def draw_spec(self, cutoff=None, already_flipped=False):
'''
Draw the spectrogram of self.samples as loaded by load_samples(),
cutting off at sample `cutoff` if provided
'''
if cutoff:
freqs, times, spect = make_spect(self.samples[:cutoff],
self.samples_per_seg,
self.overlap_percent,
self.sample_rate)
else:
freqs, times, spect = make_spect(self.samples,
self.samples_per_seg,
self.overlap_percent,
self.sample_rate)
flip_axis = True
if already_flipped:
flip_axis = False
plotter(spect,
self.ax,
upside_down=flip_axis,
title=self.files[self.position])
self.fig.canvas.draw()
def test_plot_kline(self):
k = plotter()
data_path = "../quantist3/data/pool/"
result_path = "../quantist3/data/result/"
#result_name = 'plot_kline'
data = pd.read_excel(data_path + 'sh2.xls')
k.plot_kline(data, show_num=120) # sh
def test_semeion(trainer_type, testel, trainel):
images_dir = "/Users/jian/Dropbox/AI_dropbox/progetto_2014/dummy_data_set/courier_digits_data_set/tiff_images_swidth"
results_dir = "/Users/jian/Dropbox/AI_dropbox/progetto_2014/results/test_semeion" + "/" + trainer_type
semeion_dir = "/Users/jian/Dropbox/AI_dropbox/progetto_2014/semeion_data_set/semeion.data"
filename = results_dir + "/" + "tr" + str(trainel) + "_ts"+ str(testel) + "_c" + str(corr_ratio) + "_e" + str(erase_ratio) + "_" + trainer_type + "." + filetype
dim = [16, 16]
image_dim = [dim[1], dim[0]] # changing shape for images
# Loading images data set
temp_train = iM.collectimages(image_dim, images_dir, filter)
# image conversion to 1 and -1 for Hopfield net
for i in range(temp_train.shape[0]):
temp = utl.image_converter(temp_train[i].reshape(dim))
temp_train[i] = temp.flatten()
train_input = np.zeros((trainel, dim[0] * dim[1]))
for i in range(trainel):
train_input[i] = temp_train[i]
# training the net
net = HopfieldNet.HopfieldNet(train_input, trainer_type, dim)
# loading semeion data set
test_set = np.zeros((testel, dim[0], dim[1]))
for i in range(testel):
test_set[i] = utl.semeion_loader(semeion_dir, i)
# testing the net
result_set = np.zeros((testel, dim[0], dim[1]))
for i in range(testel):
# test_set[i] = temp_train[i].reshape(dim)
if corr_ratio != 0:
test_set[i] = utl.corrupter(test_set[i], corr_ratio)
if erase_ratio != 0:
test_set[i] = utl.image_eraser(test_set[i], erase_ratio)
result_set[i] = net.test(test_set[i])
# Plotting and saving results
utl.plotter(test_set, result_set, filename, plotbool, savebool)
def get_moves_old(game_map, turns, pid, training=False, graph=False):
w = game_map.width
h = game_map.height
me = game_map.get_me()
out = np.zeros((w, h))
graph_o, all_axes, F, Z = get_gx(game_map, training)
gradU, gradV = np.gradient(Z, axis=(0, 1)) # gradient of func
for idx, ship in enumerate(me.all_ships()):
sx, sy = int(ship.x), int(ship.y)
sv = vector(ship) # vector
# Distance/Magnitude/Norm/Length = np.sqrt(x**2+y**2) = np.sqrt([x,y].dot([x,y])
dx, dy = gradU[sx][sy], gradV[sx][sy] # unit vector of grad @ sx,sy
gm = F.norm(dx, dy)
u, v = -gm * dx, -gm * dy
angle = degrees(np.arctan2(v, u)) % 360
out[sx][sy] = angle
is_last_ship = idx == len(me.all_ships()) - 1
is_my_pid = pid == 0
if is_last_ship and is_my_pid and graph:
plotter(Z, sv, graph_o, turns, pid, w, h)
graph_o.clear()
return out
def test2(trainer_type, testel, trainel):
images_dir = "/Users/jian/Dropbox/AI_dropbox/progetto_2014/dummy_data_set/courier_digits_data_set/tiff_images_swidth"
results_dir = "/Users/jian/Dropbox/AI_dropbox/progetto_2014/results/test_2" + "/" + trainer_type
filename = results_dir + "/" + filter + "_" + "tr" + str(trainel) + "_ts"+ str(testel) + "_c" + str(corr_ratio) + "_e" + str(erase_ratio) + "_" + trainer_type + "." + filetype
dim = [14, 9] # in the form rows * cols
# testel = 8 # elements for training
#corruption_val = 5
# trainers = ["hebbian","pseudoinv","storkey"]
image_dim = [dim[1], dim[0]] # changing shape for images
# Loading images data set
temp_train = iM.collectimages(image_dim, images_dir, filter)
# image conversion to 1 and -1 for Hopfield net
for i in range(temp_train.shape[0]):
temp = utl.image_converter(temp_train[i].reshape(dim))
temp_train[i] = temp.flatten()
train_input = np.zeros((trainel, dim[0] * dim[1]))
for i in range(trainel):
train_input[i] = temp_train[i]
# training the net
net = HopfieldNet.HopfieldNet(train_input, trainer_type, dim)
# testing the net
test_set = np.zeros((testel, dim[0], dim[1]))
result_set = np.zeros((testel, dim[0], dim[1]))
for i in range(testel):
test_set[i] = temp_train[i].reshape(dim)
if corr_ratio != 0:
test_set[i] = utl.corrupter(test_set[i], corr_ratio)
if erase_ratio != 0:
test_set[i] = utl.image_eraser(test_set[i], erase_ratio)
result_set[i] = net.test(test_set[i])
# Plotting and saving results
utl.plotter(test_set, result_set, filename, plotbool, savebool)
def get_moves(game_map, turns, pid, training=False, graph=False):
w = game_map.width
h = game_map.height
me = game_map.get_me()
ships = me.all_ships()
out = {}
if graph:
grad_u, grad_v, graph_objs, Func, gridZ = get_gradient(
ships, game_map, graph)
else:
grad_u, grad_v = get_gradient(ships, game_map, graph)
for idx, ship in enumerate(ships):
sx, sy = int(ship.x), int(ship.y)
# Distance/Magnitude/Norm/Length = np.sqrt(x**2+y**2) = np.sqrt([x,y].dot([x,y])
sv = vector(ship)
u, v = grad_u[idx], grad_v[idx]
angle = degrees(np.arctan2(v, u)) % 360
out[sx, sy] = angle
is_last_ship = idx == len(ships) - 1
is_my_pid = pid == 0
if graph and is_last_ship and is_my_pid:
plotter(gridZ, sv, graph_objs, turns, pid, w, h)
graph_objs.clear()
return out
hyp2 = hyperParameters()
hyp2.cov = np.array([0.0,0.0])
hyp2.lik = np.array([np.log(0.1)])
#vargout = min_wrapper(hyp2,gp,'CG',inffunc,[],covfunc,likfunc,x,y,None,None,True)
#hyp2 = vargout[0]
hyp2.cov = np.array([-0.993396880620537,0.685943441677086])
hyp2.lik = np.array([-1.902546786026883])
#vargout = gp(hyp2,inffunc,[],covfunc,likfunc,x,y,None,None,False)
#print "nlml2 = ",vargout[0]
vargout = gp(hyp2,inffunc,[],covfunc,likfunc,x,y,z)
ym = vargout[0]; ys2 = vargout[1]
m = vargout[2]; s2 = vargout[3]
## Plot results
plotter(z,ym,ys2,x,y,[-1.9, 1.9, -0.9, 3.9])
###########################################################
'''covfunc = [ ['kernels.covSEiso'] ]
hyp = hyperParameters()
hyp.cov = np.array([0.0,0.0])
hyp.mean = np.array([0.0,0.0])
hyp.lik = np.array([np.log(0.1)])
vargout = min_wrapper(hyp,gp,'BFGS',inffunc,meanfunc,covfunc,likfunc,x,y,None,None,True)
hyp = vargout[0]
hyp.mean = np.array([1.1919,1.4625])
hyp.cov = np.array([-1.1513,-0.4559])
hyp.lik = np.array([-1.9122])
vargout = gp(hyp,inffunc,meanfunc,covfunc,likfunc,x,y,z)
ym = vargout[2]; ys2 = vargout[3]
def train(self):
# initialize memory buffer
buffer = ReplayBuffer(int(500000), self.batch_size, self.num_agents, 0)
# use keep_awake to keep workspace from disconnecting
for episode in range(self.number_of_episodes):
env_info = self.env.reset(train_mode=True)[self.brain_name]
agent_episode_rewards = [0, 0]
for agent in self.maddpg.ddpg_agents:
agent.noise.reset()
for episode_t in range(self.max_episode_len):
states = env_info.vector_observations
states_t = to_tensor(states)
with torch.no_grad():
action_ts = self.maddpg.act(states_t, noise=self.noise)
self.noise *= self.noise_reduction
actions = torch.stack(action_ts).numpy()
env_info = self.env.step(actions)[self.brain_name]
next_states = env_info.vector_observations
rewards = env_info.rewards
dones = env_info.local_done
for i in range(self.num_agents):
agent_episode_rewards[i] += rewards[i]
full_state = np.concatenate(states)
full_next_state = np.concatenate(next_states)
buffer.add((states, full_state, actions, rewards, next_states, full_next_state, dones))
# update once after every episode_per_update
critic_losses = []
actor_losses = []
if len(buffer) > self.batch_size and episode % self.episode_per_update == 0:
for i in range(self.num_agents):
samples = buffer.sample()
cl, al = self.maddpg.update(samples, i)
critic_losses.append(cl)
actor_losses.append(al)
self.maddpg.update_targets() # soft update the target network towards the actual networks
if np.any(dones):
# if any of the agents are done break
break
episode_reward = max(agent_episode_rewards)
self.episode_rewards.append(episode_reward)
self.last_100_episode_rewards.append(episode_reward)
self.avg_rewards.append(np.mean(self.last_100_episode_rewards))
# scores.append(episode_reward)
print('\rEpisode {}\tAverage Score: {:.4f}\tScore: {:.4f}'.format(episode, self.avg_rewards[-1],
episode_reward),
end="")
if episode % self.print_period == 0:
print('\rEpisode {}\tAverage Score: {:.4f}'.format(episode, self.avg_rewards[-1]))
# saving successful model
# training ends when the threshold value is reached.
if self.avg_rewards[-1] >= self.threshold:
save_dict_list = []
for i in range(self.num_agents):
save_dict = {'actor_params': self.maddpg.ddpg_agents[i].actor.state_dict(),
'actor_optim_params': self.maddpg.ddpg_agents[i].actor_optimizer.state_dict(),
'critic_params': self.maddpg.ddpg_agents[i].critic.state_dict(),
'critic_optim_params': self.maddpg.ddpg_agents[i].critic_optimizer.state_dict()}
save_dict_list.append(save_dict)
torch.save(save_dict_list, self.ckpt)
raw_score_plotter(self.episode_rewards)
plotter('Tennis', len(self.episode_rewards), self.avg_rewards, self.threshold)
break
features_batch = conv_base.predict(inputs_batch)
features[i * batch_size:(i + 1) * batch_size] = features_batch
labels[i * batch_size:(i + 1) * batch_size] = labels_batch
i += 1
if i * batch_size >= sample_count:
break
return features, labels
train_features, train_labels = extract_features(train_dir, 2000)
validation_features, validation_labels = extract_features(validation_dir, 1000)
test_features, test_labels = extract_features(test_dir, 1000)
train_features = np.reshape(train_features, (2000, 4 * 4 * 512))
validation_features = np.reshape(validation_features, (1000, 4 * 4 * 512))
test_features = np.reshape(test_features, (1000, 4 * 4 * 512))
model = models.Sequential()
model.add(layers.Dense(256, activation='relu', input_dim=4 * 4 * 512))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(1, activation='sigmoid'))
model.compile(optimizer=optimizers.RMSprop(lr=2e-5),
loss='binary_crossentropy',
metrics=['accuracy'])
history = model.fit(train_features,
train_labels,
epochs=30,
batch_size=20,
validation_data=(validation_features, validation_labels))
plotter(history.history)
def train(D, G, curr_lr, lr, n_epoch, beta1, beta2, bs):
data_loader = CreateDataLoader(batchSize=bs)
dataset = data_loader.load_data()
# Create optimizers for the generators and discriminators
optimizer_G = torch.optim.Adam(G.parameters(), lr=lr, betas=(beta1, beta2))
optimizer_D = torch.optim.Adam(D.parameters(), lr=lr, betas=(beta1, beta2))
res_d = []
res_g = []
total_steps = 0
for epoch in range(1, n_epoch + 1):
print("Running epoch:", epoch)
start_time_epoch = time.time()
sum_d = 0
sum_g = 0
for i, data in enumerate(dataset):
total_steps += c.batchSize
images_X = data['A']
images_Y = data['B']
# move images to GPU if available (otherwise stay on CPU)
# train discriminator on real
real_A = Variable(images_X.to(device))
fake_B = G.forward(real_A)
real_B = Variable(images_Y.to(device))
# =======================Train the discriminator=======================#
for iter in range(5):
optimizer_D.zero_grad()
# Real images
D_real = D.forward(real_B)
# Fake images
D_fake = D.forward(fake_B.detach())
# Gradient penalty
gradient_penalty = calc_gradient_penalty(D, real_B.data, fake_B.data)
d_loss = D_fake.mean() - D_real.mean() + gradient_penalty
d_loss.backward(retain_graph=True)
optimizer_D.step()
if iter == 4:
sum_d += d_loss.item()
#========================Train the generator===========================#
optimizer_G.zero_grad()
fake_B = G.forward(real_A)
D_fake = D.forward(fake_B)
g_loss = -D_fake.mean()
g_contentloss = perceptual_loss(fake_B, real_B) * 100
g_total_loss = g_loss + g_contentloss
g_total_loss.backward()
optimizer_G.step()
# printing pnsr & SSIM metrics at certain frequency
if total_steps % c.display_freq == 4:
image_res = util.get_visuals(real_A, fake_B, real_B)
psnr = metrics.PSNR(image_res['Restored_Train'], image_res['Sharp_Train'])
print('PSNR on Train (at epoch {0}) = {1}'.format(epoch, psnr))
ssim = metrics.SSIM_my(image_res['Restored_Train'], image_res['Sharp_Train'])
print('SSIM_my on Train (at epoch {0}) = {1}'.format(epoch, ssim))
# print losses & errors
# if total_steps % c.print_freq == 0:
# err = util.get_errors(g_loss, g_contentloss, d_loss)
# t = (time.time() - start_time_epoch) / c.batchSize
# util.print_errors(epoch, i, err, t)
# sum the loss over all the image
sum_g += g_total_loss.item()
# decaying learning rate
if epoch > 150:
lrd = 0.0001 / 150
new_lr = curr_lr - lrd
for param_group in optimizer_D.param_groups:
param_group['lr'] = new_lr
for param_group in optimizer_G.param_groups:
param_group['lr'] = new_lr
print('Update learning rate: %f -> %f' % (curr_lr, new_lr))
curr_lr = new_lr
# saving model after every 50 epochs
if epoch % c.save_freq == 0:
torch.save(G.state_dict(), 'model_G_' + str(epoch) + '.pt')
torch.save(D.state_dict(), 'model_D_' + str(epoch) + '.pt')
res_d.append(np.mean(sum_d))
res_g.append(np.mean(sum_g))
end_time_epoch = time.time()
print("Time for epoch {0}: {1} | Disc loss: {2} | Gen loss: {3}".format(epoch, (end_time_epoch - start_time_epoch), res_d[epoch-1], res_g[epoch-1]))
torch.save(G.state_dict(), 'model_G_last.pt')
torch.save(D.state_dict(), 'model_D_last.pt')
print("Model Saved!")
util.plotter(res_d, res_g)
def fit(train_dataloader,
val_dataloader,
model,
optimizer,
loss_fn,
n_epochs,
post_classification=False,
with_acc=False,
with_scheduler=False):
train_is_triplet = train_dataloader.dataset.is_triplet
val_is_triplet = val_dataloader.dataset.is_triplet
train_len = len(train_dataloader)
val_len = len(val_dataloader)
history = []
for epoch in range(n_epochs):
model.train()
train_loss = 0
val_loss = 0
train_acc = 0
val_acc = 0
for inputs, labels in tqdm(train_dataloader, desc="Train iteration"):
if with_scheduler:
prev_sd = model.state_dict()
optimizer.zero_grad()
if train_is_triplet:
anchor, pos, neg = inputs
anchor = anchor.to(device='cuda')
pos = pos.to(device='cuda')
neg = neg.to(device='cuda')
outputs = model(anchor, pos, neg)
if not post_classification:
loss = loss_fn(*outputs)
elif post_classification:
y0 = torch.tensor([0 for _ in range(len(outputs[0]))
]).to(device='cuda')
y1 = torch.tensor([1 for _ in range(len(outputs[1]))
]).to(device='cuda')
y = torch.cat((y0, y1))
outputs = torch.cat((outputs[0], outputs[1]))
loss = loss_fn(outputs, y)
elif not train_is_triplet:
w0, w1 = inputs
w0 = w0.to(device='cuda')
w1 = w1.to(device='cuda')
labels = labels.to(device='cuda')
outputs = model(w0, w1)
loss = loss_fn(outputs, labels)
loss.backward()
optimizer.step()
if with_acc:
if post_classification:
train_acc += accuracy_score(outputs, y)
elif not post_classification:
train_acc += accuracy_score(outputs, labels)
train_loss += loss.item()
model.eval()
with torch.no_grad():
for inputs, labels in tqdm(val_dataloader, desc="Val iteration"):
if val_is_triplet:
anchor, pos, neg = inputs
anchor = anchor.to(device='cuda')
pos = pos.to(device='cuda')
neg = neg.to(device='cuda')
outputs = model(anchor, pos, neg)
loss = loss_fn(*outputs)
else:
w0, w1 = inputs
w0 = w0.to(device='cuda')
w1 = w1.to(device='cuda')
labels = labels.to(device='cuda')
if not post_classification:
outputs = model(w0, w1)
elif post_classification:
outputs = model.classifire_it(w0, w1)
loss = loss_fn(outputs, labels)
val_loss += loss.item()
if with_acc:
val_acc += accuracy_score(outputs, labels)
epoch_train_loss = train_loss / train_len
epoch_val_loss = val_loss / val_len
if with_acc:
epoch_train_acc = train_acc / train_len
epoch_val_acc = val_acc / val_len
if with_scheduler:
flag = False
if epoch > 2:
flag = True
if epoch_val_loss > prev_loss:
model.load_state_dict(prev_sd)
epoch_val_loss = prev_loss
optimizer.param_groups[0]['lr'] /= 1.5
flag = False
if (epoch <= 2) or (flag == True):
history.append([
epoch, epoch_train_loss, epoch_val_loss, epoch_train_acc,
epoch_val_acc
])
flag = False
prev_loss = epoch_val_loss
elif not with_scheduler:
if with_acc:
history.append([
epoch, epoch_train_loss, epoch_val_loss, epoch_train_acc,
epoch_val_acc
])
else:
history.append([epoch, epoch_train_loss, epoch_val_loss])
plotter(history)
def train(config_file):
with open(config_file) as stream:
config_data = yaml.safe_load(stream)
use_only_specified_gpu(config_data['data_creation_parameters']['GPU_id'])
base_model_type = config_data['training_parameters']['base_type']
print("Reading dataset...")
reader = HMDBDataReader(
dataset_directory=config_data['dataset_save_dir'],
batch_size=config_data['training_parameters']['batch_size'],
sequence_len=config_data['data_creation_parameters']['sequence_len'],
base_type=base_model_type)
train_ds, _, val_ds = reader.get_datasets_sequence()
if bool(config_data['data_creation_parameters']['display_training_data']):
print("Displaying sequence...")
reader.display_sequences_train()
##### CHOOSING THE BASE MODEL #####
if base_model_type == 'VGG':
base_model = tf.keras.applications.VGG16(include_top=False,
weights='imagenet')
feature_map_size = 7
image_input_size = 224
filter_no = 512
else:
raise ValueError("No proper base model chosen for training!")
##### CHOOSING THE ARCHITECTURE #####
units_first_lstm = config_data['training_parameters']['lstm_parameters'][0]
if config_data['training_parameters']['model_type'] == 'ALSTM':
print('Training ALSTM')
classifier = ALSTM(
base_model=base_model,
use_dropout=bool(
config_data['training_parameters']['use_dropout']),
train_base=bool(config_data['training_parameters']['train_base']),
units_first_lstm=units_first_lstm,
no_classes=config_data['training_parameters']['no_classes'],
feature_map_size=feature_map_size)
elif config_data['training_parameters']['model_type'] == 'ConvALSTM':
print('Training ConvALSTM')
classifier = ConvALSTM(
base_model=base_model,
use_dropout=bool(
config_data['training_parameters']['use_dropout']),
train_base=bool(config_data['training_parameters']['train_base']),
units_first_lstm=units_first_lstm,
no_classes=config_data['training_parameters']['no_classes'],
feature_map_size=feature_map_size)
else:
raise ValueError("No proper model chosen for training.")
build_attention_network(
classifier, config_data['training_parameters']['batch_size'],
config_data['data_creation_parameters']['sequence_len'],
units_first_lstm, config_data['training_parameters']['model_type'],
feature_map_size, image_input_size, filter_no)
##### START TRAINING #####
trainer = AttentionTrainer(
network=classifier,
network_name=config_data['training_parameters']['model_type'],
optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
loss_object=tf.keras.losses.SparseCategoricalCrossentropy(),
loss_object_attention_pen=tf.keras.losses.MeanSquaredError(),
train_loss=tf.keras.metrics.Mean(name='train_loss'),
train_accuracy=tf.keras.metrics.SparseCategoricalAccuracy(
name='train_accuracy'),
train_precision=tf.keras.metrics.Precision(name='train_precision'),
train_recall=tf.keras.metrics.Recall(name='train_recall'),
val_loss=tf.keras.metrics.Mean(name='val_loss'),
val_accuracy=tf.keras.metrics.SparseCategoricalAccuracy(
name='val_accuracy'),
val_precision=tf.keras.metrics.Precision(name='val_precision'),
val_recall=tf.keras.metrics.Recall(name='val_recall'),
test_loss=tf.keras.metrics.Mean(name='test_loss'),
test_accuracy=tf.keras.metrics.SparseCategoricalAccuracy(
name='test_accuracy'),
test_precision=tf.keras.metrics.Precision(name='test_precision'),
test_recall=tf.keras.metrics.Recall(name='test_recall'),
batch_size=config_data['training_parameters']['batch_size'],
penalty_coeff=config_data['training_parameters']['penalty_coeff'],
weight_decay=config_data['training_parameters']['weight_decay'])
training_data = []
for epoch in range(config_data['training_parameters']['epochs']):
trainer.train_loss.reset_states()
trainer.train_accuracy.reset_states()
trainer.train_precision.reset_states()
trainer.train_recall.reset_states()
trainer.val_loss.reset_states()
trainer.val_accuracy.reset_states()
trainer.val_precision.reset_states()
trainer.val_recall.reset_states()
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
train_log_dir = os.path.join(
config_data['model_save_dir'],
'logs/gradient_tape/') + current_time + '/train'
val_log_dir = os.path.join(
config_data['model_save_dir'],
'logs/gradient_tape/') + current_time + '/val'
train_summary_writer = tf.summary.create_file_writer(train_log_dir)
val_summary_writer = tf.summary.create_file_writer(val_log_dir)
print('Training...')
for (batch, (image, labels)) in enumerate(train_ds):
if image.shape[0] != config_data['training_parameters'][
'batch_size']:
print('Invalid batch size, skipped...')
else:
trainer.train_step(image, labels)
with train_summary_writer.as_default():
tf.summary.scalar('train loss',
trainer.train_loss.result(),
step=batch)
tf.summary.scalar('train accuracy',
trainer.train_accuracy.result(),
step=batch)
print('Validating...')
for (batch, (image_val, labels_val)) in enumerate(val_ds):
if image_val.shape[0] != config_data['training_parameters'][
'batch_size']:
print('Invalid batch size, skipped...')
else:
trainer.val_step(image_val, labels_val)
with val_summary_writer.as_default():
tf.summary.scalar('val loss',
trainer.val_loss.result(),
step=batch)
tf.summary.scalar('val accuracy',
trainer.val_accuracy.result(),
step=batch)
epoch_summary = (epoch + 1, trainer.train_loss.result(),
trainer.train_accuracy.result(),
f1_score(trainer.train_precision.result(),
trainer.train_recall.result()),
trainer.val_loss.result(),
trainer.val_accuracy.result(),
f1_score(trainer.val_precision.result(),
trainer.val_recall.result()))
training_data.append(epoch_summary)
template = 'Epoch {}, Loss: {}, Accuracy:{}, F1 Score: {}, Val Loss: {}, Val Acc: {}, Val F1 Score: {}'
print(
template.format(
epoch_summary[0],
epoch_summary[1],
epoch_summary[2],
epoch_summary[3],
epoch_summary[4],
epoch_summary[5],
epoch_summary[6],
))
model_savedir = os.path.join(
config_data['model_save_dir'],
config_data['training_parameters']['model_type'])
if not os.path.exists(model_savedir):
os.mkdir(model_savedir)
plotter(training_data, model_savedir)
with io.open(os.path.join(model_savedir, '{}'.format(config_file)),
'w',
encoding='utf8') as outfile:
yaml.dump(config_data,
outfile,
default_flow_style=False,
allow_unicode=True)
trainer.save_weights(epoch, model_savedir)
print("Weights have been saved. Epoch {} done!".format(
epoch_summary[0]))
this_L2 *= wtl2
this_DGz /= opt.update_measures_plots
this_Dx /= opt.update_measures_plots
this_Adv /= opt.update_measures_plots
this_L2 /= opt.update_measures_plots
this_G_tot /= opt.update_measures_plots
this_D_tot /= opt.update_measures_plots
D_G_zs.append(this_DGz)
D_xs.append(this_Dx)
Advs.append(this_Adv)
L2s.append(this_L2)
G_tots.append(this_G_tot)
D_tots.append(this_D_tot)
plotter(D_G_zs, D_xs, Advs, L2s, G_tots, D_tots, (len(dataloader) / opt.update_measures_plots), PATHS["plots"])
this_DGz = 0
this_Dx = 0
this_Adv = 0
this_L2 = 0
this_G_tot = 0
this_D_tot = 0
step_counter = 0
if i % opt.update_train_img == 0:
if not opt.jointD:
recon_image = input_cropped.clone()
recon_image.data[:, :,
int(opt.imageSize / 2 - opt.patchSize / 2):int(opt.imageSize / 2 + opt.patchSize / 2),
int(opt.imageSize / 2 - opt.patchSize / 2):int(opt.imageSize / 2 + opt.patchSize / 2)] = fake.data
stacked_x = y_pred.reshape(
(y_pred.shape[0], y_pred.shape[1] * y_pred.shape[2]))
# split stacked samples to train and test
stacked_x_train, stacked_x_test,\
stacked_y_train, stacked_y_test = train_test_split(stacked_x, y_test, train_size=0.75, random_state=0)
# train the same MLP against stacked data
stacked_model, stacked_history = get_simple_mlp(stacked_x_train,
stacked_y_train,
dimensions=n * classes,
classes=classes,
epochs=500)
stacked_loss, stacked_acc = stacked_model.evaluate(stacked_x_test,
stacked_y_test,
verbose=0)
print('Stacked accuracy: %.3f' % stacked_acc)
print('Stacked loss: %.3f' % stacked_loss)
plotter(stacked_history.history)
etc = ExtraTreesClassifier(n_jobs=-1)
etc.fit(stacked_x_train, stacked_y_train)
etc_pred = etc.predict(stacked_x_test)
etc_acc_score = accuracy_score(stacked_y_test, etc_pred)
print('ExtraTreesClassifier stacked accuracy score: %.3f' % etc_acc_score)
rfc = RandomForestClassifier(n_jobs=-1)
rfc.fit(stacked_x_train, stacked_y_train)
rfc_pred = rfc.predict(stacked_x_test)
rfc_acc_score = accuracy_score(stacked_y_test, rfc_pred)
print('RandomForestClassifier stacked accuracy score: %.3f' %
rfc_acc_score)
st.write(
'''In this challenge, you are asked to predict whether a user will churn after his/her subscription expires. Specifically, we want to forecast if a user make a new service subscription transaction within 30 days after the current membership expiration date.'''
)
col1 = st.beta_columns(1)
# KAPLAN MEIER CURVES
drop_cols = [
'customer_id', 'bd', 'city', 'reg_month', 'tx_last_date', 'mem_end_date',
'latest_actual_amount_paid', 'latest_payment_method_id', 'avg_tot_secs',
'avg_num_unq', 'duration'
]
st.title('Kaplan-Meier Curves')
option = st.selectbox('', [x for x in df.columns if x not in drop_cols])
plt = plotter(df, option, DURATION, EVENT, CategoricalDtype)
KM_plot = st.pyplot(plt)
st.title("Model Summary")
st.write(model.summary)
# COX PROPORTIONAL HAZARDS SUMMARY
#from lifelines import CoxPHFitter
#rossi= load_rossi()
#st.title('Regression Model Summary')
#cph = CoxPHFitter()
#cph.fit(df, duration_col=DURATION, event_col=EVENT)
#st.write("## Coefficients")
#cols = ['coef','exp(coef)', 'exp(coef) lower 95%', 'exp(coef) upper 95%', 'p']
four_cliques_graph_noise=four_cliques_graph_noise,
dim_feature=dim_feature,
num_clusters=num_clusters)
# identify cluster data
if cluster_to_idx and idx_to_cluster:
cluster_data = (cluster_to_idx, idx_to_cluster)
else:
cluster_data = None
# Load a specific agent
agent = load_agent(algorithm_name=algorithm_name,
dim_feature=dim_feature,
alpha=alpha,
graph=network,
cluster_data=cluster_data)
# the dummy agent for normalization
agent_normalized = load_agent(algorithm_name='dummy',
dim_feature=dim_feature,
alpha=alpha,
graph=network,
cluster_data=cluster_data)
# Run the experiment
regrets = runner(agent=agent,
agent_normalized=agent_normalized,
user_contexts=user_contexts,
time_steps=time_steps)
# Plot the figure and write the result
plotter(results=regrets, output_filename=output_filename)
## SET (hyper)parameters
hyp = hyperParameters()
## SET (hyper)parameters for covariance and mean
hyp.cov = np.array([np.log(67.), np.log(66.), np.log(1.3), np.log(1.0), np.log(2.4), np.log(90.), np.log(2.4), \
np.log(1.2), np.log(0.66), np.log(0.78), np.log(1.6/12.), np.log(0.18), np.log(0.19)])
hyp.mean = np.array([])
sn = 0.1
hyp.lik = np.array([np.log(sn)])
#_________________________________
# STANDARD GP:
### TEST POINTS
xs = np.arange(2004+1./24.,2024-1./24.,1./12.)
xs = xs.reshape(len(xs),1)
vargout = gp(hyp,inffunc,meanfunc,covfunc,likfunc,x,y,xs)
ym = vargout[0]; ys2 = vargout[1]
m = vargout[2]; s2 = vargout[3]
plotter(xs,ym,ys2,x,y)
#vargout = min_wrapper(hyp,gp,'CG',inffunc,meanfunc,covfunc,likfunc,x,y,None,None,True)
#hyp = vargout[0]
#vargout = gp(hyp,inffunc,meanfunc,covfunc,likfunc,x,y,xs)
#ym = vargout[0]; ys2 = vargout[1]
#m = vargout[2]; s2 = vargout[3]
#plotter(xs,ym,ys2,x,y)
