])
sens_opt_thresh_here = np.zeros([
__N_RUNS,
])
spec_opt_thresh_here = np.zeros([
__N_RUNS,
])
# Iterate over number of desired runs
for run_idx in range(__N_RUNS):
logger.info('\tWorking on run [%2d / %2d]...' %
(run_idx + 1, __N_RUNS))
# Create the cnn model
model = get_sino_cnn(input_shape=np.shape(train_data)[1:],
lr=0.001)
# Train the model
model.fit(x=train_data,
y=train_labels,
epochs=__N_EPOCHS,
shuffle=True,
batch_size=2056,
verbose=False)
# Predict on the test set
test_preds = model.predict(x=test_data)
# Get metrics at 0.50 decision threshold
auc = roc_auc_score(y_true=test_labels, y_score=test_preds)
acc = get_acc(preds=test_preds[:, 1], labels=test_labels[:, 1])
accs = np.zeros([
n_runs,
])
sens = np.zeros([
n_runs,
])
spec = np.zeros([
n_runs,
])
for run_idx in range(n_runs):
logger.info('\tWorking on run [%d / %d]...' % (run_idx + 1, n_runs))
# Get model
cnn = get_sino_cnn(input_shape=np.shape(g2_d)[1:], lr=0.001)
# Train the model
cnn.fit(x=g2_d,
y=g2_labels,
epochs=__N_EPOCHS,
shuffle=True,
batch_size=2048,
verbose=False)
# Calculate the predictions
g1_preds = cnn.predict(x=g1_d)
# Get and store ROC AUC
g1_auc = 100 * roc_auc_score(y_true=g1_labels, y_score=g1_preds)
auc_scores[run_idx] = g1_auc