# -*- coding: utf-8 -*- # ---------------------------------------------------------------------------- # Links # Titanic: Machine Learning from Disaster # https://www.kaggle.com/c/titanic # Kaggle Python Tutorial on Machine Learning # https://www.datacamp.com/community/open-courses/kaggle-python-tutorial-on-machine-learning#gs.7vaQ7NY # Auto-scaling scikit-learn with Apache Spark # https://databricks.com/blog/2016/02/08/auto-scaling-scikit-learn-with-apache-spark.html # ---------------------------------------------------------------------------- # Import libraries import numpy as np import pandas as pd #import subprocess as sp import sklearn as skl import sklearn.ensemble as skle import sklearn.model_selection as sklms # ---------------------------------------------------------------------------- # Load data def LoadData(): # print(sp.check_output(['ls', '../input']).decode('utf8')) train = pd.read_csv('../input/train.csv') test = pd.read_csv('../input/test.csv') print('\nThe source data has been loaded') print('Train DS shape: ', str(train.shape)) print('Test DS shape: ' + str(test.shape)) return (train, test) # ---------------------------------------------------------------------------- # Show data """ def ShowData(data): print(data.head()) return # ---------------------------------------------------------------------------- # Review data def ReviewData(data): d = data.describe() print('\nColumns:') print("All: " + str(data.columns.get_values())) print("Numeric: " + str(d.columns.get_values())) print("Non-null: ") print(data.count()) print('\nDescribe:') print(d) return # ---------------------------------------------------------------------------- # Print column values def PrintColumnValues(data, columns): data = data[columns] for c in data.columns: print('\nColumn ' + c) print(data[c].value_counts()) return # ---------------------------------------------------------------------------- # Preprocess data def PreprocessData(data): data = data.copy() # Sex data['Sex_'] = 0 data.set_value(data['Sex'] == 'male', 'Sex_', 1) data.set_value(data['Sex'] == 'female', 'Sex_', -1) # Embarked data['Embarked_S'] = 0 data['Embarked_C'] = 0 data['Embarked_Q'] = 0 data.set_value(data['Embarked'] == 'S', 'Embarked_S', 1) data.set_value(data['Embarked'] == 'C', 'Embarked_C', 1) data.set_value(data['Embarked'] == 'Q', 'Embarked_Q', 1) # Age data['Age_'] = data['Age'] data.set_value(data['Age_'].isnull(), 'Age_', 28.0) # = train['Age'].median() # Family data['Family_'] = data['SibSp'] + data['Parch'] # Fare data['Fare_'] = data['Fare'] data.set_value(data['Fare_'].isnull(), 'Fare_', 14.4542) # = train['Fare'].median() # Drop preprocessed columns data.drop(['Sex', 'Embarked', 'Age', 'SibSp', 'Parch', 'Fare'], axis=1, inplace=True) # Drop unused columns data.drop(['Name', 'Ticket', 'Cabin'], axis=1, inplace=True) return data # ---------------------------------------------------------------------------- # Split train data def SplitTrainData(data, train_fraction, val_fraction): # Normalize fractions fraction_sum = train_fraction + val_fraction train_fraction /= fraction_sum val_fraction /= fraction_sum # Reorder data n = len(data) ix = np.random.permutation(n) data = data.iloc[ix] # Split data cut_ix = int(n * train_fraction) train_data = data[:cut_ix] val_data = data[cut_ix:] return (train_data, val_data) # ---------------------------------------------------------------------------- # Exctract features def ExtractFeatures(data): data_ix = data['PassengerId'].values if 'Survived' in data.columns: data_X = data.drop(['Survived', 'PassengerId'], axis=1).values else: data_X = data.drop(['PassengerId'], axis=1).values return (data_ix, data_X) # ---------------------------------------------------------------------------- # Exctract features and labels def ExtractFeaturesAndLabels(data): (data_ix, data_X) = ExtractFeatures(data) data_y = data['Survived'].values return (data_ix, data_X, data_y) # ---------------------------------------------------------------------------- # Create a classifier def CreateClassifierByType(classifier_type): classifier = None if classifier_type == 'Decision Tree': classifier = skl.tree.DecisionTreeClassifier(max_depth=10, min_samples_split=5) elif classifier_type == 'Random Forest': classifier = skle.RandomForestClassifier(max_depth=10, min_samples_split=5, n_estimators=100) elif classifier_type == 'Random Forest Grid Search': param_grid = { 'max_depth': [3, 5, 10, None], 'max_features': [1, 3, 5, None], 'min_samples_split': [2, 3, 5, 10], 'min_samples_leaf': [1, 3, 5, 10], 'bootstrap': [True, False], 'criterion': ['gini', 'entropy'], 'n_estimators': [10, 20, 40, 80] } # classifier = skl.grid_search.GridSearchCV(skle.RandomForestClassifier(), # param_grid=param_grid) classifier = sklms.GridSearchCV(skle.RandomForestClassifier(), param_grid=param_grid) return classifier # ---------------------------------------------------------------------------- # Fit and validate def FitAndValidate(pretrain_X, pretrain_y, val_X, val_y, classifier_type): classifier = CreateClassifierByType(classifier_type) classifier.fit(pretrain_X, pretrain_y) val_score = classifier.score(val_X, val_y) return val_score # ---------------------------------------------------------------------------- # Validate classifiers def ValidateClassifiers(pretrain_X, pretrain_y, val_X, val_y): # classifier_types = ['Decision Tree', 'Random Forest', # 'Random Forest Grid Search'] classifier_types = ['Decision Tree', 'Random Forest'] score_best = -1 ct_best = 'None' print('\nValidate classifiers') for ct in classifier_types: val_score = FitAndValidate(pretrain_X, pretrain_y, val_X, val_y, ct) print('Score of ' + ct + ': ' + str(val_score)) if val_score > score_best: score_best = val_score ct_best = ct return ct_best # ---------------------------------------------------------------------------- # Fit and predict def CreateAndFit(train_X, train_y, classifier_type): print('\nFinal learning begins') classifier = CreateClassifierByType(classifier_type) classifier.fit(train_X, train_y) print('Final learning ends') return classifier; # ---------------------------------------------------------------------------- # Main function if __name__ == '__main__': # 1. Load and show data (train, test) = LoadData() # ShowData(train) # 2. Review data before preprocessing # ReviewData(train) # PrintColumnValues(train, ['Sex', 'Cabin', 'Embarked']) # 3. Preprocess data and review it again train_dat = PreprocessData(train) # ReviewData(train_dat) # PrintColumnValues(train_dat, # ['Sex_', 'Embarked_S', 'Embarked_C', 'Embarked_Q']) # 4. Split train data (pretrain_dat, val_dat) = SplitTrainData(train_dat, 80, 20) print('\nSplit of train dataset') print('Pretrain DS shape: ' + str(pretrain_dat.shape)) print('Validate DS shape: ' + str(val_dat.shape)) # 5. Extract features and labels for validation (pretrain_ix, pretrain_X, pretrain_y) = ExtractFeaturesAndLabels(pretrain_dat) (val_ix, val_X, val_y) = ExtractFeaturesAndLabels(val_dat) # 6.Validate classifiers ctype_best = ValidateClassifiers(pretrain_X, pretrain_y, val_X, val_y) print('\nThe best classifier is ' + ctype_best) # 7. Final learning (train_ix, train_X, train_y) = ExtractFeaturesAndLabels(train_dat) cl_best = CreateAndFit(train_X, train_y, ctype_best) # 8. Final predicting # ReviewData(test) test_dat = PreprocessData(test) ReviewData(test_dat) (test_ix, test_X) = ExtractFeatures(test_dat) test_predictions = cl_best.predict(test_X) # 9. Create and save the final prediction file prediction_ds = pd.DataFrame(index=test_ix, data=test_predictions, columns=['Survived']) print('\nFinal prediction distribution:') print(prediction_ds['Survived'].value_counts()) prediction_ds.to_csv('MyFirstPenTest.csv', index_label='PassengerId') """ # ----------------------------------------------------------------------------