def calc_acc(y_pred, y_test):
"""calculates accuracy between y_pred and y_test
input: y_pred is a 1D array of length n_image
input: y_test is a 2D array of shape n_image x 3
output: accuracy, float"""
y_act = get_labels.get_label_1D(y_test)
agree = np.where(y_pred == y_act)
return len(agree[0]) / float(len(y_pred))
def calc_acc(y_pred, y_test):
'''calculates accuracy between y_pred and y_test
input: y_pred is a 1D array of length n_image
input: y_test is a 2D array of shape n_image x 3
output: accuracy, float'''
y_act = get_labels.get_label_1D(y_test)
agree = np.where(y_pred == y_act)
return len(agree[0]) / float(len(y_pred))
# An epoch is one pass through the whole training data set, batch size is number of data points used for each iteration of gradient descent
NN_fit = model.fit(X_train_norm, y_train,
nb_epoch=20,
batch_size=150,
validation_data = (X_test_norm, y_test),
show_accuracy = True, verbose=2)
# Calculate metrics
y_pred_train = model.predict_classes(X_train_norm) # Predicted y_train classification
print "Train Acc:"
metrics.calc_acc(y_pred_train, y_train)
y_pred_test = model.predict_classes(X_test_norm) # Predicted y_test classification
print "Train Acc:"
calc_acc(y_pred_test, y_test)
# Classification report
y_act = get_labels.get_label_1D(y_test)
print(classification_report(y_act, y_pred_test, target_names=['spiral', 'elliptical', 'uncertain']))
# Plot confusion matrix
cm = confusion_matrix(y_act, y_pred_test)
plots.plot_confusion_matrix(cm, normed=True)
# Get probabilities for each classification
probas = model.predict_proba(X_test_norm)
# Calculate true positive and false positive rates treating each classifier as binary
sp_fpr, sp_tpr, _ = roc_curve(y_test20[:,0], probas[:,0])
ell_fpr, ell_tpr, _ = roc_curve(y_test20[:,1], probas[:,1])
unc_fpr, unc_tpr, _ = roc_curve(y_test20[:,2], probas[:,2])
# Plot ROC curves
plots.plot_ROC_curve(sp_tpr, sp_fpr, ell_tpr, ell_fpr, unc_tpr, unc_fpr)
# Print AUC scores
print "AUC for spirals:"
roc_auc_score(y_test[:,0], probas[:,0])
batch_size=150,
validation_data=(X_test_norm, y_test),
show_accuracy=True,
verbose=2)
# Calculate metrics
y_pred_train = model.predict_classes(
X_train_norm) # Predicted y_train classification
print "Train Acc:"
metrics.calc_acc(y_pred_train, y_train)
y_pred_test = model.predict_classes(
X_test_norm) # Predicted y_test classification
print "Train Acc:"
calc_acc(y_pred_test, y_test)
# Classification report
y_act = get_labels.get_label_1D(y_test)
print(
classification_report(
y_act,
y_pred_test,
target_names=['spiral', 'elliptical', 'uncertain']))
# Plot confusion matrix
cm = confusion_matrix(y_act, y_pred_test)
plots.plot_confusion_matrix(cm, normed=True)
# Get probabilities for each classification
probas = model.predict_proba(X_test_norm)
# Calculate true positive and false positive rates treating each classifier as binary
sp_fpr, sp_tpr, _ = roc_curve(y_test20[:, 0], probas[:, 0])
ell_fpr, ell_tpr, _ = roc_curve(y_test20[:, 1], probas[:, 1])
unc_fpr, unc_tpr, _ = roc_curve(y_test20[:, 2], probas[:, 2])
# Plot ROC curves
