Merge pull request #31 from rissangs/main

Add fit_wine_quality_predict_model.py file

src/fit_wine_quality_predict_model.py

@@ -0,0 +1,147 @@
+# author: UBC Master of Data Science - Group 33
+# date: 2020-11-28
+
+
+"""Model building and fitting the data.
+
+Usage: src/fit_wine_quality_predict_model.py --in_file_1=<in_file_1> --out_dir=<out_dir> 
+
+Options:
+--in_file_1=<in_file_1>      Path (including file name) to first raw data which is for red wine
+--out_dir=<out_dir>          Path (excluding file name) of where to locally write the file
+
+"""
+
+from docopt import docopt
+import joblib
+import os
+import string
+from collections import deque
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+
+# data
+from sklearn import datasets
+from sklearn.compose import ColumnTransformer, make_column_transformer
+from sklearn.dummy import DummyClassifier, DummyRegressor
+from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
+from sklearn.feature_extraction.text import CountVectorizer
+
+# Feature selection
+from sklearn.feature_selection import RFE, RFECV
+from sklearn.impute import SimpleImputer
+
+# classifiers / models
+from sklearn.linear_model import RidgeClassifier
+from sklearn.linear_model import LogisticRegression
+
+# other
+from sklearn.metrics import accuracy_score, log_loss, make_scorer, mean_squared_error
+from sklearn.model_selection import (
+    GridSearchCV,
+    RandomizedSearchCV,
+    ShuffleSplit,
+    cross_val_score,
+    cross_validate,
+    train_test_split,
+)
+from sklearn.pipeline import Pipeline, make_pipeline
+from sklearn.preprocessing import (
+    OneHotEncoder,
+    OrdinalEncoder,
+    PolynomialFeatures,
+    StandardScaler,
+)
+from sklearn.neighbors import KNeighborsClassifier
+from sklearn.neural_network import MLPClassifier
+from sklearn.neighbors import NearestCentroid
+from sklearn.discriminant_analysis import QuadraticDiscriminantAnalysis
+from sklearn.model_selection import cross_val_predict
+from sklearn.metrics import plot_precision_recall_curve, plot_roc_curve
+
+opt = docopt(__doc__)
+
+def store_cross_val_results(model_name, scores, results_dict):
+    """
+    Stores mean scores from cross_validate in results_dict for
+    the given model model_name.
+    Parameters
+    ----------
+    model_name :
+        scikit-learn classification model
+    scores : dict
+        object return by `cross_validate`
+    results_dict: dict
+        dictionary to store results
+    Returns
+    ----------
+        None
+    """
+    results_dict[model_name] = {
+        "mean_fit_time": "{:0.4f}".format(np.mean(scores["fit_time"])),
+        "mean_score_time": "{:0.4f}".format(np.mean(scores["score_time"])),
+        "mean_test_f1 (s)": "{:0.4f}".format(np.mean(scores["test_f1_micro"])),
+        "mean_train_f1 (s)": "{:0.4f}".format(np.mean(scores["train_f1_micro"])),
+        "mean_test_accuracy (s)": "{:0.4f}".format(np.mean(scores["test_accuracy"])),
+        "mean_train_accuracy (s)": "{:0.4f}".format(np.mean(scores["train_accuracy"])),
+    }
+
+
+def main(in_file_1, out_dir):
+    # read data and combine two data set vertically
+    train_df = pd.read_csv(in_file_1)
+    X_train = train_df.drop(columns = ['quality'])
+    y_train = train_df['quality']
+
+    #-----------------------------------------------------------------------------------------------------------------------------
+    #PreProcessor 
+    numeric_features = ['fixed acidity', 'volatile acidity', 'citric acid', 'residual sugar', 
+                        'chlorides', 'free sulfur dioxide','total sulfur dioxide', 'density', 'pH', 'sulphates', 'alcohol']
+    binary_features = ['type']
+
+    numeric_transformer = make_pipeline(SimpleImputer(), StandardScaler())
+    binary_transformer = make_pipeline(OneHotEncoder(drop="if_binary", dtype=int))
+
+    preprocessor = ColumnTransformer(
+        transformers = [
+            ('num', numeric_transformer, numeric_features),
+            ('bin', binary_transformer, binary_features)
+        ]
+    )
+    #-----------------------------------------------------------------------------------------------------------------------------
+    #Model picking
+    results_df={}
+    scoring ={'accuracy', 'f1_micro'}
+
+    pipe_iter = make_pipeline(preprocessor, MLPClassifier(random_state=1, max_iter=300))
+    scores_iter = cross_validate(pipe_iter, X_train, y_train, 
+                    return_train_score=True,scoring = scoring)
+    store_cross_val_results(f"preprocessing + MLP", scores_iter, results_df)
+    print(store_cross_val_results)
+
+    #-----------------------------------------------------------------------------------------------------------------------------
+    #Hyperparameters Tuning
+    rf_pipeline = make_pipeline(
+        preprocessor, MLPClassifier())
+
+    param_dist = {
+        'mlpclassifier__hidden_layer_sizes': [(50,50,50), (50,100,50), (100,)],
+        'mlpclassifier__activation': ['tanh', 'relu'],
+        'mlpclassifier__solver': ['sgd', 'adam'],
+        'mlpclassifier__alpha': [0.0001, 0.05],
+        'mlpclassifier__learning_rate': ['constant','adaptive'],
+        'mlpclassifier__max_iter': [300,500,450,200,300]
+    }
+
+    random_search = RandomizedSearchCV(rf_pipeline, param_distributions=param_dist, n_jobs=-1, n_iter=50, cv=5, scoring = 'f1_micro')
+    random_search.fit(X_train, y_train)
+    best_model_pipe = random_search.best_estimator_
+
+    joblib_file = out_dir + "best_Model.pkl"  
+    joblib.dump(best_model_pipe, joblib_file)
+
+
+if __name__ == "__main__":
+  main(opt["--in_file_1"], opt["--out_dir"])