From fba1481c08cc23b7b4d0e51887ba9f13e2467c66 Mon Sep 17 00:00:00 2001 From: Tanushree Tunstall Date: Fri, 24 Jun 2022 13:25:00 +0100 Subject: [PATCH] added MultClfs.py that contains my ML functions --- scripts/ml/MultClfs.py | 531 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 531 insertions(+) create mode 100644 scripts/ml/MultClfs.py diff --git a/scripts/ml/MultClfs.py b/scripts/ml/MultClfs.py new file mode 100644 index 0000000..7e04e71 --- /dev/null +++ b/scripts/ml/MultClfs.py @@ -0,0 +1,531 @@ +#!/usr/bin/env python3 +# -*- coding: utf-8 -*- +""" +Created on Fri Mar 4 15:25:33 2022 + +@author: tanu +""" +#%% +import os, sys +import pandas as pd +import numpy as np +import pprint as pp +from copy import deepcopy +from sklearn import linear_model +from sklearn import datasets +from collections import Counter + +from sklearn.linear_model import LogisticRegression, LogisticRegressionCV +from sklearn.linear_model import RidgeClassifier, RidgeClassifierCV, SGDClassifier, PassiveAggressiveClassifier + +from sklearn.naive_bayes import BernoulliNB +from sklearn.neighbors import KNeighborsClassifier +from sklearn.svm import SVC +from sklearn.tree import DecisionTreeClassifier, ExtraTreeClassifier +from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier, AdaBoostClassifier, GradientBoostingClassifier, BaggingClassifier +from sklearn.naive_bayes import GaussianNB +from sklearn.gaussian_process import GaussianProcessClassifier, kernels +from sklearn.gaussian_process.kernels import RBF, DotProduct, Matern, RationalQuadratic, WhiteKernel + +from sklearn.discriminant_analysis import LinearDiscriminantAnalysis, QuadraticDiscriminantAnalysis +from sklearn.neural_network import MLPClassifier + +from sklearn.svm import SVC +from xgboost import XGBClassifier +from sklearn.naive_bayes import MultinomialNB +from sklearn.preprocessing import StandardScaler, MinMaxScaler, OneHotEncoder + +from sklearn.compose import ColumnTransformer +from sklearn.compose import make_column_transformer + +from sklearn.metrics import make_scorer, confusion_matrix, accuracy_score, balanced_accuracy_score, precision_score, average_precision_score, recall_score +from sklearn.metrics import roc_auc_score, roc_curve, f1_score, matthews_corrcoef, jaccard_score, classification_report + +# added +from sklearn.model_selection import train_test_split, cross_validate, cross_val_score, LeaveOneOut, KFold, RepeatedKFold, cross_val_predict + +from sklearn.model_selection import train_test_split, cross_validate, cross_val_score +from sklearn.model_selection import StratifiedKFold,RepeatedStratifiedKFold, RepeatedKFold + +from sklearn.pipeline import Pipeline, make_pipeline + +from sklearn.feature_selection import RFE, RFECV + +import itertools +import seaborn as sns +import matplotlib.pyplot as plt + +from statistics import mean, stdev, median, mode + +from imblearn.over_sampling import RandomOverSampler +from imblearn.under_sampling import RandomUnderSampler +from imblearn.over_sampling import SMOTE +from sklearn.datasets import make_classification +from imblearn.combine import SMOTEENN +from imblearn.combine import SMOTETomek + +from imblearn.over_sampling import SMOTENC +from imblearn.under_sampling import EditedNearestNeighbours +from imblearn.under_sampling import RepeatedEditedNearestNeighbours + +from sklearn.model_selection import GridSearchCV +from sklearn.base import BaseEstimator +from sklearn.impute import KNNImputer as KNN +import json +import argparse +import re +#%% GLOBALS +rs = {'random_state': 42} +njobs = {'n_jobs': 10} + +scoring_fn = ({ 'mcc' : make_scorer(matthews_corrcoef) + , 'fscore' : make_scorer(f1_score) + , 'precision' : make_scorer(precision_score) + , 'recall' : make_scorer(recall_score) + , 'accuracy' : make_scorer(accuracy_score) + , 'roc_auc' : make_scorer(roc_auc_score) + , 'jcc' : make_scorer(jaccard_score) + }) + +skf_cv = StratifiedKFold(n_splits = 10 + #, shuffle = False, random_state= None) + , shuffle = True,**rs) + +rskf_cv = RepeatedStratifiedKFold(n_splits = 10 + , n_repeats = 3 + , **rs) + +mcc_score_fn = {'mcc': make_scorer(matthews_corrcoef)} +jacc_score_fn = {'jcc': make_scorer(jaccard_score)} + +############################################################################### +score_type_ordermapD = { 'mcc' : 1 + , 'fscore' : 2 + , 'jcc' : 3 + , 'precision' : 4 + , 'recall' : 5 + , 'accuracy' : 6 + , 'roc_auc' : 7 + , 'TN' : 8 + , 'FP' : 9 + , 'FN' : 10 + , 'TP' : 11 + , 'trainingY_neg': 12 + , 'trainingY_pos': 13 + , 'blindY_neg' : 14 + , 'blindY_pos' : 15 + , 'fit_time' : 16 + , 'score_time' : 17 + } + +scoreCV_mapD = {'test_mcc' : 'MCC' + , 'test_fscore' : 'F1' + , 'test_precision' : 'Precision' + , 'test_recall' : 'Recall' + , 'test_accuracy' : 'Accuracy' + , 'test_roc_auc' : 'ROC_AUC' + , 'test_jcc' : 'JCC' + } + +scoreBT_mapD = {'bts_mcc' : 'MCC' + , 'bts_fscore' : 'F1' + , 'bts_precision' : 'Precision' + , 'bts_recall' : 'Recall' + , 'bts_accuracy' : 'Accuracy' + , 'bts_roc_auc' : 'ROC_AUC' + , 'bts_jcc' : 'JCC' + } + +#%%############################################################################ +############################ +# MultModelsCl() +# Run Multiple Classifiers +############################ +# Multiple Classification - Model Pipeline +def MultModelsCl(input_df, target, skf_cv + , blind_test_df + , blind_test_target + , tts_split_type + , resampling_type = 'none' # default + , add_cm = True # adds confusion matrix based on cross_val_predict + , add_yn = True # adds target var class numbers + , var_type = ['numerical', 'categorical','mixed'] + , return_formatted_output = True): + + ''' + @ param input_df: input features + @ type: df with input features WITHOUT the target variable + + @param target: target (or output) feature + @type: df or np.array or Series + + @param skv_cv: stratifiedK fold int or object to allow shuffle and random state to pass + @type: int or StratifiedKfold() + + @var_type: numerical, categorical and mixed to determine what col_transform to apply (MinMaxScalar and/or one-ho t encoder) + @type: list + + returns + Dict containing multiple classification scores for each model and mean of each Stratified Kfold including training + ''' + + #====================================================== + # Determine categorical and numerical features + #====================================================== + numerical_ix = input_df.select_dtypes(include=['int64', 'float64']).columns + numerical_ix + categorical_ix = input_df.select_dtypes(include=['object', 'bool']).columns + categorical_ix + + #====================================================== + # Determine preprocessing steps ~ var_type + #====================================================== + if var_type == 'numerical': + t = [('num', MinMaxScaler(), numerical_ix)] + + if var_type == 'categorical': + t = [('cat', OneHotEncoder(), categorical_ix)] + + if var_type == 'mixed': + t = [('num', MinMaxScaler(), numerical_ix) + , ('cat', OneHotEncoder(), categorical_ix) ] + + col_transform = ColumnTransformer(transformers = t + , remainder='passthrough') + + #====================================================== + # Specify multiple Classification Models + #====================================================== + models = [('AdaBoost Classifier' , AdaBoostClassifier(**rs) ) + # , ('Bagging Classifier' , BaggingClassifier(**rs, **njobs, bootstrap = True, oob_score = True) ) + # , ('Decision Tree' , DecisionTreeClassifier(**rs) ) + # , ('Extra Tree' , ExtraTreeClassifier(**rs) ) + # , ('Extra Trees' , ExtraTreesClassifier(**rs) ) + # , ('Gradient Boosting' , GradientBoostingClassifier(**rs) ) + # , ('Gaussian NB' , GaussianNB() ) + # , ('Gaussian Process' , GaussianProcessClassifier(**rs) ) + # , ('K-Nearest Neighbors' , KNeighborsClassifier() ) + # , ('LDA' , LinearDiscriminantAnalysis() ) + , ('Logistic Regression' , LogisticRegression(**rs) ) + # , ('Logistic RegressionCV' , LogisticRegressionCV(cv = 3, **rs)) + # , ('MLP' , MLPClassifier(max_iter = 500, **rs) ) + # , ('Multinomial' , MultinomialNB() ) + # , ('Naive Bayes' , BernoulliNB() ) + # , ('Passive Aggresive' , PassiveAggressiveClassifier(**rs, **njobs) ) + # , ('QDA' , QuadraticDiscriminantAnalysis() ) + # , ('Random Forest' , RandomForestClassifier(**rs, n_estimators = 1000 ) ) + # , ('Random Forest2' , RandomForestClassifier(min_samples_leaf = 5 + # , n_estimators = 1000 + # , bootstrap = True + # , oob_score = True + # , **njobs + # , **rs + # , max_features = 'auto') ) + # , ('Ridge Classifier' , RidgeClassifier(**rs) ) + # , ('Ridge ClassifierCV' , RidgeClassifierCV(cv = 3) ) + # , ('SVC' , SVC(**rs) ) + # , ('Stochastic GDescent' , SGDClassifier(**rs, **njobs) ) + # , ('XGBoost' , XGBClassifier(**rs, verbosity = 0, use_label_encoder =False) ) + ] + + mm_skf_scoresD = {} + + print('\n==============================================================\n' + , '\nRunning several classification models (n):', len(models) + ,'\nList of models:') + for m in models: + print(m) + print('\n================================================================\n') + + index = 1 + for model_name, model_fn in models: + print('\nRunning classifier:', index + , '\nModel_name:' , model_name + , '\nModel func:' , model_fn) + index = index+1 + + model_pipeline = Pipeline([ + ('prep' , col_transform) + , ('model' , model_fn)]) + + print('\nRunning model pipeline:', model_pipeline) + skf_cv_modD = cross_validate(model_pipeline + , input_df + , target + , cv = skf_cv + , scoring = scoring_fn + , return_train_score = True) + + ####################################################################### + #====================================================== + # Option: Add confusion matrix from cross_val_predict + # Understand and USE with caution + # cross_val_score, cross_val_predict, "Passing these predictions into an evaluation metric may not be a valid way to measure generalization performance. Results can differ from cross_validate and cross_val_score unless all tests sets have equal size and the metric decomposes over samples." + # https://stackoverflow.com/questions/65645125/producing-a-confusion-matrix-with-cross-validate + #====================================================== + if add_cm: + + #----------------------------------------------------------- + # Initialise dict of Confusion Matrix (cm) + #----------------------------------------------------------- + cmD = {} + + # Calculate cm + y_pred = cross_val_predict(model_pipeline, input_df, target, cv = skf_cv, **njobs) + #_tn, _fp, _fn, _tp = confusion_matrix(y_pred, y).ravel() # internally + tn, fp, fn, tp = confusion_matrix(y_pred, target).ravel() + + # Build dict + + cmD = {'TN' : tn + , 'FP': fp + , 'FN': fn + , 'TP': tp} + #--------------------------------- + # Update cv dict with cmD and tbtD + #---------------------------------- + skf_cv_modD.update(cmD) + else: + skf_cv_modD = skf_cv_modD + ####################################################################### + #============================================= + # Option: Add targety numbers for data + #============================================= + if add_yn: + + #----------------------------------------------------------- + # Initialise dict of target numbers: training and blind (tbt) + #----------------------------------------------------------- + tbtD = {} + + # training y + tyn = Counter(target) + tyn_neg = tyn[0] + tyn_pos = tyn[1] + + # blind test y + btyn = Counter(blind_test_target) + btyn_neg = btyn[0] + btyn_pos = btyn[1] + + # Build dict + tbtD = {'n_trainingY_neg' : tyn_neg + , 'n_trainingY_pos' : tyn_pos + , 'n_blindY_neg' : btyn_neg + , 'n_blindY_pos' : btyn_pos} + + #--------------------------------- + # Update cv dict with cmD and tbtD + #---------------------------------- + skf_cv_modD.update(tbtD) + else: + skf_cv_modD = skf_cv_modD + + ####################################################################### + #============================== + # Extract mean values for CV + #============================== + mm_skf_scoresD[model_name] = {} + + for key, value in skf_cv_modD.items(): + print('\nkey:', key, '\nvalue:', value) + print('\nmean value:', np.mean(value)) + mm_skf_scoresD[model_name][key] = round(np.mean(value),2) + + #return(mm_skf_scoresD) +#%% + #========================= + # Blind test: BTS results + #========================= + # Build the final results with all scores for the model + #bts_predict = gscv_fs.predict(blind_test_df) + model_pipeline.fit(input_df, target) + bts_predict = model_pipeline.predict(blind_test_df) + + bts_mcc_score = round(matthews_corrcoef(blind_test_target, bts_predict),2) + print('\nMCC on Blind test:' , bts_mcc_score) + print('\nAccuracy on Blind test:', round(accuracy_score(blind_test_target, bts_predict),2)) + + # Diff b/w train and bts test scores + # train_test_diff_MCC = cvtrain_mcc - bts_mcc_score + # print('\nDiff b/w train and blind test score (MCC):', train_test_diff) + + mm_skf_scoresD[model_name]['bts_mcc'] = bts_mcc_score + mm_skf_scoresD[model_name]['bts_fscore'] = round(f1_score(blind_test_target, bts_predict),2) + mm_skf_scoresD[model_name]['bts_precision'] = round(precision_score(blind_test_target, bts_predict),2) + mm_skf_scoresD[model_name]['bts_recall'] = round(recall_score(blind_test_target, bts_predict),2) + mm_skf_scoresD[model_name]['bts_accuracy'] = round(accuracy_score(blind_test_target, bts_predict),2) + mm_skf_scoresD[model_name]['bts_roc_auc'] = round(roc_auc_score(blind_test_target, bts_predict),2) + mm_skf_scoresD[model_name]['bts_jcc'] = round(jaccard_score(blind_test_target, bts_predict),2) + #mm_skf_scoresD[model_name]['diff_mcc'] = train_test_diff_MCC + + #return(mm_skf_scoresD) +#%% + # ADD more info: meta data related to input and blind and resampling + + # target numbers: training + yc1 = Counter(target) + yc1_ratio = yc1[0]/yc1[1] + + # target numbers: test + yc2 = Counter(blind_test_target) + yc2_ratio = yc2[0]/yc2[1] + + mm_skf_scoresD[model_name]['resampling'] = resampling_type + + mm_skf_scoresD[model_name]['n_training_size'] = len(input_df) + mm_skf_scoresD[model_name]['n_trainingY_ratio'] = round(yc1_ratio, 2) + + mm_skf_scoresD[model_name]['n_test_size'] = len(blind_test_df) + mm_skf_scoresD[model_name]['n_testY_ratio'] = round(yc2_ratio,2) + mm_skf_scoresD[model_name]['n_features'] = len(input_df.columns) + mm_skf_scoresD[model_name]['tts_split'] = tts_split_type + + #return(mm_skf_scoresD) + #============================ + # Process the dict to have WF + #============================ + if return_formatted_output: + CV_BT_metaDF = ProcessMultModelsCl(mm_skf_scoresD) + return(CV_BT_metaDF) + else: + return(mm_skf_scoresD) + +#%% Process output function ################################################### +############################ +# ProcessMultModelsCl() +############################ +#Processes the dict from above if use_formatted_output = True + +def ProcessMultModelsCl(inputD = {}): + + scoresDF = pd.DataFrame(inputD) + + #------------------------ + # Extracting split_name + #----------------------- + tts_split_nameL = [] + for k,v in inputD.items(): + tts_split_nameL = tts_split_nameL + [v['tts_split']] + + if len(set(tts_split_nameL)) == 1: + tts_split_name = str(list(set(tts_split_nameL))[0]) + print('\nExtracting tts_split_name:', tts_split_name) + + #------------------------ + # WF: only CV and BTS + #----------------------- + scoresDFT = scoresDF.T + + scoresDF_CV = scoresDFT.filter(regex='^test_.*$', axis = 1); scoresDF_CV.columns + # map colnames for consistency to allow concatenting + scoresDF_CV.columns = scoresDF_CV.columns.map(scoreCV_mapD); scoresDF_CV.columns + scoresDF_CV['source_data'] = 'CV' + + scoresDF_BT = scoresDFT.filter(regex='^bts_.*$', axis = 1); scoresDF_BT.columns + # map colnames for consistency to allow concatenting + scoresDF_BT.columns = scoresDF_BT.columns.map(scoreBT_mapD); scoresDF_BT.columns + scoresDF_BT['source_data'] = 'BT' + + # dfs_combine_wf = [baseline_BT, smnc_BT, ros_BT, rus_BT, rouC_BT, + # baseline_CV, smnc_CV, ros_CV, rus_CV, rouC_CV] + + #baseline_all = baseline_all_scores.filter(regex = 'bts_.*|test_.*|.*_time|TN|FP|FN|TP|.*_neg|.*_pos', axis = 0) + + #metaDF = scoresDFT.filter(regex='training_size|blind_test_size|_time|TN|FP|FN|TP|.*_neg|.*_pos|resampling', axis = 1); scoresDF_BT.columns + #metaDF = scoresDFT.filter(regex='n_.*$|_time|TN|FP|FN|TP|.*_neg|.*_pos|resampling|tts.*', axis = 1); metaDF.columns + metaDF = scoresDFT.filter(regex='^(?!test_.*$|bts_.*$|train_.*$).*'); metaDF.columns + + print('\nTotal cols in each df:' + , '\nCV df:', len(scoresDF_CV.columns) + , '\nBT_df:', len(scoresDF_BT.columns) + , '\nmetaDF:', len(metaDF.columns)) + if len(scoresDF_CV.columns) == len(scoresDF_BT.columns): + print('\nFirst proceeding to rowbind CV and BT dfs:') + expected_ncols_out = len(scoresDF_BT.columns) + len(metaDF.columns) + print('\nFinal output should have:',expected_ncols_out, 'columns' ) + + #----------------- + # Combine WF + #----------------- + dfs_combine_wf = [scoresDF_CV, scoresDF_BT] + + print('\nCombinig', len(dfs_combine_wf), 'using pd.concat by row ~ rowbind' + , '\nChecking Dims of df to combine:' + , '\nDim of CV:', scoresDF_CV.shape + , '\nDim of BT:', scoresDF_BT.shape) + #print(scoresDF_CV) + #print(scoresDF_BT) + + dfs_nrows_wf = [] + for df in dfs_combine_wf: + dfs_nrows_wf = dfs_nrows_wf + [len(df)] + dfs_nrows_wf = max(dfs_nrows_wf) + + dfs_ncols_wf = [] + for df in dfs_combine_wf: + dfs_ncols_wf = dfs_ncols_wf + [len(df.columns)] + dfs_ncols_wf = max(dfs_ncols_wf) + print(dfs_ncols_wf) + + expected_nrows_wf = len(dfs_combine_wf) * dfs_nrows_wf + expected_ncols_wf = dfs_ncols_wf + + common_cols_wf = list(set.intersection(*(set(df.columns) for df in dfs_combine_wf))) + print('\nNumber of Common columns:', dfs_ncols_wf + , '\nThese are:', common_cols_wf) + + if len(common_cols_wf) == dfs_ncols_wf : + combined_baseline_wf = pd.concat([df[common_cols_wf] for df in dfs_combine_wf], ignore_index=False) + print('\nConcatenating dfs with different resampling methods [WF]:' + , '\nSplit type:', tts_split_name + , '\nNo. of dfs combining:', len(dfs_combine_wf)) + #print('\n================================================^^^^^^^^^^^^') + if len(combined_baseline_wf) == expected_nrows_wf and len(combined_baseline_wf.columns) == expected_ncols_wf: + #print('\n================================================^^^^^^^^^^^^') + + print('\nPASS:', len(dfs_combine_wf), 'dfs successfully combined' + , '\nnrows in combined_df_wf:', len(combined_baseline_wf) + , '\nncols in combined_df_wf:', len(combined_baseline_wf.columns)) + else: + print('\nFAIL: concatenating failed' + , '\nExpected nrows:', expected_nrows_wf + , '\nGot:', len(combined_baseline_wf) + , '\nExpected ncols:', expected_ncols_wf + , '\nGot:', len(combined_baseline_wf.columns)) + sys.exit('\nFIRST IF FAILS') + else: + print('\nConcatenting dfs not possible [WF],check numbers ') + + + #------------------------------------- + # Combine WF+Metadata: Final output + #------------------------------------- + # checking indices for the dfs to combine: + c1 = list(set(combined_baseline_wf.index)) + c2 = list(metaDF.index) + + if c1 == c2: + print('\nPASS: proceeding to merge metadata with CV and BT dfs') + combDF = pd.merge(combined_baseline_wf, metaDF, left_index = True, right_index = True) + else: + sys.exit('\nFAIL: Could not merge metadata with CV and BT dfs') + + if len(combDF.columns) == expected_ncols_out: + print('\nPASS: Combined df has expected ncols') + else: + sys.exit('\nFAIL: Length mismatch for combined_df') + + print('\n=========================================================' + , '\nSUCCESS: Ran multiple classifiers' + , '\n=======================================================') + + #resampling_methods_wf = combined_baseline_wf[['resampling']] + #resampling_methods_wf = resampling_methods_wf.drop_duplicates() + #, '\n', resampling_methods_wf) + + return combDF + +############################################################################### + + \ No newline at end of file