def neural_network_test_table_gen():
X, y = default_img_set()
X = [[full_prepare(img) for img in same_sample] for same_sample in X]
Xs = count_blobs_with_all_methods(X)
Xs = [np.array(X_count) for X_count in Xs]
y = np.array(y)
row_names = ('Wszystkie detale', 'Śledzone detale',
'Stosunek śledzonych detali')
with open('exports/neural_network_test.csv', 'w') as csvfile:
filewriter = csv.writer(csvfile, delimiter=';')
# Header
filewriter.writerow(('Metoda zliczania detali', 'Wskaźnik',
'wskaźnik'))
filewriter.writerow(('Metoda zliczania detali', 'Błąd', 'Dokładność'))
for X, name in zip(Xs, row_names):
X = np.array(X)
X_train, X_test, y_train, y_test = train_test_split(
X, y, stratify=y, test_size=0.33, random_state=1)
model = default_grain_classifier_model()
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
model.fit(X_train, y_train, epochs=300, verbose=0)
score = model.evaluate(X_test, y_test, verbose=0)
filewriter.writerow((name, *score))
def neural_network_validation_table_gen():
X, y = default_img_set()
X = [[full_prepare(img) for img in same_sample] for same_sample in X]
Xs = count_blobs_with_all_methods(X)
Xs = [np.array(X_count) for X_count in Xs]
y = np.array(y)
row_names = ('Wszystkie detale', 'Śledzone detale',
'Stosunek śledzonych detali')
with open('exports/neural_network_validation.csv', 'w') as csvfile:
filewriter = csv.writer(csvfile, delimiter=';')
# Header
filewriter.writerow(('Metoda zliczania detali', 'Wskaźnik',
'wskaźnik'))
filewriter.writerow(('Metoda zliczania detali', 'Błąd', 'Dokładność'))
for X, name in zip(Xs, row_names):
model = default_grain_classifier_model()
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
scores = np.array(network_cross_validation(model, X, y, 3))
score = np.round(scores.mean(axis=0), 2)
filewriter.writerow((name, *score))
def neural_network_trainig_plots_gen():
X, y = default_img_set()
X = [[full_prepare(img) for img in same_sample] for same_sample in X]
Xs = count_blobs_with_all_methods(X)
files_suffixes = ('all', 'remaining', 'ratio')
for X, suffix in zip(Xs, files_suffixes):
X = np.array(X)
y = np.array(y)
X_train, _, y_train, _ = train_test_split(
X, y, stratify=y, test_size=0.33, random_state=1)
model = default_grain_classifier_model()
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
history = model.fit(X_train, y_train, epochs=300, verbose=0)
plt.figure()
plt.title('Historia treningu modelu')
plt.xlabel('Epoka')
plt.plot(history.history['accuracy'], c='b')
plt.plot(history.history['loss'], c='r')
plt.legend(('Dokładność', 'Błąd'))
tikzplotlib.save('exports/neural_network_trainig_' + suffix)
def confusion_matrix_table_gen():
X, y = default_img_set()
X = [[full_prepare(img) for img in same_sample] for same_sample in X]
X = count_blobs_with_all_methods(X)[2]
X = np.array(X)
y = np.array(y)
model = default_grain_classifier_model()
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
mcm = mean_confusion_matrix(model, X, y, 3)
np.savetxt(
"exports/mean_confusion_matrix_ratio.csv",
mcm,
fmt='%.2f',
delimiter=";")
def network_comparison_table_gen():
X, y = default_img_set()
X = [[full_prepare(img) for img in same_sample] for same_sample in X]
X = [
ratio_of_remaining_blobs_in_stages(find_blob_series(img_series))
for img_series in X
]
X = np.array(X)
y = np.array(y)
with open('exports/neural_network_comparison.csv', 'w') as csvfile:
filewriter = csv.writer(csvfile, delimiter=';')
# Header
filewriter.writerow(('Parametr', 'Wartość', 'Błąd', 'Dokładność'))
# Activation functions
activation_funcs = ('sigmoid', 'relu', 'elu', 'tanh')
for func in activation_funcs:
model = keras.Sequential([
keras.layers.Dense(5, activation=func),
keras.layers.Dense(256, activation=func),
keras.layers.Dense(128, activation=func),
keras.layers.Dense(4, activation='softmax')
])
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
scores = np.array(network_cross_validation(model, X, y, 3))
score = np.round(scores.mean(axis=0), 2)
filewriter.writerow(('Funkcja aktywacji', func, *score))
# Number of hidden layers
models = []
models.append(
keras.Sequential([
keras.layers.Dense(5, activation='tanh'),
keras.layers.Dense(512, activation='tanh'),
keras.layers.Dense(4, activation='softmax')
]))
models.append(
keras.Sequential([
keras.layers.Dense(5, activation='tanh'),
keras.layers.Dense(256, activation='tanh'),
keras.layers.Dense(128, activation='tanh'),
keras.layers.Dense(4, activation='softmax')
]))
for model, i in zip(models, (1, 2)):
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
scores = np.array(network_cross_validation(model, X, y, 3))
score = np.round(scores.mean(axis=0), 2)
filewriter.writerow(('Liczba warstw ukrytych', i, *score))
# Number of neurons in hidden layers
neurons_num = ((128, 64), (256, 128), (512, 126))
for num in neurons_num:
model = keras.Sequential([
keras.layers.Dense(5, activation='tanh'),
keras.layers.Dense(num[0], activation='tanh'),
keras.layers.Dense(num[1], activation='tanh'),
keras.layers.Dense(4, activation='softmax')
])
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
scores = np.array(network_cross_validation(model, X, y, 3))
score = np.round(scores.mean(axis=0), 2)
filewriter.writerow(('Liczba neuronów w warstwach ukrytych',
'{} i {}'.format(num[0], num[1]), *score))
# Optimizer
model = default_grain_classifier_model()
optimizers = ('sgd', 'adam')
for opt in optimizers:
model.compile(
optimizer=opt,
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
scores = np.array(network_cross_validation(model, X, y, 3))
score = np.round(scores.mean(axis=0), 2)
filewriter.writerow(('Algorytm uczenia', opt, *score))