def train_and_evaluate_model(hp):
"""Trains and evaluates a model."""
hp.set(run_id=datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
_record_hyperparams(hp)
for path in [_checkpoints_dir(hp.run_id), _logs_dir(hp.run_id)]:
os.makedirs(path, exist_ok=True)
if hp.word_emb.pretrained == "none":
word2vec = {}
elif hp.word_emb.pretrained == "glove":
word2vec = embedding_utils.read_glove(hp.word_emb.dim, pruned=True)
else:
assert False, hp.word_emb.pretrained
v = Vocab()
v.add_dataset(dataset_iter(TRAIN_FILENAME))
v.add_words(word2vec.keys())
d_train = Dataset(v, dataset_iter(TRAIN_FILENAME))
d_test = Dataset(v, dataset_iter(TEST_FILENAME))
# Drop rare words so that UNK appears in the training set.
d_train.drop_rare_words(hp.drop_rare_words_freq)
model = build_model(d_train, word2vec, hp)
print(model.summary())
x_train, y_train, x_test, y_test, history = train_model(
model, d_train, d_test, hp)
plotting.plot_history(history, filepath=_plot_history_filepath(hp))
evaluate_model("train", model, d_train, x_train, y_train, hp)
evaluate_model("test", model, d_test, x_test, y_test, hp)
def main():
nn = NN([None, layers.ConvolutionalLayer(10, 10, 28, 28, 5, list()),
layers.FullyConnectedLayer(10, 19 * 19, list(), unit=neuron.Logistic())])
#nn = NN([layers.FullyConnectedLayer(784,784,list()), layers.ConvolutionalLayer(10,10,28,28,5,list()), layers.FullyConnectedLayer(10, 19*19,list(),unit=neuron.Logistic())])
#nn = NN([layers.FullyConnectedLayer(784, 784, list()), layers.FullyConnectedLayer(28, 784, list()), layers.FullyConnectedLayer(10, 28, list(), unit=neuron.Logistic())])
# read in data
"""
raw_data = pd.read_csv(
"/Users/delbalso/projects/nn1/data/handwriting.csv",
sep=",",
header=None)
raw_data = raw_data.reindex(np.random.permutation(raw_data.index))
data = np.array(raw_data.transpose())
num_labels = 10
num_test_data = int(data.shape[1] * 0.2)
features = data[:-num_labels, :] # num_features x num_examples
labels = data[-1 * num_labels:, :] # num_labels x num_examples
weights = train(
labels[
:,
:-
1 *
num_test_data],
features[
:,
:-
num_test_data],
nn)
test(labels[:, -num_test_data:], features[:, -num_test_data:], weights, nn)
"""
training_data, validation_data, test_data = mnist.load_data_wrapper_1()
random.shuffle(training_data)
training_features, training_labels = zip(*training_data[:500])
training_data = MLDataSet(
np.squeeze(training_features).transpose(),
np.squeeze(training_labels).transpose())
validation_features, validation_labels = zip(*validation_data[:])
validation_data = MLDataSet(
np.squeeze(validation_features).transpose(),
np.squeeze(validation_labels).transpose())
test_features, test_labels = zip(*test_data)
test_data = MLDataSet(
np.squeeze(test_features).transpose(),
np.squeeze(test_labels).transpose())
#hyperparam_search(1, 10, 1, 100, nn, training_data, validation_data)
train(training_data, nn, 30, 10, validation_data=validation_data)
test(test_data, nn)
plot.plot_history(
training_accuracy_history,
validation_accuracy_history,
training_cost_history,
validation_cost_history)
def main():
nn = NN([
None,
layers.ConvolutionalLayer(10, 10, 28, 28, 5, list()),
layers.FullyConnectedLayer(10, 19 * 19, list(), unit=neuron.Logistic())
])
#nn = NN([layers.FullyConnectedLayer(784,784,list()), layers.ConvolutionalLayer(10,10,28,28,5,list()), layers.FullyConnectedLayer(10, 19*19,list(),unit=neuron.Logistic())])
#nn = NN([layers.FullyConnectedLayer(784, 784, list()), layers.FullyConnectedLayer(28, 784, list()), layers.FullyConnectedLayer(10, 28, list(), unit=neuron.Logistic())])
# read in data
"""
raw_data = pd.read_csv(
"/Users/delbalso/projects/nn1/data/handwriting.csv",
sep=",",
header=None)
raw_data = raw_data.reindex(np.random.permutation(raw_data.index))
data = np.array(raw_data.transpose())
num_labels = 10
num_test_data = int(data.shape[1] * 0.2)
features = data[:-num_labels, :] # num_features x num_examples
labels = data[-1 * num_labels:, :] # num_labels x num_examples
weights = train(
labels[
:,
:-
1 *
num_test_data],
features[
:,
:-
num_test_data],
nn)
test(labels[:, -num_test_data:], features[:, -num_test_data:], weights, nn)
"""
training_data, validation_data, test_data = mnist.load_data_wrapper_1()
random.shuffle(training_data)
training_features, training_labels = zip(*training_data[:500])
training_data = MLDataSet(
np.squeeze(training_features).transpose(),
np.squeeze(training_labels).transpose())
validation_features, validation_labels = zip(*validation_data[:])
validation_data = MLDataSet(
np.squeeze(validation_features).transpose(),
np.squeeze(validation_labels).transpose())
test_features, test_labels = zip(*test_data)
test_data = MLDataSet(
np.squeeze(test_features).transpose(),
np.squeeze(test_labels).transpose())
#hyperparam_search(1, 10, 1, 100, nn, training_data, validation_data)
train(training_data, nn, 30, 10, validation_data=validation_data)
test(test_data, nn)
plot.plot_history(training_accuracy_history, validation_accuracy_history,
training_cost_history, validation_cost_history)
def mccv(x, y, model, plot=(), epochs=10, batch_size=10, iterations=500):
"""
:param plot: To plot something, the first value of "plot" has to be the title of the plot, and the second has to
be the name of the output file (without any extensions). These values are passed as a tuple.
"""
auc = [] # ROC AUC.
spec = [
] # Proportion of actual negatives that are correctly identified as such.
sens = [
] # Proportion of actual positives that are correctly identified as such.
history = [] # List of history objects. Used for plotting.
for _ in tqdm(
range(iterations)
): # Rewrite to "for _ in range(iterations)" to remove tqdm dependency.
x_new, x_tst, y_new, y_tst = train_test_split(x,
y,
test_size=16,
stratify=y) # Test data.
x_tr, x_val, y_tr, y_val = train_test_split(
x_new, y_new, test_size=0.3, stratify=y_new) # Train and val data.
history.append(
model.fit(x_tr,
y_tr,
validation_data=(x_val, y_val),
epochs=epochs,
batch_size=batch_size,
verbose=0))
predictions = [i[0] for i in model.predict(x=x_tst).tolist()
] # Flattening prediction list.
auc.append(roc_auc_score(y_tst, predictions))
report = classification_report(
y_tst, [0 if p < 0.5 else 1 for p in predictions],
output_dict=True)
spec.append(report["0"]["recall"])
sens.append(report["1"]["recall"])
print(ci(auc))
print(ci(spec))
print(ci(sens))
if plot:
history_avg = deepcopy(history[0])
history_avg.history["acc"] = mean([h.history["acc"] for h in history],
axis=0)
history_avg.history["loss"] = mean(
[h.history["loss"] for h in history], axis=0)
history_avg.history["val_acc"] = mean(
[h.history["val_acc"] for h in history], axis=0)
history_avg.history["val_loss"] = mean(
[h.history["val_loss"] for h in history], axis=0)
plot_history(plot[0], history_avg, plot[1])
def fine_tuning(preprocess, pretrained_vgg_model, debug=False):
if debug:
pretrained_vgg_model.summary()
number_of_classes = 10
pretrained_layers = pretrained_vgg_model.layers
fine_tuned = list()
for idx, num_samples in enumerate([100, 1000, 10000]):
model = keras.Sequential()
for pretrained_layer in pretrained_layers[:
-2]: # Without last Dense and its Activation
pretrained_layer.trainable = False
model.add(pretrained_layer)
# Adding the trainable layer and its activation
model.add(Dense(number_of_classes))
model.add(Activation('softmax'))
if debug:
model.summary()
batch_sizes = 8, 32, 128 #TODO: are you sure you want different?
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train, _, y_train, _ = train_test_split(x_train,
y_train,
train_size=num_samples,
random_state=42,
stratify=y_train)
x_train = preprocess(x_train)
x_test = preprocess(x_test)
y_train = keras.utils.to_categorical(y_train, number_of_classes)
y_test = keras.utils.to_categorical(y_test, number_of_classes)
batch_size = batch_sizes[idx]
max_epochs = 200 if 100 == num_samples else 100
learning_rate = 0.0001
lr_decay = 1e-6
lr_drop = 20
def lr_scheduler(epoch):
return learning_rate * (0.5**(epoch // lr_drop))
reduce_lr = keras.callbacks.LearningRateScheduler(lr_scheduler)
if debug:
max_epochs = 3
x_test = x_test[:20]
y_test = y_test[:20]
# data augmentation
datagen = ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset
samplewise_center=False, # set each sample mean to 0
featurewise_std_normalization=
False, # divide inputs by std of the dataset
samplewise_std_normalization=False, # divide each input by its std
# zca_whitening=False, # apply ZCA whitening
# rotation_range=15, # randomly rotate images in the range (degrees, 0 to 180)
# width_shift_range=0.1, # randomly shift images horizontally (fraction of total width)
# height_shift_range=0.1, # randomly shift images vertically (fraction of total height)
# horizontal_flip=True, # randomly flip images
vertical_flip=False) # randomly flip images
# (std, mean, and principal components if ZCA whitening is applied).
datagen.fit(x_train)
# optimization details
#sgd = optimizers.SGD(lr=learning_rate, decay=lr_decay, momentum=0.9, nesterov=True)
sgd = optimizers.Adam(lr=learning_rate, decay=lr_decay)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
model.summary()
history = model.fit_generator(
datagen.flow(x_train, y_train, batch_size=batch_size),
steps_per_epoch=x_train.shape[0] // batch_size,
epochs=max_epochs,
validation_data=(x_test, y_test),
# callbacks=[reduce_lr],
verbose=1)
model.save_weights(
'cifar10vgg_finetuning_{}_trainset.h5'.format(num_samples))
fine_tuned.append((model, history))
plot_history(
history,
'Fine tuning with {} training samples'.format(num_samples))
return fine_tuned
def collect():
name_list = [
"mnist",
#"mnistcov"
]
nr_of_epochs = 8000
record_all_flag = False
rec_test_flag = True
learning_r = [0.0004]
save_data_flag = False
show_flag = True
separate_flag = False
save_MI_and_plot_flag = False
bin_size_or_nr = [True]
bins = [0.01, 0.07, 0.15, 0.3]
color_list = [
"red", "blue", "green", "orange", "purple", "brown", "pink", "teal",
"goldenrod"
]
models = [3, 4, 2, 1, 5, 6]
for set_name in name_list:
nrs = [3, 8, 1]
#samples = "full"
samples = 1600
seed(1337)
set_random_seed(1337)
X_train, X_test, y_train, y_test = data_selection.select_data(
set_name, shuffle=True, samples_per_class=samples, list_of_nrs=nrs)
batch_size = [256, X_train.shape[0], 128, 512]
print("calculations starting for: ", set_name)
for i in models:
if (i <= 4) and ("cov" in set_name):
continue
if (i > 4) and ("cov" not in set_name):
continue
for batch in batch_size:
seed(1337)
set_random_seed(1337)
# objects to record parameters
outputs = classes.Outputs()
# define and train model
output_recording = LambdaCallback(
on_epoch_end=lambda
epoch, logs: Callbacks.record_activations(
outputs, model, epoch, X_train, X_test, y_test, batch,
record_all_flag, rec_test_flag))
model, architecture = model_selection.select_model(
i, nr_of_epochs, set_name, X_train.shape, y_train)
adam = optimizers.Adam(lr=learning_r)
model.compile(loss="categorical_crossentropy",
optimizer=adam,
metrics=["accuracy"])
history = model.fit(X_train,
y_train,
epochs=nr_of_epochs,
batch_size=batch,
validation_split=0.2,
callbacks=[output_recording])
# final model score
score = model.evaluate(X_test, y_test, verbose=0)
score = score[1]
# save data
common_name = architecture + "_lr_" + str(
learning_r) + "_batchsize_" + str(batch)
if "mnist" in set_name:
common_name = str(samples) + str(nrs) + common_name
aname = common_name + "_activations"
outputs.model_score = score
if save_data_flag == True:
util.save(outputs, aname)
hname = common_name + "_history"
h_obj = history.history
h_obj["model_score"] = score
if save_data_flag == True:
util.save(h_obj, hname)
plotting.plot_history(h_obj, common_name, show_flag,
save_MI_and_plot_flag)
if rec_test_flag == True:
plotting.plot_test_development(outputs.int_model_score,
common_name, show_flag,
save_MI_and_plot_flag)
# compute binning MI
for flag in bin_size_or_nr:
for nr_of_bins in bins:
if flag == True and nr_of_bins > 1:
continue
if flag == False and nr_of_bins < 1:
continue
seed(1337)
set_random_seed(1337)
est_type_flag = 1
info_plane.create_infoplane(common_name,
X_train,
y_train,
outputs,
est_type_flag,
color_list,
nr_of_bins,
flag,
show_flag,
separate_flag,
save_MI_and_plot_flag,
par_flag=False)
seed(1337)
set_random_seed(1337)
# compute EDGE
if batch == 256:
est_type_flag = 2
info_plane.create_infoplane(
common_name,
X_train,
y_train,
outputs,
est_type_flag,
color_list,
show_flag,
separate_flag,
save_flag=save_MI_and_plot_flag,
par_flag=True)
seed(1337)
set_random_seed(1337)
# compute KDE upper
est_type_flag = 3
info_plane.create_infoplane(common_name,
X_train,
y_train,
outputs,
est_type_flag,
color_list,
show_flag,
separate_flag,
save_flag=save_MI_and_plot_flag,
par_flag=False)
seed(1337)
set_random_seed(1337)
# compute KDE lower
if batch == 256:
est_type_flag = 4
info_plane.create_infoplane(
common_name,
X_train,
y_train,
outputs,
est_type_flag,
color_list,
show_flag,
separate_flag,
save_flag=save_MI_and_plot_flag,
par_flag=False)
seed(1337)
set_random_seed(1337)
# compute KSG discrete
est_type_flag = 5
info_plane.create_infoplane(common_name,
X_train,
y_train,
outputs,
est_type_flag,
color_list,
show_flag,
separate_flag,
save_flag=save_MI_and_plot_flag,
par_flag=False)
cuda_particles = cuda.mem_alloc(4 * N * (6 + 6*MAX_LANDMARKS))
cuda_measurements = cuda.mem_alloc(4 * 2 * MAX_MEASUREMENTS)
cuda_cov = cuda.mem_alloc(4 * 4)
# plt.pause(5)
for i in range(u.shape[0]):
print(i)
if PLOT:
plt.pause(0.05)
ax[0].clear()
ax[0].set_xlim([-5, 20])
ax[0].set_ylim([-5, 20])
plot_landmarks(ax[0], landmarks)
plot_history(ax[0], real_position_history, color='green')
plot_history(ax[0], predicted_position_history, color='orange')
plot_particles_grey(ax[0], particles)
vehicle.move_noisy(u[i])
real_position_history.append(vehicle.position)
FlatParticle.predict(particles, u[i], sigmas=movement_variance, dt=1)
visible_measurements = sensor.get_noisy_measurements(vehicle.position[:2])
particles = update(
particles, THREADS, BLOCK_SIZE, visible_measurements, measurement_variance,
cuda_particles, cuda_measurements, cuda_cov, THRESHOLD
)
validation_generator = test_datagen.flow_from_directory(validation_dir,
target_size=(150, 150),
batch_size=BATCH_SIZE,
class_mode='binary')
model_2.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=2e-5),
metrics=['acc'])
history = model_2.fit_generator(train_generator,
steps_per_epoch=N_TRAIN_IMAGES // BATCH_SIZE,
epochs=10,
validation_data=validation_generator,
validation_steps=50)
plot_history(history)
conv_base.trainable = True
set_trainable = False
for layer in conv_base.layers:
if layer.name == 'block5_conv1':
set_trainable = True
if set_trainable:
layer.trainable = True
else:
layer.trainable = False
model_2.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=1e-5),
metrics=['acc'])
def run_cem(env_id,
epochs=50,
batch_size=4096,
elite_frac=0.0625,
randomize=True,
only_success_elite=False,
reward_type="sparse",
env_index=0,
single_obj_reward=-1,
extra_std=2.0,
extra_decay_time=10,
num_process=8):
now = datetime.now()
time_now = now.strftime("%m-%d-%H-%M-%S")
output_dir = './{}/{}/'.format(env_id, time_now)
print("output_dir: ", output_dir)
ensure_dir(output_dir)
start = time.time()
num_episodes = epochs * num_process * batch_size
print('expt of {} total episodes'.format(num_episodes))
num_elite = int(batch_size * elite_frac)
history = defaultdict(list)
z_dim = 4 * 12
means = np.zeros(z_dim)
stds = np.ones(z_dim)
for epoch in tqdm(range(epochs)):
print("current epoch number: ", epoch)
extra_cov = max(1.0 - epoch / extra_decay_time, 0) * extra_std**2
zs = np.random.multivariate_normal(
mean=means,
cov=np.diag(np.array(stds**2) + extra_cov),
size=batch_size)
with Pool(num_process) as p:
returns_successes = p.map(
partial(evaluate_z,
randomize=randomize,
reward_type=reward_type,
output_dir=output_dir,
epoch=epoch,
env_index=env_index,
single_obj_reward=single_obj_reward), zs)
print(returns_successes)
returns = [rs[0] for rs in returns_successes]
successes = [rs[1] for rs in returns_successes]
log_prob_sum = [rs[2] for rs in returns_successes]
print("successes: ")
print(successes)
print("log prob sum: ")
print(log_prob_sum)
returns = np.array(returns)
successes = np.array(successes)
indexes = get_elite_indicies(num_elite, returns, successes,
only_success_elite)
elites = zs[indexes]
means = elites.mean(axis=0)
stds = elites.std(axis=0)
history['epoch'].append(epoch)
history['avg_ret'].append(np.mean(returns))
history['std_ret'].append(np.std(returns))
history['avg_ret_elites'].append(np.mean(returns[indexes]))
history['std_ret_elites'].append(np.std(returns[indexes]))
history['avg_suc'].append(np.mean(successes))
history['std_suc'].append(np.std(successes))
history['avg_suc_elites'].append(np.mean(successes[indexes]))
history['std_suc_elites'].append(np.std(successes[indexes]))
print('epoch {} - population returns: {} {} - elite returns: {} {}'.
format(epoch, history['avg_ret'][-1], history['std_ret'][-1],
history['avg_ret_elites'][-1],
history['std_ret_elites'][-1]))
print(
'epoch {} - population successes: {} {} - elite successes: {} {}'.
format(epoch, history['avg_suc'][-1], history['std_suc'][-1],
history['avg_suc_elites'][-1],
history['std_suc_elites'][-1]))
if True: #epoch % 5 == 0:
end = time.time()
expt_time = end - start
plot_history(history, output_dir, epoch, expt_time)
save_path_history = os.path.join(output_dir,
"history_{}.npy".format(epoch))
np.save(save_path_history, history)
save_path_elites = os.path.join(output_dir,
"elites_{}.npy".format(epoch))
np.save(save_path_elites, elites)
end = time.time()
expt_time = end - start
print('expt took {:2.1f} seconds'.format(expt_time))
plot_history(history, output_dir, epochs, expt_time)
num_optimal = 5
print('epochs done - evaluating {} best zs'.format(num_optimal))
best_z_rewards = [
evaluate_z(z,
randomize=randomize,
reward_type=reward_type,
output_dir=output_dir,
epoch=epoch,
env_index=env_index,
single_obj_reward=single_obj_reward)
for z in elites[:num_optimal]
]
print('best rewards - {} across {} samples'.format(best_z_rewards,
num_optimal))
EPOCH = 1
train_losses, train_ious, valid_losses, valid_ious = [], [], [], []
for metrics in net.train_and_validate(train_loader,
valid_loader,
args.epochs,
optimizer=optimizer,
transform_data=transform_data,
denormalizer=denormalize):
estimate_masked, train_loss, train_iou, valid_loss, valid_iou = metrics
save_batch(
estimate_masked,
os.path.join(figure_dir, "estimates",
f"estimates-{EPOCH}.png"))
train_losses.append(train_loss)
train_ious.append(train_iou)
valid_losses.append(valid_loss)
valid_ious.append(valid_iou)
EPOCH += 1
fig = plot_history({
"train_loss": train_losses,
"train_iou": train_ious,
"valid_loss": valid_losses,
"valid_iou": valid_ious
})
fig.savefig(os.path.join(figure_dir, "output_figure.pdf"))
plt.close(fig)
torch.save(net.state_dict(), os.path.join(figure_dir, "model.pth"))
torch.save(net, os.path.join(figure_dir, "model_object.pt"))
else:
print(f"{option} not implemented yet")
def run_cem(env_id,
epochs=10,
batch_size=4096,
elite_frac=0.2,
extra_std=2.0,
extra_decay_time=10,
num_process=8):
ensure_dir('./{}/'.format(env_id))
start = time.time()
num_episodes = epochs * num_process * batch_size
print('expt of {} total episodes'.format(num_episodes))
num_elite = int(batch_size * elite_frac)
history = defaultdict(list)
env, obs_shape, act_shape = setup_env(env_id)
theta_dim = (obs_shape + 1) * act_shape
means = np.random.uniform(size=theta_dim)
stds = np.ones(theta_dim)
for epoch in range(epochs):
extra_cov = max(1.0 - epoch / extra_decay_time, 0) * extra_std**2
thetas = np.random.multivariate_normal(
mean=means,
cov=np.diag(np.array(stds**2) + extra_cov),
size=batch_size)
with Pool(num_process) as p:
rewards = p.map(partial(evaluate_theta, env_id=env_id), thetas)
rewards = np.array(rewards)
indicies = get_elite_indicies(num_elite, rewards)
elites = thetas[indicies]
means = elites.mean(axis=0)
stds = elites.std(axis=0)
history['epoch'].append(epoch)
history['avg_rew'].append(np.mean(rewards))
history['std_rew'].append(np.std(rewards))
history['avg_elites'].append(np.mean(rewards[indicies]))
history['std_elites'].append(np.std(rewards[indicies]))
print('epoch {} - {:2.1f} {:2.1f} pop - {:2.1f} {:2.1f} elites'.format(
epoch, history['avg_rew'][-1], history['std_rew'][-1],
history['avg_elites'][-1], history['std_elites'][-1]))
end = time.time()
expt_time = end - start
print('expt took {:2.1f} seconds'.format(expt_time))
plot_history(history, env_id, num_episodes, expt_time)
num_optimal = 3
print('epochs done - evaluating {} best thetas'.format(num_optimal))
best_theta_rewards = [
evaluate_theta(theta, env_id, monitor=True)
for theta in elites[:num_optimal]
]
print('best rewards - {} acoss {} samples'.format(best_theta_rewards,
num_optimal))
np.tile([0.0, 0.7], (14, 1))))
# s_history = [s]
ss_history = [ss]
slam_history = [get_mean(particles)]
for i in range(68):
print(i)
plt.pause(0.01)
ax.clear()
ax.set_xlim([0, 17])
ax.set_ylim([0, 17])
plot_landmarks(ax, landmarks)
plot_history(ax, ss_history, color='green')
plot_history(ax, slam_history, color='orange')
# plot_connection(ax, ss, z_real[landmark_indices, :] + ss[:2])
plot_particles_grey(ax, particles)
# s = move_vehicle_exact(s, u[i], dt=1)
ss = move_vehicle_stochastic(ss, u[i], dt=1, sigmas=[0.05, 0.05])
# s_history.append(s)
ss_history.append(ss)
particles = predict(particles, u[i], sigmas=[0.05, 0.05], dt=1)
# print(particles)
# sigmas = [0.05, 0.05]
R = np.array([[sigmas[0], 0], [0, sigmas[1]]], dtype=np.float)
def show_history_test(filenames, path):
for f in filenames:
with open(path + f, 'r') as pkl_fd:
history = pickle.load(pkl_fd)
plotting.plot_history(history, 'acc',
'Densenet121 - ' + f.split('_')[-3])
class_mode='binary')
history = model.fit_generator(train_generator,
steps_per_epoch=100,
epochs=EPOCHS,
validation_data=validation_generator,
validation_steps=50,
callbacks=callbacks_list)
else:
train_datagen = ImageDataGenerator(rescale=1. / 255)
test_datagen = ImageDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(train_dir,
target_size=(150, 150),
batch_size=BATCH_SIZE,
class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
validation_dir,
target_size=(150, 150),
batch_size=BATCH_SIZE,
class_mode='binary')
history = model.fit_generator(train_generator,
steps_per_epoch=2 * N_TRAIN_IMAGES //
BATCH_SIZE,
epochs=20,
validation_data=validation_generator,
validation_steps=50,
callbacks=callbacks_list)
model.save(osp.join(CONV_NETS_DIR, 'cats_and_dogs_small_1.h5'))
plot_history(history)