def plot_wrong_classification(x, y_true, y_pred, filename):
samples = []
fig_size = (10, 10)
for i in range(0, x.shape[0]):
if y_true[i] != y_pred[i]:
samples.append((y_true[i], y_pred[i], i))
def plot_sample(x, axis):
img = x.reshape(x.shape[0], x.shape[1])
axis.imshow(img, cmap='gray')
fig = plt.figure(figsize=fig_size)
for i in range(len(samples)):
y_t, y_p, index = samples[i]
ax = fig.add_subplot(*fig_size, i + 1, xticks=[], yticks=[])
title = map_int_to_letter(y_t) + " -> " + map_int_to_letter(y_p)
ax.title.set_text(title)
ax.title.set_fontsize(10)
plot_sample(x[index], ax)
fig.tight_layout()
plt.savefig(filename)
samples.append((y_true[i], y_pred[i], i))
def plot_sample(x, axis):
img = x.reshape(30, 30)
axis.imshow(img, cmap='gray')
fig = plt.figure(figsize=(10, 6))
print()
for i in range(len(samples)):
y_t, y_p, index = samples[i]
ax = fig.add_subplot(10, 6, i + 1, xticks=[], yticks=[])
title = cro_mapper.map_int_to_letter(
y_t) + " -> " + cro_mapper.map_int_to_letter(y_p)
ax.title.set_text(title)
ax.title.set_fontsize(10)
plot_sample(x_test[index], ax)
fig.tight_layout()
# fig.subplots_adjust(top=0.88)
plt.show()
# def twoway_confusion_matrix(cm, i):
# tp = cm[i, i]
# fn = np.sum(cm[i,:]) - tp
# fp = np.sum(cm[:,i]) - tp
# tn = np.sum(cm) - fp - fn - tp
# return np.matrix([[tp, fp], [fn, tn]]).astype(float)
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("weights-improvement-645-0.94.hdf5")
print("Loaded model from disk")
image = misc.imread("/Users/filipgulan/Desktop/test.png", mode='F')
print("Image loaded...")
X = []
X.append(image)
X = np.asarray(X)
X = X.reshape(X.shape[0], img_rows, img_cols, 1)
X = X.astype('float32')
X /= 255
prediction = loaded_model.predict(X)[0]
print("Predictions:")
prediction_dict = dict(enumerate(prediction))
prediction_sorted = sorted(prediction_dict.items(),
key=lambda x: x[1],
reverse=True)
counter = 0
for key, value in prediction_sorted:
print(cro_mapper.map_int_to_letter(key), value)
counter += 1
if counter == 5:
break
index = np.argmax(prediction)
print("Recognized letter:", cro_mapper.map_int_to_letter(index))
if tw[0, 0] != 0:
precision += tw[0, 0] / (tw[0, 0] + tw[1, 0])
return precision / size
else:
conf = np.zeros((2, 2)).astype(float)
for tw in tw_ms:
conf += tw
conf /= size
return conf[0, 0] / (conf[0, 0] + conf[1, 0])
def twoway_confusion_matrix(cm, i):
tp = cm[i, i]
fn = np.sum(cm[i, :]) - tp
fp = np.sum(cm[:, i]) - tp
tn = np.sum(cm) - fp - fn - tp
return np.matrix([[tp, fp], [fn, tn]]).astype(float)
test_confusion = confusion_matrix(y_true, y_pred)
for i in range(test_confusion.shape[0]):
print("Matrica zabune za klasu", cro_mapper.map_int_to_letter(i))
tw = twoway_confusion_matrix(test_confusion, i)
print("Preciznost", precision_c(tw, averaging="macro"))
print("Odziv", recall_c(tw, averaging="macro"))
print("Tocnost", accuracy_c(tw, averaging="macro"))
print(tw)
np.set_printoptions(threshold=np.inf)
# with open("confmatrix_10000.txt", 'w') as f:
# f.write(np.array2string(confusion_matrix(y_true, y_pred), separator=', '))
def save_confusion_matrix_csv(confusion_matrix, filename):
num_classes = confusion_matrix.shape[0]
header = [map_int_to_letter(letter_int) for letter_int in range(num_classes)]
header = ",".join(header)
np.savetxt(filename, confusion_matrix.astype(int), fmt='%i', delimiter=",")