def train_model(model, x_train, y_train, num_of_epochs, batch_size,
position_or_section):
class_weights = None
if position_or_section.upper() == "POSITION":
position_or_section = "Position"
else:
position_or_section = "Section"
class_weights = {0: 1., 1: 2., 2: 2., 3: 1.}
dashes = "-" * 50
print(dashes + " " + position_or_section + " model training " + dashes)
ANN.train_NN(model, x_train, y_train, class_weights, num_of_epochs,
batch_size)
print(dashes * 2)
def main():
config = Config.instance()
loader = UCIStudentPerformance(batch_size=config.BATCH_SIZE,
is_eval=config.TRAIN_MODE,
debug=config.DEBUG_MODE)
model = ANN(config.DEBUG_MODE)
optimizer = torch.optim.SGD(model.parameters(), lr=config.LEARNING_RATE)
trainer = ANNTrainer(model, loader, optimizer)
if config.TRAIN_MODE:
trainer.train(config.MAX_EPOCH, config.BATCH_SIZE)
else:
trainer.evaluate()
def do_ANN(x_train, y_train, x_test, y_test, position_or_section):
if position_or_section.upper() == "POSITION":
model = ANN.make_NN(x_train, 27)
elif position_or_section.upper() == "SECTION":
model = ANN.make_NN(x_train, 4)
else:
raise ValueError("Invalid value for position_or_section parameter")
# tried number of epochs: 10, 100, 1000 - none of them gave an accuracy of over 89%
# tried batch sizes: 10, 20, 50, 100, 200, 500 - none of them gave an accuracy of over 89%
train_model(model, x_train, y_train, 10, 100, position_or_section)
print("Test set evaluation: ",
model.evaluate(x_test, y_test, batch_size=100))
predictions = model.predict(x_test)
def __init__(self, state_size, action_size, seed):
"""Initialize an Agent object.
Params
======
state_size (int): dimension of each state
action_size (int): dimension of each action
seed (int): random seed
"""
self.state_size = state_size
self.action_size = action_size
self.seed = random.seed(seed)
# Q-Network
self.qnetwork_local = ANN(state_size, action_size, seed).to(device)
self.qnetwork_target = ANN(state_size, action_size, seed).to(device)
self.optimizer = optim.Adam(self.qnetwork_local.parameters(), lr=LR)
# Replay memory
self.memory = ReplayBuffer(action_size, BUFFER_SIZE, BATCH_SIZE, seed)
# Initialize time step (for updating every UPDATE_EVERY steps)
self.t_step = 0
def train():
# load data
X, y = load_data(DATA_PATH, IMP_M, SCALE_M)
# split train/valid/test
print('\nSplit... (65% train, 20% valid, 15% test)')
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.15,
random_state=seed)
print('Number of training+validation data:', X_train.shape[0])
print('Number of testing data:', X_test.shape[0])
# model
if (not os.path.isfile(ANN_PATH)) | ANN_AGAIN:
print('\nTraining ANN...')
from model import ANN
model = ANN(X_train,
y_train,
isW,
W_TRAINABLE,
ANN_PATH,
isPlot=OUTPUT_DIR)
else:
print('\nLoading ANN...')
model = load_model(ANN_PATH)
# evaluate model
train_scores = model.evaluate(X_train, y_train, verbose=0)
print("Train %s: %.2f%%" % (model.metrics_names[1], train_scores[1] * 100))
test_scores = model.evaluate(X_test, y_test, verbose=0)
print("Test %s: %.2f%%" % (model.metrics_names[1], test_scores[1] * 100))
# evaluation
if isEval:
print('\nEvaluating...')
from eval import eval
eval(model, X_test, y_test, OUTPUT_DIR)
class Agent():
"""Interacts with and learns from the environment."""
def __init__(self, state_size, action_size, seed):
"""Initialize an Agent object.
Params
======
state_size (int): dimension of each state
action_size (int): dimension of each action
seed (int): random seed
"""
self.state_size = state_size
self.action_size = action_size
self.seed = random.seed(seed)
# Q-Network
self.qnetwork_local = ANN(state_size, action_size, seed).to(device)
self.qnetwork_target = ANN(state_size, action_size, seed).to(device)
self.optimizer = optim.Adam(self.qnetwork_local.parameters(), lr=LR)
# Replay memory
self.memory = ReplayBuffer(action_size, BUFFER_SIZE, BATCH_SIZE, seed)
# Initialize time step (for updating every UPDATE_EVERY steps)
self.t_step = 0
def step(self, state, action, reward, next_state, done):
# Save experience in replay memory
self.memory.add(state, action, reward, next_state, done)
# Learn every UPDATE_EVERY time steps.
self.t_step = (self.t_step + 1) % UPDATE_EVERY
if self.t_step == 0:
# If enough samples are available in memory, get random subset and learn
if len(self.memory) > BATCH_SIZE:
experiences = self.memory.sample()
self.learn(experiences, GAMMA)
def act(self, state, eps=0.):
"""Returns actions for given state as per current policy.
Params
======
state (array_like): current state
eps (float): epsilon, for epsilon-greedy action selection
"""
state = torch.from_numpy(state).float().unsqueeze(0).to(device)
self.qnetwork_local.eval()
with torch.no_grad():
action_values = self.qnetwork_local(state)
self.qnetwork_local.train()
# Epsilon-greedy action selection
if random.random() > eps:
return np.argmax(action_values.cpu().data.numpy())
else:
return random.choice(np.arange(self.action_size))
def learn(self, experiences, gamma):
"""Update value parameters using given batch of experience tuples.
Params
======
experiences (Tuple[torch.Tensor]): tuple of (s, a, r, s', done) tuples
gamma (float): discount factor
"""
states, actions, rewards, next_states, dones = experiences
## TODO: compute and minimize the loss
"*** YOUR CODE HERE ***"
# states, actions, rewards, next_states, dones = experiences
# Double Q learning
best_next_action = self.qnetwork_local(next_states).detach().max(
1)[1].unsqueeze(1)
Q_targets_next = self.qnetwork_target(next_states).detach().gather(
1, best_next_action)
# Q Learning
# Get max predicted Q values (for next states) from target model
# Q_targets_next = self.qnetwork_target(next_states).detach().max(1)[0].unsqueeze(1)
# Compute Q targets for current states
Q_targets = rewards + (gamma * Q_targets_next * (1 - dones))
# Get expected Q values from local model
Q_expected = self.qnetwork_local(states).gather(1, actions)
# Compute loss
loss = F.mse_loss(Q_expected, Q_targets)
# Minimize the loss
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
# ------------------- update target network ------------------- #
self.soft_update(self.qnetwork_local, self.qnetwork_target, TAU)
def soft_update(self, local_model, target_model, tau):
"""Soft update model parameters.
θ_target = τ*θ_local + (1 - τ)*θ_target
Params
======
local_model (PyTorch model): weights will be copied from
target_model (PyTorch model): weights will be copied to
tau (float): interpolation parameter
"""
for target_param, local_param in zip(target_model.parameters(),
local_model.parameters()):
target_param.data.copy_(tau * local_param.data +
(1.0 - tau) * target_param.data)
import argparse
import numpy as np
import tensorflow as tf
from model import ANN
import data
parser = argparse.ArgumentParser(description='Visualize ANN')
parser.add_argument('-d',
'--dataset',
type=str,
default='mnist',
choices=data.get_names())
parser.add_argument('--num_iter', type=int, default=5000)
args = parser.parse_args()
dataset = data.init_dataset(name=args.dataset)
model = ANN(dataset.shape)
model.train(dataset.tr_data, dataset.tr_labels, num_iter=args.num_iter)
torch.from_numpy(test_subject_labels))
# tensor to DataLoader
train_dataloader = DataLoader(train_data,
batch_size=1024,
shuffle=True,
drop_last=True)
test_dataloader = DataLoader(test_data,
batch_size=1024,
shuffle=True,
drop_last=True)
# ----------------------------------------- Training & Validation ----------------------------------------- #
precision = 1e-8
ann = ANN(input_channel=448, number_of_classes=7).cuda()
class_criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(ann.parameters(), lr=1e-3)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer=optimizer,
mode='min',
factor=0.1,
patience=6,
verbose=True,
eps=precision)
epoch_num = 120
patience = 12
patience_increase = 12
best_loss = float('inf')
for epoch in range(epoch_num):
# print(test_label.shape)
# time.sleep(100)
# numpy to tensor
test_data = TensorDataset(torch.from_numpy(test_data),
torch.from_numpy(test_label))
# tensor to DataLoader
test_dataloader = DataLoader(test_data,
batch_size=4096,
shuffle=True,
drop_last=True)
print('Test segments loaded. Batch_size = 4096.\n')
print('Loading trained ANN network parameters...')
ann_test = ANN(input_channel=448, number_of_classes=7).cuda()
ann_test.load_state_dict(torch.load(r'saved_model\ANN_8_subjects.pkl'))
ann_test.eval()
total_batch_num = len(test_dataloader)
running_loss = 0.
correct_gesture_label, total_num = 0.0, 0.0
print('Start testing...\n')
for i, (data, label) in enumerate(test_dataloader):
data = data.cuda()
data = data.view(4096, -1)
gesture_label = label.cuda()
pred_label = ann_test(data)
'--num_workers',
type=int,
default=8,
help='The numbers of processors used for loading the data')
args = parser.parse_args()
#device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
os.makedirs('saved_models', exist_ok=True)
data = DataLoader(Data("./data/training"),
batch_size=args.batch_size,
shuffle=True,
num_workers=args.num_workers)
model = ANN()
criterion = torch.nn.L1Loss()
optimizer = torch.optim.Adam(model.parameters(),
lr=0.0002,
betas=(0.9, 0.999))
if args.epoch > 0:
model.load_state_dict(
torch.load('saved_models/model_%d.pth' % args.epoch))
for epoch in range(args.epoch, args.num_epochs):
for i, items in enumerate(data):
init = items['init']
init_pos, init_vel, init_t = init['pos'], init['vel'], init['t']
init_t = init_t.view(-1, 1)