def train_support_vector_machine(self,
X_train,
X_test,
y_train,
y_test,
save_mdl=False,
save_loc=Weights_File,
max_epochs=17000,
cost_thresh=0.001,
l_r=0.000001,
r_s=10000):
# set global parameters
global Learning_rate
global Reg_strength
Learning_rate = l_r
Reg_strength = r_s
# normalize the data
x_train_cpy = Utils.normalize_numpy_array(X_train,
self.__min_max_scalar)
x_test_cpy = Utils.normalize_numpy_array(X_test, self.__min_max_scalar)
# print(type(x_train_cpy)) # debug
# creating and appending col of 1's to train and test data (this col is the intercept)
intercept_train = np.ones(x_train_cpy.shape[0])
intercept_test = np.ones(x_test_cpy.shape[0])
x_train_cpy = np.hstack(
(x_train_cpy, np.atleast_2d(intercept_train).T))
x_test_cpy = np.hstack((x_test_cpy, np.atleast_2d(intercept_test).T))
# setting attributes
self.__x_train = x_train_cpy
self.__x_test = x_test_cpy
self.__y_train = y_train
self.__y_test = y_test
# using stochastic gradient to find the optimal weights for linear classifier
print("Training Support Vector Machine") # debug
self.__trained_weights = stochastic_gradient_descent(
self.__x_train,
self.__y_train,
max_epochs=max_epochs,
cost_thresh=cost_thresh)
print("Trained Weights: ", self.__trained_weights)
print("Finished Training SupportVectorMachine")
self.__test_support_vector_machine()
if save_mdl:
self.__save_trained_weights(save_loc)
def __test_support_vector_machine(self):
if self.__trained_weights is not None and self.__x_test is not None and self.__y_test is not None:
print("Testing SupportVectorMachine")
y_test_predicted = np.array([])
# for i in range(self.x_test.shape[0]):
for r in self.__x_test:
y_predict = np.sign(np.dot(r, self.__trained_weights))
y_test_predicted = np.append(y_test_predicted, y_predict)
accuracy_score = Utils.calculate_accuracy_score(
self.__y_test, y_test_predicted)
print("accuracy of the model: {}".format(accuracy_score))
confusion_mat = Utils.calculate_confusion_matrix(
self.__y_test, y_test_predicted)
print("SVM confusion Matrix")
print(confusion_mat[0])
print(confusion_mat[1])
# write accuracy and confusion matrix to file
try:
report_fl_name = "C:/Users/Brijesh Prajapati/Documents/Projects/Autism_Detection_Hons_Proj/Classifiers/" \
"Classifier_Reports/svm_current_report.json"
classifier_desc = "SVM l_r: " + str(
Learning_rate) + ", r_s: " + str(Reg_strength)
dict_report = {
"desc": classifier_desc,
"accuracy_score": str(accuracy_score),
"tp": str(confusion_mat[0][0]),
"fn": str(confusion_mat[0][1]),
"fp": str(confusion_mat[1][0]),
"tn": str(confusion_mat[1][1]),
"specificity": str(confusion_mat[2][0]),
"sensitivity": str(confusion_mat[2][1])
}
with open(report_fl_name, 'w') as report_fl:
report_fl.write(json.dumps(dict_report))
except Exception as ex:
print("Exception occurred saving SVM report", ex)
# print("SK Learn Metrics")
# print(confusion_matrix(self.y_test, y_test_predicted))
# print(classification_report(self.y_test, y_test_predicted))
else:
print("Train SVM before Testing.")
def predict_sample(self, sample_data, return_lbl=False):
try:
if self.__trained_weights is not None:
# min_max_scalar = Utils.calculate_min_max_scalar(pd.read_csv(self.training_data_dir))
# sample_data_copy = sample_data.copy(deep=True)
sample_data_copy = copy.deepcopy(sample_data)
# sample_data_copy = Utils.normalize_dataset(sample_data_copy, self.__min_max_scalar)
sample_data_copy = Utils.normalize_numpy_array(
sample_data_copy, self.__min_max_scalar)
# sample_data_copy.insert(loc=len(sample_data_copy.columns), column='intercept', value=1)
intercept_train = np.ones(sample_data_copy.shape[0])
sample_data_copy = np.hstack(
(sample_data_copy, np.atleast_2d(intercept_train).T))
y_predict = np.sign(
np.dot(sample_data_copy, self.__trained_weights))
if return_lbl is True:
if y_predict > 0:
return 'ASD'
else:
return 'Normal'
return y_predict
else:
print("Train the SVM first OR load the trained weights")
except Exception as e:
print("Error occurred predicting the sample in SVM.")
print('Exception', e)
traceback.print_exc()
def trained_decision_tree(max_depth, x, y):
# Create decision tree
# print("Called trained_decision_tree") # debug
dcsn_tree = DecisionTree(max_depth=max_depth)
x_sample, y_sample = Utils.bootstrap_samples(x, y) # bootstrapping samples
dcsn_tree.train_decision_tree(x_sample, y_sample)
return dcsn_tree
def train_classifiers(self, svm_param_dict=None, rf_param_dict=None, mlp_param_dict=None,
adab_param_dict=None):
self.__x_train, self.__x_test, self.__y_train, self.__y_test = prepare_data_for_training(Training_Data_Dir)
save_mdl = True
# ******************************************************************************Setting Up SVM for training
x_temp, y_temp = Utils.bootstrap_samples(self.__x_train, self.__y_train) # bootstrap samples
if svm_param_dict is not None:
self.svm.train_support_vector_machine(x_temp, self.__x_test, y_temp, self.__y_test, save_mdl=save_mdl,
**svm_param_dict)
else:
self.svm.train_support_vector_machine(x_temp, self.__x_test, y_temp, self.__y_test)
# ******************************************************************************Setting Up RF for training
x_temp, y_temp = Utils.bootstrap_samples(self.__x_train, self.__y_train) # bootstrap samples
if rf_param_dict is not None:
self.random_forest.train_random_forest(x_temp, self.__x_test, y_temp, self.__y_test, save_mdl=save_mdl,
**rf_param_dict)
else:
self.random_forest.train_random_forest(x_temp, self.__x_test, y_temp, self.__y_test, save_mdl=save_mdl)
# ******************************************************************************Setting Up MLP for training
x_temp, y_temp = Utils.bootstrap_samples(self.__x_train, self.__y_train)
if mlp_param_dict is not None:
self.multi_layer_perceptron.train_MultiLayerPerceptron(x_temp, self.__x_test, y_temp, self.__y_test,
save_mdl=save_mdl, **mlp_param_dict)
else:
self.multi_layer_perceptron.train_MultiLayerPerceptron(x_temp, self.__x_test, y_temp, self.__y_test,
save_mdl=save_mdl)
# ******************************************************************************Setting Up Adaboost for training
x_temp, y_temp = Utils.bootstrap_samples(self.__x_train, self.__y_train)
if adab_param_dict is not None:
self.adaboost.train_adaboost(x_temp, self.__x_test, y_temp, self.__y_test, save_classifiers=save_mdl,
**adab_param_dict)
else:
self.adaboost.train_adaboost(x_temp, self.__x_test, y_temp, self.__y_test, save_classifiers=save_mdl)
# self.__models_loaded = True
# self.test_main_classifier()
return 0
def __init__(self, training_data_dir):
self.training_data_dir = training_data_dir
self.__trained_weights = None
self.__x_train = None
self.__x_test = None
self.__y_train = None
self.__y_test = None
# finding the min max scalar
self.__min_max_scalar = Utils.calculate_min_max_np_scalar(
pd.read_csv(training_data_dir))
def train_MultiLayerPerceptron(self,
X_train,
X_test,
y_train,
y_test,
save_mdl=False,
save_loc=MultilayerPerceptronMdlFl,
hidden_layer_sizes=16,
max_iter=400):
x_train_cpy = Utils.normalize_numpy_array(X_train,
self.__min_max_scalar)
x_test_cpy = Utils.normalize_numpy_array(X_test, self.__min_max_scalar)
self.__x_train = x_train_cpy
self.__x_test = x_test_cpy
self.__y_train = y_train
self.__y_test = y_test
self.__max_iter = max_iter
self.__hidden_layer_size = hidden_layer_sizes
print("Training MultiLayerPerceptron...")
t0 = time.time() # test purposes
self.__mlp_classifier = MLPClassifier(
hidden_layer_sizes=self.__hidden_layer_size,
activation='relu',
max_iter=self.__max_iter,
solver='adam')
self.__mlp_classifier = self.__mlp_classifier.fit(
self.__x_train, self.__y_train)
t1 = time.time() # test purposes
print("Finished Training MultiLayerPerceptron in: ", t1 - t0)
self.__test_MultiLayerPerceptron()
if save_mdl:
self.__save_MultiLayerPerceptron(save_loc)
def prepare_data_for_training(file_dir):
dataframe = pd.read_csv(file_dir)
# replacing the labels
dataframe['feature_class']. \
replace({'ASD': 1.0, 'TD': -1.0}, inplace=True)
# splitting data into train and test
x_train, x_test, y_train, y_test = Utils.train_test_split(dataframe=dataframe, test_size=0.2)
# convert to dataframe to numpy array and return
return x_train.to_numpy(), x_test.to_numpy(), y_train.to_numpy(), y_test.to_numpy()
def __test_adaboost(self):
print("Testing AdaBoost")
y_predictions = np.ones(len(self.__x_test))
for test, i in zip(self.__x_test, range(len(self.__x_test))):
test = np.reshape(test, (-1, len(test)))
y_predictions[i] = self.predict_sample_adaboost(test)
accuracy_score = Utils.calculate_accuracy_score(
self.__y_test, y_predictions)
print("Accuracy: ", accuracy_score)
cf = Utils.calculate_confusion_matrix(self.__y_test, y_predictions)
print("Adaboost Matrix")
print(cf[0])
print(cf[1])
# write accuracy and confusion matrix to file
try:
report_fl_name = "C:/Users/Brijesh Prajapati/Documents/Projects/Autism_Detection_Hons_Proj/Classifiers/" \
"Classifier_Reports/adaboost_current_report.json"
classifier_desc = "Adaboost number of stumps: " + str(
self.__num_classifiers)
dict_report = {
"desc": classifier_desc,
"accuracy_score": str(accuracy_score),
"tp": str(cf[0][0]),
"fn": str(cf[0][1]),
"fp": str(cf[1][0]),
"tn": str(cf[1][1]),
"specificity": str(cf[2][0]),
"sensitivity": str(cf[2][1])
}
with open(report_fl_name, 'w') as report_fl:
report_fl.write(json.dumps(dict_report))
except Exception as ex:
print("Exception occurred saving adaboost report", ex)
def __init__(self, training_data_dir):
self.__max_iter = 400
self.__hidden_layer_size = 16
self.training_data_dir = training_data_dir
self.__mlp_classifier = MLPClassifier()
self.__x_train = None
self.__x_test = None
self.__y_train = None
self.__y_test = None
# calculating the min max scalar for normalization
self.__min_max_scalar = Utils.calculate_min_max_np_scalar(
pd.read_csv(training_data_dir))
def process_training_data(fl_dir, min_max_scalar=None):
df = pd.read_csv(fl_dir)
# replace labels
df['feature_class'].replace({'ASD': 1.0, 'TD': -1.0}, inplace=True)
x_train, x_test, y_train, y_test = Utils.train_test_split(df, 0.2)
# if min_max_scalar is not None:
# x_train = Utils.normalize_dataset(x_train, min_max_scalar)
# x_test = Utils.normalize_dataset(x_test, min_max_scalar)
return x_train, x_test, y_train, y_test
def test_main_classifier(self):
self.__load_classifiers()
y_pred = []
for test_data in self.__x_test:
svm_p = self.svm.predict_sample([test_data])
rf_p = self.random_forest.predict([test_data])
mlp_p = self.multi_layer_perceptron.predict(test_data)
adb_p = self.adaboost.predict_sample_adaboost(np.array([test_data]))
predictions = [float(svm_p), rf_p[0], mlp_p[0], adb_p[0]]
y_pred.append(most_common_label(predictions))
accuracy_score = Utils.calculate_accuracy_score(self.__y_test, y_pred)
print("Accuracy of main model: ", accuracy_score)
confusion_mat = Utils.calculate_confusion_matrix(self.__y_test, y_pred)
print("Main Classifier Confusion Matrix")
print(confusion_mat[0])
print(confusion_mat[1])
# write accuracy and confusion matrix to file
try:
report_fl_name = "C:/Users/Brijesh Prajapati/Documents/Projects/Autism_Detection_Hons_Proj/Classifiers/" \
"Classifier_Reports/mainmodel_current_report.json"
classifier_desc = "Main Model"
dict_report = {"desc": classifier_desc, "accuracy_score": str(accuracy_score),
"tp": str(confusion_mat[0][0]), "fn": str(confusion_mat[0][1]),
"fp": str(confusion_mat[1][0]), "tn": str(confusion_mat[1][1]),
"specificity": str(confusion_mat[2][0]), "sensitivity": str(confusion_mat[2][1])}
with open(report_fl_name, 'w') as report_fl:
report_fl.write(json.dumps(dict_report))
except Exception as ex:
print("Exception occurred saving Main Model report", ex)
def __test_MultiLayerPerceptron(self):
print("Testing Multi Layer Perceptron")
mlp_predictions = self.__mlp_classifier.predict(self.__x_test)
accuracy_score = Utils.calculate_accuracy_score(
self.__y_test, mlp_predictions)
print("Accuracy: ", accuracy_score)
cf = Utils.calculate_confusion_matrix(self.__y_test, mlp_predictions)
print("Multi Layer Perceptron Confusion Matrix")
print(cf[0])
print(cf[1])
print(classification_report(self.__y_test, mlp_predictions))
# write accuracy and confusion matrix to file
try:
report_fl_name = "C:/Users/Brijesh Prajapati/Documents/Projects/Autism_Detection_Hons_Proj/Classifiers/" \
"Classifier_Reports/mlp_current_report.json"
classifier_desc = "MLP hidden_layer_size: " + str(self.__hidden_layer_size) + ", max_iteration: " + \
str(self.__max_iter)
dict_report = {
"desc": classifier_desc,
"accuracy_score": str(accuracy_score),
"tp": str(cf[0][0]),
"fn": str(cf[0][1]),
"fp": str(cf[1][0]),
"tn": str(cf[1][1]),
"specificity": str(cf[2][0]),
"sensitivity": str(cf[2][1])
}
with open(report_fl_name, 'w') as report_fl:
report_fl.write(json.dumps(dict_report))
except Exception as ex:
print("Exception occurred saving MLP report", ex)
def __test_random_forest(self):
print("Testing Random Forest Classifier")
y_predictions = self.predict(self.__x_test)
accuracy_score = Utils.calculate_accuracy_score(
self.__y_test, y_predictions)
print("Accuracy: ", accuracy_score)
cf = Utils.calculate_confusion_matrix(self.__y_test, y_predictions)
print("Random Forest Confusion Matrix")
print(cf[0])
print(cf[1])
# write accuracy and confusion matrix to file
try:
report_fl_name = "C:/Users/Brijesh Prajapati/Documents/Projects/Autism_Detection_Hons_Proj/Classifiers/" \
"Classifier_Reports/rf_current_report.json"
classifier_desc = "RF number of trees: " + str(self.__num_tree) + ", max depth: " + \
str(self.__max_depth)
dict_report = {
"desc": classifier_desc,
"accuracy_score": str(accuracy_score),
"tp": str(cf[0][0]),
"fn": str(cf[0][1]),
"fp": str(cf[1][0]),
"tn": str(cf[1][1]),
"specificity": str(cf[2][0]),
"sensitivity": str(cf[2][1])
}
with open(report_fl_name, 'w') as report_fl:
report_fl.write(json.dumps(dict_report))
except Exception as ex:
print("Exception occurred saving random forest report", ex)
def predict(self, sample_data, return_lbl=False):
# normalize the sample data
# sample_data = Utils.normalize_dataset(sample_data, self.__min_max_scalar)
sample_data = Utils.normalize_numpy_array(sample_data,
self.__min_max_scalar)
prediction = self.__mlp_classifier.predict([sample_data])
if return_lbl is True:
if prediction > 0:
return 'ASD'
else:
return 'Normal'
else:
return prediction
def process_training_data(fl_dir):
df = pd.read_csv(fl_dir)
# finding min max scalar
# min_max_scalar = Utils.calculate_min_max_scalar(pd.read_csv(fl_dir))
# replace labels
df['feature_class'].replace({'ASD': 1.0, 'TD': -1.0}, inplace=True)
X_train, X_test, y_train, y_test = Utils.train_test_split(df, 0.2)
# X_train = Utils.normalize_dataset(X_train, min_max_scalar)
# X_test = Utils.normalize_dataset(X_test, min_max_scalar)
X_train = X_train.to_numpy()
X_test = X_test.to_numpy()
y_train = y_train.to_numpy()
y_test = y_test.to_numpy()
return X_train, X_test, y_train, y_test
def process_training_data(fl_dir, min_max_scalar):
dataframe = pd.read_csv(fl_dir)
# calculates the min max value of each col in the dataframe
# min_max_scalar = Utils.calculate_min_max_scalar(dataset=dataframe)
# replacing the labels
dataframe['feature_class']. \
replace({'ASD': 1.0, 'TD': -1.0}, inplace=True)
# splitting data into train and test
x_train, x_test, y_train, y_test = Utils.train_test_split(
dataframe=dataframe, test_size=0.2)
# normalizing train and test data
# x_train = Utils.normalize_dataset(df=x_train, min_max=min_max_scalar)
# x_test = Utils.normalize_dataset(df=x_test, min_max=min_max_scalar)
#
# # insert intercept col 'b' in (W * Xi + b)
# x_train.insert(loc=len(x_train.columns), column='intercept', value=1)
# x_test.insert(loc=len(x_test.columns), column='intercept', value=1)
return x_train.to_numpy(), x_test.to_numpy(), y_train.to_numpy(
), y_test.to_numpy()
def is_leaf_node(self):
# if self.value is not None:
# return True
# else:
# return False
return self.value is not None
if __name__ == "__main__":
df = pd.read_csv("D:/TrainingDataset_YEAR_PROJECT/TrainingSet.csv")
# replace labels
df['feature_class'].replace({'ASD': 1.0, 'TD': -1.0}, inplace=True)
X_train, X_test, y_train, y_test = Utils.train_test_split(df, 0.2)
X_train = X_train.values
X_test = X_test.values
y_train = y_train.values
y_test = y_test.values
decision_tree = DecisionTree(max_depth=10)
decision_tree.train_decision_tree(X_train, y_train)
y_pred = decision_tree.predict(X_test)
accuracy_score = Utils.calculate_accuracy_score(y_test, y_pred)
print("Accuracy: ", accuracy_score)
# References:
def train_svm_model(fl_dir):
# X_train, X_test, Y_train, Y_test = SupportVectorMachine.proc_CSV_data(fl_dir)
dataframe = pd.read_csv(fl_dir)
dataframe["feature_class"]. \
replace({"ASD": 1, "TD": 0},
inplace=True)
dataframe = dataframe.sample(frac=1)
# DATA NORMALIZATION
# min_max_scalar = Utils.calculate_min_max_scalar(dataframe)
# dataframe = Utils.normalize_dataset(dataframe, min_max_scalar)
#
# X = dataframe.drop(labels="feature_class", axis=1)
# Y = dataframe['feature_class']
#
# # feature selection BEGIN
# print("Shape before feature selection", X.shape)
#
# # L1-based feature selection
# lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(X, Y)
# model = SelectFromModel(lsvc, prefit=True)
# X = model.transform
#
# # Univariate feature selection
# X = SelectKBest(chi2, k=6).fit_transform(X, Y)
# Tree-based feature selection
# clf = ExtraTreesClassifier(n_estimators=50)
# clf = clf.fit(X, Y)
# clf.feature_importances_
# model = SelectFromModel(clf, prefit=True)
# X = model.transform(X)
# print("Shape after feature seleciton", X.shape)
# feature selection END
# train test split
# X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.20)
X_train, X_test, Y_train, Y_test = Utils.train_test_split(
dataframe=dataframe, test_size=0.20)
# SVM
# scalar = MinMaxScaler()
# X_train = pd.DataFrame(scalar.fit_transform(X_train.values))
# X_test = pd.DataFrame(scalar.transform(X_test.values))
#
# ns_probs = [0 for _ in range(len(Y_test))]
# svm_model_linear = SVC(kernel='rbf', probability=True).fit(X_train, Y_train) # polynomial kernel
#
# # load the saved model
# # load_model = joblib.load(filename=saved_mdl_path)
#
# svm_prediction = svm_model_linear.predict(X_test)
#
# accuracy = svm_model_linear.score(X_test, Y_test)
# print(accuracy) # debug
#
# # creating a confusion matrix
# cm = confusion_matrix(Y_test, svm_prediction)
#
# print(cm)
# print(classification_report(Y_test, svm_prediction))
#
# # predict probabilities
# lr_probs = svm_model_linear.predict_proba(X_test)
# # keep probabilities for the positive outcome only
# lr_probs = lr_probs[:, 1]
# # calculate scores
# ns_auc = roc_auc_score(Y_test, ns_probs)
# lr_auc = roc_auc_score(Y_test, lr_probs)
# # summarize scores
# print('No Skill: ROC AUC=%.3f' % (ns_auc))
# print('Logistic: ROC AUC=%.3f' % (lr_auc))
# # calculate roc curves
# ns_fpr, ns_tpr, _ = roc_curve(Y_test, ns_probs)
# lr_fpr, lr_tpr, _ = roc_curve(Y_test, lr_probs)
# # plot the roc curve for the model
# pyplot.plot(ns_fpr, ns_tpr, linestyle='--', label='No Skill')
# pyplot.plot(lr_fpr, lr_tpr, marker='.', label='Logistic')
# # axis labels
# pyplot.xlabel('False Positive Rate')
# pyplot.ylabel('True Positive Rate')
#
# # save the model
# # saved_mdl_path = 'normal_leaf_model.sav'
# # joblib.dump(svm_model_linear, saved_mdl_path)
#
# # show the legend
# pyplot.legend()
# # show the plot
# pyplot.show()
# RANDOM FOREST CLASSIFIER
# scalar = MinMaxScaler()
# X_train = pd.DataFrame(scalar.fit_transform(X_train.values))
# X_test = pd.DataFrame(scalar.transform(X_test.values))
#
# ns_probs = [0 for _ in range(len(Y_test))]
# regressor = RandomForestClassifier(n_estimators=18, max_depth=10).fit(X_train, Y_train)
# y_pred = regressor.predict(X_test)
#
# print(confusion_matrix(Y_test, y_pred))
# print(classification_report(Y_test, y_pred))
# # print(Y_test)
# # print(y_pred)
# print(SupportVectorMachine.calc_accuracy_score(Y_test, y_pred))
#
# accuracy = regressor.score(X_test, Y_test)
# print(accuracy) # debug
#
# print(accuracy_score(Y_test, y_pred.round(), normalize=False))
# NN..............................................................
# mlp = MLPClassifier(hidden_layer_sizes=9, activation='relu', max_iter=400, solver='adam').fit(X_train, Y_train)
#
# predictions = mlp.predict(X_test)
# print(confusion_matrix(Y_test, predictions))
# print(classification_report(Y_test, predictions))
# print(Utils.calculate_accuracy_score(Y_test, predictions))
# KNN
# scalar = MinMaxScaler()
# X_train = pd.DataFrame(scalar.fit_transform(X_train.values))
# X_test = pd.DataFrame(scalar.transform(X_test.values))
model = KNeighborsClassifier(n_neighbors=2).fit(X_train, Y_train)
y_pred = model.predict(X_test)
#
accuracy = model.score(X_test, Y_test)
print(accuracy) # debug
print(confusion_matrix(Y_test, y_pred))
print(classification_report(Y_test, y_pred))
accuracy = model.score(X_test, Y_test)
print(accuracy) # debug