tanu/ML_AI_training
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity

added exmaples and practice run for imbalanced data sets

Browse source
This commit is contained in:
Tanushree Tunstall 2022-03-14 18:43:29 +00:00
parent 1016430ae0
commit 29306e77ee
2 changed files with 217 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

112
imbalance_p1.py Normal file
View file

@ -0,0 +1,112 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Mar 14 11:42:14 2022
@author: tanu
"""
#%%
#Link: https://www.justintodata.com/imbalanced-data-machine-learning-classification/
# Seems good:https://learn-scikit.oneoffcoder.com/imbalanced-learn.html
#https://medium.com/analytics-vidhya/what-is-an-imbalanced-data-how-to-handle-imbalanced-data-in-python-e6067792950f
# VV GOOD and knowledge: https://www.analyticsvidhya.com/blog/2017/03/imbalanced-data-classification/
#Oversampling: which adds examples of the minority class to balance the dataset
# since in our case it is DM muts that are 'minority'
# Undersampling: We will downsize the majority class to balance with the minority class.
# not good idea coz loses data
from imblearn.over_sampling import RandomOverSampler
from imblearn.over_sampling import SMOTE
#%%
my_df['mutation_info_labels'].value_counts()
# IMP
#dm_om_map = {'DM': 0, 'OM': 1} # for minority oversampling pncA, code 1 for what is minority
dm_om_map = {'DM': 1, 'OM': 0} # for all other genes
my_df['mutation_info_labels'] = my_df['mutation_info_labels'].map(dm_om_map)
my_df['mutation_info_labels'].value_counts()
df = my_df[x_stabilityN + ['mutation_info_labels']]
df.columns
#%% Data preparation
# 1) Simple random oversampling: the basic approach of random sampling with replacement from the minority class.
# 2) Oversampling with shrinkage: based on random sampling, adding some noise/shrinkage to disperse the new samples.
# 3) Oversampling using SMOTE: synthesize new samples based on the minority class.
#%% 1)Data preparation
df_train, df_test = train_test_split(df
, test_size = 0.2
, stratify = df['mutation_info_labels']
, random_state = 888)
features = df_train.drop(columns=['mutation_info_labels']).columns
df['mutation_info_labels'].value_counts(normalize=True)
df['mutation_info_labels'].value_counts().plot(kind='bar')
df_train['mutation_info_labels'].value_counts()
df_test['mutation_info_labels'].value_counts()
#%% 1) Simple Random Oversampling
msk = df_train['mutation_info_labels'] == 1 # series with name and T/F
msk.value_counts()
num_to_oversample = len(df_train) - 2*msk.sum()
print('\nNo to oversample:', num_to_oversample)
df_positive_oversample = df_train[msk].sample(n = num_to_oversample
, replace = True
, random_state = 888)
df_train_oversample = pd.concat([df_train, df_positive_oversample])
df_train_oversample['mutation_info_labels'].value_counts()
df_train_oversample['mutation_info_labels'].value_counts().plot(kind = 'bar')
# apply model
clf = LogisticRegression(random_state=888)
clf.fit(df_train_oversample[features], df_train_oversample['mutation_info_labels'])
y_pred = clf.predict_proba(df_test[features])[:, 1]
roc_auc_score(df_test['mutation_info_labels'], y_pred)
y_pred1 = clf.predict(df_test[features])
f1_score(df_test['mutation_info_labels'], y_pred1)
accuracy_score(df_test['mutation_info_labels'], y_pred1)
precision_score(df_test['mutation_info_labels'], y_pred1)
recall_score(df_test['mutation_info_labels'], y_pred1)
roc_auc_score(df_test['mutation_info_labels'], y_pred1)
roc_curve(df_test['mutation_info_labels'], y_pred1) # tuple
matthews_corrcoef(df_test['mutation_info_labels'], y_pred1)
print(classification_report(df_test['mutation_info_labels'], y_pred1))
#%%imbalanced-learn
# FIXME: INSTALL
ros = RandomOverSampler(random_state=888)
X_resampled, y_resampled = ros.fit_resample(df_train[features], df_train['mutation_info_labels'])
y_resampled.value_counts()
# Oversampling with shrinkage
# shrinkage can take values greater than 0
# When shrinkage=0, it will be the same as simple random sampling.
# The larger the shrinkage value, the more noise we add, so the more dispersed
# the new samples will be. This is useful when we don’t always want to repeat the samples.
ros = RandomOverSampler(random_state = 888, shrinkage = 0.1)
X_resampled, y_resampled = ros.fit_resample(df_train[features], df_train['mutation_info_labels'])
clf.fit(X_resampled, y_resampled)
y_pred = clf.predict_proba(df_test[features])[:, 1]
roc_auc_score(df_test['mutation_info_labels'], y_pred)
#%%
smote = SMOTE(random_state=888)
X_resampled, y_resampled = smote.fit_resample(df_train[features]
, df_train['mutation_info_labels'])
clf.fit(X_resampled, y_resampled)
y_pred = clf.predict_proba(df_test[features])[:, 1]
roc_auc_score(df_test['mutation_info_labels'], y_pred)
y_pred1 = clf.predict(df_test[features])
matthews_corrcoef(df_test['mutation_info_labels'], y_pred1)
print(classification_report(df_test['mutation_info_labels'], y_pred1))

105
imbalance_p2.py Normal file
View file

@ -0,0 +1,105 @@
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Mar 14 11:42:14 2022
@author: tanu
"""
#%%
#Link: https://www.justintodata.com/imbalanced-data-machine-learning-classification/
# Seems good:https://learn-scikit.oneoffcoder.com/imbalanced-learn.html
#https://medium.com/analytics-vidhya/what-is-an-imbalanced-data-how-to-handle-imbalanced-data-in-python-e6067792950f
# VV GOOD and knowledge: https://www.analyticsvidhya.com/blog/2017/03/imbalanced-data-classification/
# plots basic
#https://data36.com/plot-histogram-python-pandas/
#Oversampling: which adds examples of the minority class to balance the dataset
# since in our case it is DM muts that are 'minority'
# Undersampling: We will downsize the majority class to balance with the minority class.
# not good idea coz loses data
#%% numerical data only
num_df_wtgt['mutation_class'].value_counts()
ax = num_df_wtgt['mutation_class'].value_counts().plot(kind = 'bar')
#_ = ax.set_title('Frequency of Classes, Imbalanced')
#Split the Features (X) and Target (Y)
Y = num_df_wtgt['mutation_class'].values
X = num_df_wtgt.drop('mutation_class',axis = 1)
#%% No Imbalance Handling: model
# https://towardsdatascience.com/imbalanced-classification-in-python-smote-enn-method-db5db06b8d50
model_ori = AdaBoostClassifier()
# Define evaluation procedure (here we use Repeated Stratified K-Fold CV)
cv_ori = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=1)
# Evaluate model
scoring = ['accuracy','precision_macro','recall_macro']
scores_ori = cross_validate(model_ori, X, Y, scoring=scoring, cv=cv_ori, n_jobs=-1)
# summarize performance
# print('Mean Accuracy: %.4f' % np.mean(scores_ori['test_accuracy']))
# print('Mean Precision: %.4f' % np.mean(scores_ori['test_precision_macro']))
# print('Mean Recall: %.4f' % np.mean(scores_ori['test_recall_macro']))
#%% Balance Handling: Using SMOTE-ENN: model
model = AdaBoostClassifier()
#model = LogisticRegression()
# Define SMOTE-ENN
resample = SMOTEENN(enn=EditedNearestNeighbours(sampling_strategy='all'))
# Define pipeline
pipeline = Pipeline(steps=[('r', resample), ('m', model)])
# Define evaluation procedure (here we use Repeated Stratified K-Fold CV)
cv = RepeatedStratifiedKFold(n_splits=2, n_repeats=3, random_state=1)
# Evaluate model
scoring=['accuracy','precision_macro','recall_macro']
scores = cross_validate(pipeline, X, Y, scoring=scoring, cv=cv, n_jobs=-1)
# summarize performance
# print('Mean Accuracy: %.4f' % np.mean(scores['test_accuracy']))
# print('Mean Precision: %.4f' % np.mean(scores['test_precision_macro']))
# print('Mean Recall: %.4f' % np.mean(scores['test_recall_macro']))
#%%
print('\nOriginal data performance:')
print('Mean Accuracy: %.4f' % np.mean(scores_ori['test_accuracy']))
print('Mean Precision: %.4f' % np.mean(scores_ori['test_precision_macro']))
print('Mean Recall: %.4f' % np.mean(scores_ori['test_recall_macro']))
print('\nBalance handling data performance:')
print('Mean Accuracy: %.4f' % np.mean(scores['test_accuracy']))
print('Mean Precision: %.4f' % np.mean(scores['test_precision_macro']))
print('Mean Recall: %.4f' % np.mean(scores['test_recall_macro']))
#%%
# pnca: works (not needed strictly)
# alr: fails
# embb: works
# gid: fails
# katG: works
# rpoB: works (but quite similar)
#%%
#https://towardsdatascience.com/dealing-with-imbalanced-dataset-642a5f6ee297
y = num_df_wtgt['mutation_class']
X_train, X_test, y_train, y_test = train_test_split(num_df, y
, test_size = 0.33
, random_state = 2
, shuffle = True
, stratify = y)
sm = SMOTE(random_state = 42)
X_train_new, y_train_new = sm.fit_resample(X_train, y_train)
y_train_df = y_train.to_frame()
y_train_df.value_counts().plot(kind = 'bar')
y_train_new_df = y_train_new.to_frame()
y_train_new_df.value_counts().plot(kind = 'bar')