MLFChain API Reference

The MLFChain class is the core orchestrator for building and executing ML pipelines in mlfcrafter.

Class Definition

class MLFChain:
    """
    Initialize MLFChain with multiple crafters
    Usage: MLFChain(DataIngestCrafter(...), CleanerCrafter(...), ...)
    """

Constructor

__init__(self, *crafters)

Creates a new MLFChain instance with the specified crafters.

Parameters:

• *crafters: Variable number of crafter instances to add to the pipeline

Example:

from mlfcrafter import MLFChain
from mlfcrafter.crafters import DataIngestCrafter, CleanerCrafter, ModelCrafter

# Initialize with crafters
pipeline = MLFChain(
    DataIngestCrafter(data_path="data.csv"),
    CleanerCrafter(strategy="auto"),
    ModelCrafter(model_name="random_forest")
)

# Or initialize empty and add crafters later
pipeline = MLFChain()

Methods

add_crafter(crafter)

Adds a single crafter to the pipeline chain.

Parameters:

• crafter: The crafter instance to add to the pipeline

Example:

from mlfcrafter.crafters import DataIngestCrafter, CleanerCrafter

pipeline = MLFChain()
pipeline.add_crafter(DataIngestCrafter(data_path="data.csv"))
pipeline.add_crafter(CleanerCrafter(strategy="auto"))

run(target_column=None, **kwargs)

Runs the entire pipeline chain.

Parameters:

• target_column (str, optional): Target column name for ML tasks
• **kwargs: Additional parameters to pass to the first crafter

Returns: dict - Context dictionary containing all pipeline results

Raises:

• RuntimeError: If any crafter in the pipeline fails
• TypeError: If a crafter doesn't return a dict (context)
• ValueError: If required data or parameters are missing

Example:

# Run pipeline with target column
results = pipeline.run(target_column="target")

# Access results
print(f"Model accuracy: {results['test_score']:.4f}")
print(f"Dataset shape: {results['original_shape']}")

Properties

crafters

List of all crafters in the pipeline.

Type: list

Example:

print(f"Pipeline has {len(pipeline.crafters)} crafters")
for i, crafter in enumerate(pipeline.crafters, 1):
    print(f"Crafter {i}: {type(crafter).__name__}")

Context Dictionary

The run() method returns a context dictionary containing all pipeline results. Key elements include:

Data Processing Results

• data (pd.DataFrame): Current state of the data
• original_shape (tuple): Shape of original data (rows, columns)
• cleaned_shape (tuple): Shape after cleaning (if cleaning was performed)
• missing_values_handled (bool): Whether missing values were processed

Model Training Results

• model: Trained model object
• X_train, X_test (pd.DataFrame): Feature splits
• y_train, y_test (pd.Series): Target splits
• y_pred (np.array): Predictions on test set
• train_score (float): Training accuracy
• test_score (float): Test accuracy
• model_name (str): Name of algorithm used
• features (list): List of feature column names

Scaling Results

• scaler: Fitted scaler object for future use
• scaled_columns (list): Names of columns that were scaled
• scaler_type (str): Type of scaler used

Encoding Results

• encoder: Fitted encoder object for future use
• encoded_columns (list): Names of columns that were encoded
• encoder_type (str): Type of encoder used

Scoring Results

• scores (dict): Dictionary containing calculated metrics

Deployment Results

• deployment_path (str): Path where model was saved
• artifacts_saved (list): List of artifact keys that were saved
• deployment_successful (bool): Whether deployment completed successfully

Example:

results = pipeline.run(target_column="target")

# Access various results
print(f"Original data shape: {results['original_shape']}")
print(f"Model used: {results['model_name']}")
print(f"Test accuracy: {results['test_score']:.4f}")
print(f"Model saved to: {results.get('deployment_path', 'Not saved')}")

Error Handling

MLFChain handles errors that occur during pipeline execution:

Common Exceptions:

• RuntimeError: Raised when a crafter fails during execution
• TypeError: Raised when a crafter doesn't return a dict (context)
• ValueError: Raised when required data or parameters are missing

Example:

try:
    results = pipeline.run(target_column="target")
    print("Pipeline completed successfully!")
except RuntimeError as e:
    print(f"Pipeline failed: {e}")
except ValueError as e:
    print(f"Configuration error: {e}")

Examples

Basic Pipeline

from mlfcrafter import MLFChain
from mlfcrafter.crafters import *

# Create pipeline with all crafters
pipeline = MLFChain(
    DataIngestCrafter(data_path="data.csv"),
    CleanerCrafter(strategy="auto"),
    CategoricalCrafter(encoder_type="onehot")
    ScalerCrafter(scaler_type="standard"),
    ModelCrafter(model_name="random_forest"),
    ScorerCrafter(),
    DeployCrafter()
)

# Run pipeline
results = pipeline.run(target_column="target")
print(f"Model accuracy: {results['test_score']:.4f}")

Adding Crafters Individually

# Create empty pipeline
pipeline = MLFChain()

# Add crafters one by one
pipeline.add_crafter(DataIngestCrafter(data_path="data.csv"))
pipeline.add_crafter(CleanerCrafter(strategy="median"))
pipeline.add_crafter(ModelCrafter(model_name="xgboost"))

# Run pipeline
results = pipeline.run(target_column="target")

Error Handling

try:
    results = pipeline.run(target_column="target")
    print(f"Success! Model accuracy: {results['test_score']:.4f}")
    print(f"Model saved to: {results.get('deployment_path', 'Not saved')}")
except RuntimeError as e:
    print(f"Pipeline failed: {e}")
except ValueError as e:
    print(f"Configuration error: {e}")