owen Harris	male	22.0	1	0	A/5 21171	7.2500	NaN	S
1	2	1	1	Cumings, Mrs. John Bradley (Florence Briggs Th...	female	38.0	1	0	PC 17599	71.2833	C85	C
2	3	1	3	Heikkinen, Miss. Laina	female	26.0	0	0	STON/O2. 3101282	7.9250	NaN	S
3	4	1	1	Futrelle, Mrs. Jacques Heath (Lily May Peel)	female	35.0	1	0	113803	53.1000	C123	S
4	5	0	3	Allen, Mr. William Henry	male	35.0	0	0	373450	8.0500	NaN	S

Clean the Data¶

In [3]:

dataframe.isnull().sum()

dataframe.isnull().sum()

Out[3]:

PassengerId      0
Survived         0
Pclass           0
Name             0
Sex              0
Age            177
SibSp            0
Parch            0
Ticket           0
Fare             0
Cabin          687
Embarked         2
dtype: int64

We see that there are missing values in the 'Age', 'Cabin', and 'Embarked' columns. We'll fill the missing values in the 'Age' column with the median age.

In [4]:

# Fill age missing values
median_age = dataframe['Age'].median()
dataframe['Age'].fillna(median_age, inplace=True)

# Fill age missing values median_age = dataframe['Age'].median() dataframe['Age'].fillna(median_age, inplace=True)

Sex is a categorical variable composed by "male" and "female" string values. In order to train our model we need to convert them to numerical values: male=1, female=0

In [5]:

# convert string labels to numbers: male=1, female=0
le = LabelEncoder()
dataframe["Sex"] = le.fit_transform(dataframe["Sex"])

# convert string labels to numbers: male=1, female=0 le = LabelEncoder() dataframe["Sex"] = le.fit_transform(dataframe["Sex"])

Visualize the data¶

In [6]:

sns.pairplot(dataframe[["Age", "Sex", "Fare", "Survived"]], hue="Survived")

sns.pairplot(dataframe[["Age", "Sex", "Fare", "Survived"]], hue="Survived")

Out[6]:

<seaborn.axisgrid.PairGrid at 0x7f9b740b76d0>

Split the data¶

In [7]:

RANDOM_SEED = 42  # Set a random seed for reproducibility
TEST_SIZE = 0.2  # Use 20% of the data for testing

# Prepare the dataset for training
features = ['Sex', 'Age', 'Fare']
X = dataframe[features]
Y = dataframe['Survived']

# Split the dataset into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=TEST_SIZE, random_state=RANDOM_SEED)

RANDOM_SEED = 42 # Set a random seed for reproducibility TEST_SIZE = 0.2 # Use 20% of the data for testing # Prepare the dataset for training features = ['Sex', 'Age', 'Fare'] X = dataframe[features] Y = dataframe['Survived'] # Split the dataset into training and testing sets X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=TEST_SIZE, random_state=RANDOM_SEED)

Train the Model¶

We'll train the model and we will use MLflow to log that model: logging a model means saving the model to a file or a database so that it can be used later for inference.

Connect to the MLflow server¶

In [8]:

# Set the URI where the MLflow server is running
mlflow.set_tracking_uri(uri=config.MLFLOW_TRACKING_URI)
print("✅ Successfully connected to the MLflow server")

# Set the URI where the MLflow server is running mlflow.set_tracking_uri(uri=config.MLFLOW_TRACKING_URI) print("✅ Successfully connected to the MLflow server")

✅ Successfully connected to the MLflow server

Create the experiment¶

In [9]:

EXPERIMENT_NAME = "Titanic Linear Regression"  # change this to your experiment name


# Create an experiment if it doesn't exist
try:
    experiment_id = mlflow.create_experiment(EXPERIMENT_NAME)
    print(f"✅ Created '{EXPERIMENT_NAME}'!")
except mlflow.exceptions.RestException:
    experiment_id = mlflow.get_experiment_by_name(EXPERIMENT_NAME).experiment_id
    print(f"✅ Experiment '{EXPERIMENT_NAME}' already exists!")

experiment = mlflow.get_experiment(experiment_id)

EXPERIMENT_NAME = "Titanic Linear Regression" # change this to your experiment name # Create an experiment if it doesn't exist try: experiment_id = mlflow.create_experiment(EXPERIMENT_NAME) print(f"✅ Created '{EXPERIMENT_NAME}'!") except mlflow.exceptions.RestException: experiment_id = mlflow.get_experiment_by_name(EXPERIMENT_NAME).experiment_id print(f"✅ Experiment '{EXPERIMENT_NAME}' already exists!") experiment = mlflow.get_experiment(experiment_id)

✅ Experiment 'Titanic Linear Regression' already exists!

Train the model and log it to MLflow¶

In [10]:

run_name = datetime.now().strftime("%Y-%m-%d %H:%M:%S")

with mlflow.start_run(
    experiment_id=experiment_id,
    run_name=run_name,
) as run:
    
    # Train the logistic regression model
    model = LogisticRegression(random_state=RANDOM_SEED)
    model.fit(X_train.values, y_train)

    # Log the model itself to MLflow
    mlflow.sklearn.log_model(model, "model")

    # Print the run ID
    print(f"Run ID: {run.info.run_id}")

run_name = datetime.now().strftime("%Y-%m-%d %H:%M:%S") with mlflow.start_run( experiment_id=experiment_id, run_name=run_name, ) as run: # Train the logistic regression model model = LogisticRegression(random_state=RANDOM_SEED) model.fit(X_train.values, y_train) # Log the model itself to MLflow mlflow.sklearn.log_model(model, "model") # Print the run ID print(f"Run ID: {run.info.run_id}")

Run ID: f14b6da63a47497eb837518537dfcdb2

/usr/local/lib/python3.9/site-packages/_distutils_hack/__init__.py:33: UserWarning: Setuptools is replacing distutils.
  warnings.warn("Setuptools is replacing distutils.")