def run_single_patient(self):
predictor_result = []
for iteration in range(self.cross_validation):
if iteration % 10 is 0:
print(str(iteration) + " ...")
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(
self.data, break_amount=1, split_percentage=0)
clf = MLPClassifier(solver='lbfgs',
alpha=1e-3,
hidden_layer_sizes=(15, 5),
random_state=1)
clf.fit(train_data, train_labels)
pred_test_probabilities = clf.predict_proba(self.patient_list)
iteration_result = []
for row in pred_test_probabilities:
if row[0] > row[1]:
iteration_result.append(["NEG", row[0]])
else:
iteration_result.append(["POS", row[1]])
predictor_result.append(iteration_result)
return predictor_result
def run_cross_validation(self):
pos_predictor_result = []
neg_predictor_result = []
for iteration in range(self.cross_validation):
if iteration % 10 is 0:
print(str(iteration) + " ...")
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(
self.data, break_amount=1.2, split_percentage=20)
clf = MLPClassifier(solver='lbfgs',
alpha=1e-3,
hidden_layer_sizes=(len(self.data.ab_names),
2),
random_state=1)
clf.fit(train_data, train_labels)
pos_pred_test = clf.predict(pos_test_data)
neg_pred_test = clf.predict(neg_test_data)
pos_predictor_result.append(
len([1 for x in pos_pred_test if x == "POS"]) /
float(len(pos_pred_test)))
neg_predictor_result.append(
len([1 for x in neg_pred_test if x == "NEG"]) /
float(len(neg_pred_test)))
return pos_predictor_result, neg_predictor_result
def run_single_patient(self):
predictor_result = []
for iteration in range(self.cross_validation):
if iteration%20 is 0:
print(str(iteration) + " ...")
# Create training set (split percentage is 0)
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(self.pos_training_data,
self.neg_training_data,
break_amount=1,
split_percentage=0)
# Train a perceptron classifier
clf = MLPClassifier(solver='lbfgs', alpha=1e-3, hidden_layer_sizes=(self.layers[0], self.layers[1]), random_state=0)
clf.fit(train_data, train_labels)
# Output prediction probabilities
pred_test_probabilities = clf.predict_proba(self.patient_list)
# Append labels
iteration_result = []
for row in pred_test_probabilities:
if row[0] > row[1]:
iteration_result.append(["NEG", row[0]])
else:
iteration_result.append(["POS", row[1]])
predictor_result.append(iteration_result)
# Process results, sums over all suscebtible and certainty values per patient
patient_pos_probability = []
patient_certainty = []
for patient_index in range(len(self.patient_list)):
pos_count = 0
certainty_sum = 0
for iteration_result in predictor_result:
if iteration_result[patient_index][0] == "POS":
pos_count+=1
certainty_sum+=iteration_result[patient_index][1]
patient_pos_probability.append(pos_count/float(len(predictor_result)))
patient_certainty.append(certainty_sum/float(len(predictor_result)))
return patient_pos_probability, patient_certainty
def run_single_patient(self):
predictor_result = []
for iteration in range(self.cross_validation):
if iteration%2000 is 0:
print(str(iteration) + " ...")
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(self.pos_training_data,
self.neg_training_data,
break_amount=1,
split_percentage=0)
clf = tree.DecisionTreeClassifier()
clf.fit(train_data, train_labels)
pred_test_probabilities = clf.predict_proba(self.patient_list)
iteration_result = []
for row in pred_test_probabilities:
if row[0] > row[1]:
iteration_result.append(["NEG", row[0]])
else:
iteration_result.append(["POS", row[1]])
predictor_result.append(iteration_result)
# Process results
patient_pos_probability = []
patient_certainty = []
for patient_index in range(len(self.patient_list)):
pos_count = 0
certainty_sum = 0
for iteration_result in predictor_result:
if iteration_result[patient_index][0] == "POS":
pos_count+=1
certainty_sum+=iteration_result[patient_index][1]
patient_pos_probability.append(pos_count/float(len(predictor_result)))
patient_certainty.append(certainty_sum/float(len(predictor_result)))
return patient_pos_probability, patient_certainty
def run_cross_validation(self):
pos_predictor_result = []
neg_predictor_result = []
importance = np.zeros(len(self.data.relevant_ab_names))
for iteration in range(self.cross_validation):
if iteration%2000 is 0:
print(str(iteration) + " ...")
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(self.pos_training_data,
self.neg_training_data,
break_amount=1,
split_percentage=10)
clf = tree.DecisionTreeClassifier()
clf.fit(train_data, train_labels)
importance+=clf.feature_importances_
pos_pred_test = clf.predict(pos_test_data)
neg_pred_test = clf.predict(neg_test_data)
pos_predictor_result.append(len([1 for x in pos_pred_test if x == "POS"])/float(len(pos_pred_test)))
neg_predictor_result.append(len([1 for x in neg_pred_test if x == "NEG"])/float(len(neg_pred_test)))
print("")
results = []
for i in range(len(self.data.relevant_ab_names)):
results.append([importance[i]/self.cross_validation, self.data.relevant_ab_names[i]])
results.sort(reverse=True)
print(self.data.relevant_ab_names)
for i in range(len(self.data.relevant_ab_names)):
print("{}: {}".format(results[i][1], results[i][0]))
print("")
return pos_predictor_result, neg_predictor_result
def run_cross_validation(self):
pos_predictor_result = []
neg_predictor_result = []
self.iteration_timer_estimate = 0
importance = np.zeros(len(self.data.relevant_ab_names))
for iteration in range(self.cross_validation):
# Start of timer
iteration_start = time()
if iteration%50 is 0:
print(str(iteration) + " ...")
print("Time left: {}".format((self.iteration_timer_estimate/(iteration+1))*(self.cross_validation-iteration)))
# Split dataset into train and test
train_data, pos_test_data, neg_test_data, train_labels = process_data.generate_training_data(self.pos_training_data,
self.neg_training_data,
break_amount=1,
split_percentage=10)
# Train percpetron on training set
clf = MLPClassifier(solver='lbfgs', alpha=1e-3, hidden_layer_sizes=(self.layers[0], self.layers[1]), random_state=0)
clf.fit(train_data, train_labels)
# Test the classification on a test and control set
pos_pred_test = clf.predict(pos_test_data)
neg_pred_test = clf.predict(neg_test_data)
# Calculate accuracy and store it
pos_predictor_result.append(len([1 for x in pos_pred_test if x == "POS"])/float(len(pos_pred_test)))
neg_predictor_result.append(len([1 for x in neg_pred_test if x == "NEG"])/float(len(neg_pred_test)))
self.iteration_timer_estimate+=(time()-iteration_start)
return pos_predictor_result, neg_predictor_result