training_model2.py

import numpy as np 
import pandas as pd 
import os
import cv2 as cv
import matplotlib.pyplot as plt
import seaborn as sns


all_0 = "./C-NMC_Leukemia/training_data/fold_0/all"
all_1 = "./C-NMC_Leukemia/training_data/fold_1/all"
all_2 = "./C-NMC_Leukemia/training_data/fold_2/all"

hem_0 = "./C-NMC_Leukemia/training_data/fold_0/hem"
hem_1 = "./C-NMC_Leukemia/training_data/fold_1/hem"
hem_2 = "./C-NMC_Leukemia/training_data/fold_2/hem"      

def get_path_image(folder):
    image_paths = []
    image_fnames = os.listdir(folder) 
    for img_id in range(len(image_fnames)):
        img = os.path.join(folder,image_fnames[img_id])
        image_paths.append(img)
    
    return image_paths

img_data = []

for i in [all_0,all_1,all_2,hem_0,hem_1,hem_2]:
    paths = get_path_image(i)
    img_data.extend(paths)
print(len(img_data))


data = {"img_data":img_data,
        "labels":[np.nan for x in range(len(img_data))]}

data = pd.DataFrame(data) 

data["labels"][0:7272] = 1 # ALL
data["labels"][7272:10661] = 0 # HEM

data["labels"] = data["labels"].astype("int64")

image = cv.imread(data["img_data"][1000])
plt.imshow(image)
plt.title("Sample image before cropping")
plt.show()

img_list = []
for i in range(len(img_data)):
    image = cv.imread(data["img_data"][i])
    gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
    thresh = cv.threshold(gray, 0, 255, cv.THRESH_BINARY_INV + cv.THRESH_OTSU)[1]

    result = cv.bitwise_and(image, image, mask=thresh)
    result[thresh==0] = [255,255,255] 
    (x, y, z_) = np.where(result > 0)
    mnx = (np.min(x))
    mxx = (np.max(x))
    mny = (np.min(y))
    mxy = (np.max(y))
    crop_img = image[mnx:mxx,mny:mxy,:]
    crop_img_r = cv.resize(crop_img, (224,224))
    img_list.append(crop_img_r)

plt.imshow(img_list[1000])
plt.title("Sample image after cropping")
plt.show()

from tensorflow.keras.applications import ResNet50, ResNet101
from keras.applications.vgg19 import VGG19
from tensorflow.keras.preprocessing import image
from tensorflow.keras.models import Model
from tensorflow.keras.applications.resnet50 import preprocess_input

def feature_extract(model):
    if model == "VGG19": model = VGG19(weights='imagenet',include_top=False, pooling="avg")
    elif model == "ResNet50": model = ResNet50(weights='imagenet',include_top=False,pooling="avg")
    elif model == "ResNet101": model = ResNet101(weights='imagenet',include_top=False,pooling="avg")
    return model

model = feature_extract("ResNet50") # or "VGG19", "ResNet101"

features_list = []
for i in range(len(img_list)):
            
    image = img_list[i].reshape(-1, 224, 224, 3)
    image = preprocess_input(image)
    
    """
    # Reshaping when VGG19 model is selected
    features = model.predict(image).reshape(512,)
    """

    #Reshaping  when ResNet50 or ResNet101 model is selected
    features = model.predict(image).reshape(2048,)
    
    features_list.append(features)

features_df = pd.DataFrame(features_list)

features_df["labels"] = data["labels"]

x = features_df.drop(['labels'], axis = 1)
y = features_df.loc[:,"labels"].values

print(f"Number of features before feature selection: {x.shape[1]}")

from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
scaler.fit(x)
x_ = scaler.transform(x)

x_ = pd.DataFrame(x_)


from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_classif

def anova_fs():
    
    selector = SelectKBest(f_classif, k=500) # k is number of features
    selector.fit(x_, y)

    cols = selector.get_support(indices=True)
    anova_x = x_[cols]
    return anova_x

from sklearn.feature_selection import RFE
from sklearn.ensemble import RandomForestClassifier

def RFE_fs():
    rfe_selector = RFE(estimator=RandomForestClassifier())
    rfe_selector.fit(x_, y)
    
    rfe_support = rfe_selector.get_support()
    rfe_feature = x_.loc[:,rfe_support].columns.tolist()
    
    rfe_x = x_[rfe_feature]
    return rfe_x

from sklearn.feature_selection import SelectFromModel
from sklearn.ensemble import RandomForestClassifier

def rf_fs():
    embeded_rf_selector = SelectFromModel(RandomForestClassifier(n_estimators=200, random_state=5), threshold='1.25*median')
    embeded_rf_selector.fit(x, y)
    
    embeded_rf_support = embeded_rf_selector.get_support()
    embeded_rf_feature = x.loc[:,embeded_rf_support].columns.tolist()
    
    rf_x = x[embeded_rf_feature]
    return rf_x

fs_x = rf_fs() # feature selection methods "anova_fs", "RFE_fs"

print(f"Number of features after feature selection: {fs_x.shape[1]}")


from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(fs_x, y, test_size = 0.2, random_state = 42)

from sklearn.model_selection import cross_val_score,cross_val_predict
from sklearn.ensemble import RandomForestClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.neighbors import KNeighborsClassifier
from sklearn import svm
from sklearn.metrics import confusion_matrix
from sklearn.metrics import f1_score,precision_score,recall_score,accuracy_score
from sklearn.model_selection import GridSearchCV

neig = np.arange(1, 25)
train_accuracy = []
test_accuracy = []

for i, k in enumerate(neig):

    knn = KNeighborsClassifier(n_neighbors=k)
    knn.fit(x_train,y_train)
    prediction_ = knn.predict(x_test)
    train_accuracy.append(knn.score(x_train, y_train))
    test_accuracy.append(knn.score(x_test, y_test))
    
print("Best accuracy is {} with K = {}".format(np.max(test_accuracy),1+test_accuracy.index(np.max(test_accuracy))))

knn = KNeighborsClassifier(n_neighbors=17)
knn.fit(x_train,y_train)
predicted = knn.predict(x_test)
score = knn.score(x_test, y_test)
knn_score_ = np.mean(score)

print('Accuracy : %.3f' % (knn_score_))

p=precision_score(y_test, predicted) 
print('Precision : %.3f' % (p))

r=recall_score(y_test, predicted) 
print('Recall : %.3f' % (r))

f1=f1_score(y_test, predicted) 
print('F1-score: %.3f' % (f1))

f1_w=f1_score(y_test, predicted, average='weighted') 
print('Weighted f1-score: %.3f' % (f1_w))

cf_matrix = confusion_matrix(y_test, predicted)
sns.heatmap(cf_matrix, cmap="PuBu", annot=True, fmt='.0f')
plt.show()


#SVM   
param_grid_svm = {'C': [0.1, 1, 10, 100, 1000],
              'gamma': [1, 0.1, 0.01, 0.001, 0.0001],
              'kernel': ['rbf', 'poly']}
 
SVM_grid = GridSearchCV(svm.SVC(), param_grid_svm, cv=5)
SVM_grid.fit(x_train, y_train)

print(SVM_grid.best_params_)
 
print(SVM_grid.best_estimator_)

svm_clf = svm.SVC(C=100, gamma=0.01, kernel='rbf')
svm_clf.fit(x_train,y_train)
predicted = svm_clf.predict(x_test)
score = svm_clf.score(x_test, y_test)
svm_score_ = np.mean(score)

print('Accuracy : %.3f' % (svm_score_))

p=precision_score(y_test, predicted) 
print('precision : %.3f' % (p))

r=recall_score(y_test, predicted) 
print('recall : %.3f' % (r))

f1=f1_score(y_test, predicted) 
print('f1-score: %.3f' % (f1))

f1_w=f1_score(y_test, predicted, average='weighted') 
print('weighted f1-score: %.3f' % (f1_w))

cf_matrix = confusion_matrix(y_test, predicted)
sns.heatmap(cf_matrix, cmap="PuBu", annot=True, fmt='.0f')
plt.show()


#Random Forest: 
param_grid_rf = { 
    'n_estimators': [200, 500],
    'max_depth' : [4,5,6,7,8]}

RF_grid = GridSearchCV(estimator=RandomForestClassifier(), param_grid=param_grid_rf, cv= 5)
RF_grid.fit(x_train, y_train)

print(RF_grid.best_params_)

r_forest = RandomForestClassifier(500,max_depth=8, random_state=5)
r_forest.fit(x_train,y_train)
predicted = r_forest.predict(x_test)
score = r_forest.score(x_test, y_test)
rf_score_ = np.mean(score)

print('Accuracy : %.3f' % (rf_score_))

p=precision_score(y_test, predicted) 
print('precision : %.3f' % (p))

r=recall_score(y_test, predicted) 
print('recall : %.3f' % (r))

f1=f1_score(y_test, predicted) 
print('f1-score: %.3f' % (f1))

f1_w=f1_score(y_test, predicted, average='weighted') 
print('weighted f1-score: %.3f' % (f1_w))


cf_matrix = confusion_matrix(y_test, predicted)
sns.heatmap(cf_matrix, cmap="PuBu", annot=True, fmt='.0f')
plt.show()

#Naive Bayes:

nb_model = GaussianNB()
nb_model.fit(x_train,y_train)
predicted = nb_model.predict(x_test)
score = nb_model.score(x_test, y_test)
nb_score_ = np.mean(score)

print('Accuracy : %.3f' % (nb_score_))

p=precision_score(y_test, predicted) 
print('precision : %.3f' % (p))

r=recall_score(y_test, predicted) 
print('recall : %.3f' % (r))

f1=f1_score(y_test, predicted) 
print('f1-score: %.3f' % (f1))

f1_w=f1_score(y_test, predicted, average='weighted') 
print('weighted f1-score: %.3f' % (f1_w))

cf_matrix = confusion_matrix(y_test, predicted)
sns.heatmap(cf_matrix, cmap="PuBu", annot=True, fmt='.0f')
plt.show()