prediction.py

import joblib
import pandas as pd
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.preprocessing import OneHotEncoder
from sklearn.cluster import KMeans

# Custom Transformer: UnitPriceTransformer
class UnitPriceTransformer(BaseEstimator, TransformerMixin):
    def fit(self, X, y=None):
        return self
    
    def transform(self, X):
        X = X.copy()  # Work on a copy to avoid SettingWithCopyWarning
        X['unit_price'] = X['sales'] / X['quantity']
        return X[['unit_price']]

# Custom Transformer: KMeansAndLabelTransformer
class KMeansAndLabelTransformer(BaseEstimator, TransformerMixin):
    def __init__(self, n_clusters=3):
        self.n_clusters = n_clusters
        self.kmeans = KMeans(n_clusters=n_clusters, random_state=42)
    
    def fit(self, X, y=None):
        self.kmeans.fit(X[['unit_price']])
        return self
    
    def transform(self, X):
        X = X.copy()  # Work on a copy to avoid SettingWithCopyWarning
        cluster_labels = self.kmeans.predict(X[['unit_price']])
        X['distinct_cluster_label'] = cluster_labels.astype(str) + "_" + X['sub_category']
        return X[['distinct_cluster_label']]

# Custom Transformer: DynamicOneHotEncoder
class DynamicOneHotEncoder(BaseEstimator, TransformerMixin):
    def __init__(self):
        self.encoder = OneHotEncoder(handle_unknown='ignore')
    
    def fit(self, X, y=None):
        self.encoder.fit(X[['distinct_cluster_label']])
        return self
    
    def transform(self, X):
        X = X.copy()  # Work on a copy to avoid SettingWithCopyWarning
        encoded_features = self.encoder.transform(X[['distinct_cluster_label']]).toarray()
        # Create a DataFrame with the encoded features
        encoded_df = pd.DataFrame(encoded_features, columns=self.encoder.get_feature_names_out(['distinct_cluster_label']))
        return encoded_df

# Load the pipeline and model
pipeline = joblib.load('full_pipeline_with_unit_price.pkl')
model = joblib.load('best_model.pkl')

def make_prediction(input_features):
    processed_features = pipeline.transform(pd.DataFrame([input_features]))
    prediction = model.predict(processed_features)
    return prediction[0]