def Acc(y_t, y_p):
"""Computes accuracy for each label
Argument: ground truth label Y (2D reshaped), prediction (2D reshaped)
Returns: array of accuracy for each label"""
a = BinaryAccuracy() #handles the fact that y_p is not given as a binary vector
Acc_per_label = []
for i in range(8):
a.update_state(y_t[:,i], y_p[:,i] )
acc = a.result().numpy()
a.reset_states()
Acc_per_label.append(acc)
return Acc_per_label
def print_bin_predictions_match(y_test, yhat, binary=True):
accuracy = BinaryAccuracy() if binary else SparseCategoricalAccuracy()
for i in range(y_test.shape[0]):
ix = colored(str(f'{i:02d} |'), 'grey')
sep = colored('|', 'grey')
y = int(y_test[i][0])
pred = f"{yhat[i][0]:.02f}" if binary else f"{np.argmax(yhat[i])}"
accuracy.reset_states()
accuracy.update_state(y, yhat[i])
true_pred = accuracy.result().numpy() == 1.0
color = "green" if true_pred else "red"
pred = colored(pred, color)
if i % 9 == 0:
print(f"\n{ix} ", end='')
print(f"{y} {sep} {pred} {sep} ", end="")
print(colored("\n", "white"))