def print_perf(params, iter, gradient):
if iter % 30 == 0:
#save_images(sigmoid(params), 'q4plot.png')
fig = plt.figure(1)
fig.clf()
ax = fig.add_subplot(111)
plot_images(sigmoid(params), ax)
print(batched_loss(params, iter))
log_prob = log_softmax(X, weight)
avg = avg_log_likelihood(((log_prob)), Y)
pred = np.argmax(np.exp(log_prob), axis=1)
return pred, avg
if __name__ == "__main__":
train_images = binarize_data(train_images)
test_images = binarize_data(test_images)
testY = np.argmax(test_labels, axis=1)
trainY = np.argmax(train_labels, axis=1)
weight, loss = optimization(0.1, train_images[:300], train_labels[:300])
data.save_images(weight.T, "unreg_Q3_weight")
data.plot_images(weight.T, plt.figure().add_subplot(111))
plt.show()
print("Train Set Result:")
class_label = np.argmax(train_labels[:300], axis=1)
pred, avg = prediction(train_images[:300], weight, class_label)
print("avg prediction Error is {}".format(np.mean(pred != class_label)))
print("The predictive avg log-likelihood is {}".format(avg))
print("\n")
print("Test Set Result:")
class_label = np.argmax(test_labels, axis=1)
pred, avg = prediction(test_images, weight, class_label)
print("avg prediction Error is {}".format(np.mean(pred != class_label)))
print("The predictive avg log-likelihood is {}".format(avg))
# Get gradient of objective using autograd.
objective_grad = elementwise_grad(batched_loss)
def print_perf(params, iter, gradient):
if iter % 30 == 0:
#save_images(sigmoid(params), 'q4plot.png')
fig = plt.figure(1)
fig.clf()
ax = fig.add_subplot(111)
plot_images(sigmoid(params), ax)
print(batched_loss(params, iter))
# The optimizers provided by autograd can optimize lists, tuples, or dicts of parameters.
# You may use these optimizers for Q4, but implement your own gradient descent optimizer for Q3!
optimized_params = adam(objective_grad,
init_params,
step_size=0.2,
num_iters=10,
callback=print_perf)
#optimized_params = adam(objective_grad, init_params, step_size=0.2, num_iters=10000, callback=print_perf)
#Q4D code: just need to run 2fs code given optimized_params
results = advanced_bayes(train_images[0:20, :], optimized_params)
fig = plt.figure(1)
fig.clf()
ax = fig.add_subplot(111)
plot_images(images, ax)
save_images(results, 'Q4D')
from tensorflow.keras.callbacks import ModelCheckpoint
from model import build_model
if __name__ == '__main__':
X_train, y_train = data.get_data_generator(data.IMAGE_TRAIN_PATH,
data.MASK_TRAIN_PATH)
X_val, y_val = data.get_data_generator(data.IMAGE_VALIDATION_PATH,
data.IMAGE_VALIDATION_PATH)
model = build_model()
model_checkpoint = ModelCheckpoint('unet_200_steps.hdf5',
monitor='loss',
verbose=1,
save_best_only=True)
history = model.fit(zip(X_train, y_train),
callbacks=[
model_checkpoint,
],
epochs=200,
steps_per_epoch=234 // 16,
batch_size=16,
validation_data=(zip(X_val, y_val)),
validation_steps=10)
X_test, y_test = data.get_data_generator(data.IMAGE_TEST_PATH,
data.MASK_TEST_PATH)
x = X_test.next()
y = y_test.next()
data.plot_images(y, model.predict(x)[0])