Application 11

src/appli11a/.gitignore

@@ -1,3 +1,4 @@
 config.yaml
 __pycache__/
+data/
 titanic/

src/appli11a/README.md

@@ -6,17 +6,11 @@ ayant la structure suivante:
 
 ```yaml
 jeton_api: ####
-train_path: ####
-test_path: ####
-test_fraction: ####
+data_path: https://minio.lab.sspcloud.fr/lgaliana/ensae-reproductibilite/data/raw/data.csv
 ```
 
-## Réutilisation
+Pour installer les dépendances
 
-Pour pouvoir tester ce projet, le code suivant
-suffit:
-
-```python
+```bash
 pip install -r requirements.txt
-python main.py
-```
+```

src/appli11a/main.py

@@ -2,67 +2,70 @@
 Prediction de la survie d'un individu sur le Titanic
 """
 
-# GESTION ENVIRONNEMENT --------------------------------
-
 import argparse
-import src.data.import_data as imp
-import src.features.build_features as bf
-import src.models.train_evaluate as te
-
-
+import pathlib
+import pandas as pd
 
-# PARAMETRES -------------------------------
+from src.data.import_data import import_yaml_config, split_and_count
+from src.pipeline.build_pipeline import split_train_test, create_pipeline
+from src.models.train_evaluate import evaluate_model
 
-# Paramètres ligne de commande
 parser = argparse.ArgumentParser(description="Paramètres du random forest")
-parser.add_argument(
-    "--n_trees", type=int, default=20, help="Nombre d'arbres"
-)
+parser.add_argument("--n_trees", type=int, default=20, help="Nombre d'arbres")
 args = parser.parse_args()
 
-# Paramètres YAML
-config = imp.import_yaml_config("configuration/config.yaml")
-base_url = "https://minio.lab.sspcloud.fr/projet-formation/ensae-reproductibilite/data/raw"
-API_TOKEN = config.get("jeton_api")
-LOCATION_TRAIN = config.get("train_path", f"{base_url}/train.csv")
-LOCATION_TEST = config.get("test_path", f"{base_url}/test.csv")
-TEST_FRACTION = config.get("test_fraction", .1)
-N_TREES = args.n_trees
+n_trees = args.n_trees
+
+URL_RAW = "https://minio.lab.sspcloud.fr/lgaliana/ensae-reproductibilite/data/raw/data.csv"
+config = import_yaml_config("configuration/config.yaml")
+jeton_api = config.get("jeton_api")
+data_path = config.get("data_path", URL_RAW)
+data_train_path = config.get("train_path", "data/derived/train.csv")
+data_test_path = config.get("test_path", "data/derived/test.csv")
 
+MAX_DEPTH = None
+MAX_FEATURES = "sqrt"
 
-# FEATURE ENGINEERING --------------------------------
 
-print(LOCATION_TRAIN)
+# IMPORT ET EXPLORATION DONNEES --------------------------------
 
-TrainingData = imp.import_data(LOCATION_TRAIN)
-TestData = imp.import_data(LOCATION_TEST)
+TrainingData = pd.read_csv(data_path)
 
-# Create a 'Title' variable
-TrainingData = bf.create_variable_title(TrainingData)
-TestData = bf.create_variable_title(TestData)
 
+# Usage example:
+ticket_count = split_and_count(TrainingData, "Ticket", "/")
+name_count = split_and_count(TrainingData, "Name", ",")
 
-## IMPUTATION DES VARIABLES ================
 
-TrainingData = bf.fill_na_titanic(TrainingData)
-TestData = bf.fill_na_titanic(TestData)
+# SPLIT TRAIN/TEST --------------------------------
 
-TrainingData = bf.label_encoder_titanic(TrainingData)
-TestData = bf.label_encoder_titanic(TestData)
+p = pathlib.Path("data/derived/")
+p.mkdir(parents=True, exist_ok=True)
 
+X_train, X_test, y_train, y_test = split_train_test(
+    TrainingData, test_size=0.1,
+    train_path=data_train_path,
+    test_path=data_test_path
+)
+
+
+# PIPELINE ----------------------------
+
+
+# Create the pipeline
+pipe = create_pipeline(
+    n_trees, max_depth=MAX_DEPTH, max_features=MAX_FEATURES
+)
 
-# Making a new feature hasCabin which is 1 if cabin is available else 0
-TrainingData = bf.check_has_cabin(TrainingData)
-TestData = bf.check_has_cabin(TestData)
 
-TrainingData = bf.ticket_length(TrainingData)
-TestData = bf.ticket_length(TestData)
+# ESTIMATION ET EVALUATION ----------------------
 
+pipe.fit(X_train, y_train)
 
-# MODELISATION: RANDOM FOREST ----------------------------
 
-model = te.random_forest_titanic(
-    data=TrainingData,
-    fraction_test=TEST_FRACTION,
-    n_trees=N_TREES
-)
+# Evaluate the model
+score, matrix = evaluate_model(pipe, X_test, y_test)
+print(f"{score:.1%} de bonnes réponses sur les données de test pour validation")
+print(20 * "-")
+print("matrice de confusion")
+print(matrix)