def __init__(self):
self.n_games = 0
self.random_game = 50 # randomness
self.gamma = 0.9 # discount rate
self.memory = deque(maxlen=MAX_MEMORY) # popleft()
self.model = NN()
self.trainer = Train(self.model, gamma=self.gamma)
def __init__(self,
pop_size,
rows,
randomize_weights=True,
input_file=None):
# layer layout of each NN
layers = [24, 16, 16, 4]
# calculate the number of genes in a dna strand
self.dna_size = 0
for i in range(1, len(layers)):
self.dna_size += layers[i - 1] * layers[i]
# initialize the population
self.population = np.empty(pop_size, dtype=NN)
for i in range(pop_size):
self.population[i] = NN(layers, self.dna_size)
# insert the given weights if the user so chooses
if not randomize_weights:
self.population[0].set_weights(
np.loadtxt('data.csv', delimiter=','))
# size of the field
self.rows = rows
def predict(X):
nn = NN([6, 8, 4, 1])
nn.load_state_dict(torch.load('model_state/state'))
nn.eval()
result = nn.forward(X).detach().numpy().reshape(-1, 1)
result_transformed = scaler.inverse_transform(result)
return result_transformed
def train(self, epochs=10):
Model = NN(batch_size=30)
opt = optim.Adam(Model.parameters(), lr=0.005)
criterion = nn.BCELoss()
softmax = nn.Softmax(dim=0)
loss = 0
print(self.labels.shape)
for i in range(epochs):
item_loss = 0
for i, (feat, lab) in enumerate(self.dataloader):
feat = feat[0][:, :1, :, :]
output = Model.forward(feat)
loss = criterion(output, lab.view((30)))
loss.backward()
opt.step()
item_loss += loss.item()
print("LOSS ---->", item_loss / len(self.dataloader))
torch.save(Model.state_dict(), "1")
def net_loader(self, path = None):
testm = NN(self.Layer_s).to(device)
if path is None:
model_name = self.part_number + '_model_lr_' + str(self.Learning_r) + '_layer_size_' + str(self.Layer_s) + '.pt'
path = os.path.join(self.checkpoint_dir, model_name)
testm.load_state_dict(torch.load(path))
print(model_name,' Was loaded successfully loaded.\t\t\t [loaded]')
else:
testm.load_state_dict(torch.load(path))
print(model_name,' Was loaded successfully loaded from the path.\t\t\t [loaded from Path]')
return testm
def train_model(DATA_LENGTH, num_file):
#文件读取
x_train = pd.read_csv("x_train.csv")
x_test = pd.read_csv("x_test.csv")
y_train = pd.read_csv("y_train.csv")
y_test = pd.read_csv("y_test.csv")
#取出dataframe中的值，组成二维矩阵，行index表示样本index，列表示特征index，可以直接输入到算法中
x_train = x_train.iloc[:, :].values.reshape(-1, DATA_LENGTH)
x_test = x_test.iloc[:, :].values.reshape(-1, DATA_LENGTH)
y_train = y_train.iloc[:, :].values.reshape(-1, num_file)
y_test = y_test.iloc[:, :].values.reshape(-1, num_file)
fnn_layers = [600, 500, 400, 300, 200, 100, 50, 11]
my_netural_netweok = NN(fnn_layers=fnn_layers,
input_dim=DATA_LENGTH,
output_dim=num_file,
batch_size=100,
act=tf.nn.relu,
learning_rate=0.01,
keep_rate=0.05)
init = tf.initialize_all_variables()
feed_train = {
my_netural_netweok.x: x_train,
my_netural_netweok.y: y_train,
my_netural_netweok.model: 'train'
}
feed_test = {
my_netural_netweok.x: x_test,
my_netural_netweok.y: y_test,
my_netural_netweok.model: 'test'
}
epcho = 1000
with tf.Session() as sess:
sess.run(init)
for i in range(epcho):
error, _ = sess.run(
[my_netural_netweok.loss, my_netural_netweok.train_op],
feed_dict=feed_train)
if (i % 10 == 0):
print('train loss', error)
acc = sess.run(my_netural_netweok.accuracy,
feed_dict=feed_test)
print('test accuracy', acc)
def main(option):
logging.basicConfig(
stream=sys.stdout,
level=logging.DEBUG,
format=
'%(asctime)s - %(filename)s[line:%(lineno)d] - %(levelname)s: %(message)s'
)
dataset = HardSimilarityDataset()
if option.dataset_cache is None:
glove = Glove(option.emb_file)
logging.info('Embeddings loaded')
dataset.load(option.dataset_file, glove)
else:
dataset.load_cache(option.dataset_cache)
logging.info('Dataset loaded')
embeddings = nn.Embedding(option.vocab_size, option.emb_dim, padding_idx=1)
if option.model == 'NTN':
model = NeuralTensorNetwork(embeddings, option.em_k)
elif option.model == 'LowRankNTN':
model = LowRankNeuralTensorNetwork(embeddings, option.em_k,
option.em_r)
elif option.model == 'RoleFactor':
model = RoleFactoredTensorModel(embeddings, option.em_k)
elif option.model == 'Predicate':
model = PredicateTensorModel(embeddings)
elif option.model == 'NN':
model = NN(embeddings, 2 * option.em_k, option.em_k)
elif option.model == 'EMC':
model = EMC(embeddings, 2 * option.em_k, option.em_k)
else:
logging.info('Unknown model type: ' + option.model)
exit(1)
checkpoint = torch.load(option.model_file, map_location='cpu')
if type(checkpoint) == dict:
if 'event_model_state_dict' in checkpoint:
state_dict = checkpoint['event_model_state_dict']
else:
state_dict = checkpoint['model_state_dict']
else:
state_dict = checkpoint
model.load_state_dict(state_dict)
logging.info(option.model_file + ' loaded')
# embeddings = nn.Embedding(option.vocab_size, option.emb_dim, padding_idx=1)
# embeddings.weight.data = torch.from_numpy(glove.embd).float()
# model = Averaging(embeddings)
if option.use_gpu:
model.cuda()
model.eval()
data_loader = torch.utils.data.DataLoader(
dataset,
collate_fn=HardSimilarityDataset_collate_fn,
shuffle=False,
batch_size=len(dataset))
batch = next(iter(data_loader))
pos_e1_subj_id, pos_e1_subj_w, pos_e1_verb_id, pos_e1_verb_w, pos_e1_obj_id, pos_e1_obj_w, \
pos_e2_subj_id, pos_e2_subj_w, pos_e2_verb_id, pos_e2_verb_w, pos_e2_obj_id, pos_e2_obj_w, \
neg_e1_subj_id, neg_e1_subj_w, neg_e1_verb_id, neg_e1_verb_w, neg_e1_obj_id, neg_e1_obj_w, \
neg_e2_subj_id, neg_e2_subj_w, neg_e2_verb_id, neg_e2_verb_w, neg_e2_obj_id, neg_e2_obj_w = batch
if option.use_gpu:
pos_e1_subj_id = pos_e1_subj_id.cuda()
pos_e1_subj_w = pos_e1_subj_w.cuda()
pos_e1_verb_id = pos_e1_verb_id.cuda()
pos_e1_verb_w = pos_e1_verb_w.cuda()
pos_e1_obj_id = pos_e1_obj_id.cuda()
pos_e1_obj_w = pos_e1_obj_w.cuda()
pos_e2_subj_id = pos_e2_subj_id.cuda()
pos_e2_subj_w = pos_e2_subj_w.cuda()
pos_e2_verb_id = pos_e2_verb_id.cuda()
pos_e2_verb_w = pos_e2_verb_w.cuda()
pos_e2_obj_id = pos_e2_obj_id.cuda()
pos_e2_obj_w = pos_e2_obj_w.cuda()
neg_e1_subj_id = neg_e1_subj_id.cuda()
neg_e1_subj_w = neg_e1_subj_w.cuda()
neg_e1_verb_id = neg_e1_verb_id.cuda()
neg_e1_verb_w = neg_e1_verb_w.cuda()
neg_e1_obj_id = neg_e1_obj_id.cuda()
neg_e1_obj_w = neg_e1_obj_w.cuda()
neg_e2_subj_id = neg_e2_subj_id.cuda()
neg_e2_subj_w = neg_e2_subj_w.cuda()
neg_e2_verb_id = neg_e2_verb_id.cuda()
neg_e2_verb_w = neg_e2_verb_w.cuda()
neg_e2_obj_id = neg_e2_obj_id.cuda()
neg_e2_obj_w = neg_e2_obj_w.cuda()
pos_e1_emb = model(pos_e1_subj_id, pos_e1_subj_w, pos_e1_verb_id,
pos_e1_verb_w, pos_e1_obj_id, pos_e1_obj_w)
pos_e2_emb = model(pos_e2_subj_id, pos_e2_subj_w, pos_e2_verb_id,
pos_e2_verb_w, pos_e2_obj_id, pos_e2_obj_w)
neg_e1_emb = model(neg_e1_subj_id, neg_e1_subj_w, neg_e1_verb_id,
neg_e1_verb_w, neg_e1_obj_id, neg_e1_obj_w)
neg_e2_emb = model(neg_e2_subj_id, neg_e2_subj_w, neg_e2_verb_id,
neg_e2_verb_w, neg_e2_obj_id, neg_e2_obj_w)
if option.distance_metric == 'cosine':
distance_func = cosine_distance
elif option.distance_metric == 'euclid':
distance_func = euclid_distance
pos_dist = distance_func(pos_e1_emb, pos_e2_emb)
neg_dist = distance_func(neg_e1_emb, neg_e2_emb)
num_correct = (pos_dist < neg_dist).sum().item()
accuracy = num_correct / len(dataset)
if option.output_file.strip() != '':
output_file = open(option.output_file, 'w')
for i, j, k in zip(pos_dist, neg_dist, (pos_dist < neg_dist)):
output_file.write(
' '.join([str(i.item(
)), str(j.item()), str(k.item())]) + '\n')
output_file.close()
logging.info('Output saved to ' + option.output_file)
logging.info('Num correct: %d' % (num_correct, ))
logging.info('Num total: %d' % (len(dataset), ))
logging.info('Accuracy: %.4f' % (accuracy, ))
if option.output_vectors is not None:
vectors = torch.stack([pos_e1_emb, pos_e2_emb, neg_e1_emb, neg_e2_emb],
dim=1).cpu()
torch.save(vectors, option.output_vectors)
print('Vectors saved to %s' % (option.output_vectors, ))
from torch.optim import Adam
from trainer import TradeComm
from game import Game
import numpy as np
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--jobnum", default=0)
args = parser.parse_args()
num_items = 15
num_utterances = 15
input_size = 2 * num_items + 2 * num_utterances
hidden_size = 256
num_samples = 10000
epsilon = 1 / 10
policy_weight = 1 / 5
horizon = 2000
write_every = 10
lr = 1e-4
directory = 'results'
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
nn = NN(input_size, hidden_size, num_items, num_utterances)
opt = Adam(nn.parameters(), lr=lr)
g = Game(num_items, num_utterances)
agent = Agent(num_items, num_utterances, nn, opt, num_samples, epsilon,
policy_weight, device)
trainer = TradeComm(g, agent, directory, args.jobnum)
trainer.run(horizon, write_every)
t = np.array(target)[:, np.newaxis]
# Use only the attributes "Petal Length and Petal Width"
X = iris_data.data[:, 2:]
plt.figure(2, figsize=(8, 6))
plt.clf()
# Plot the training points
plt.scatter(X[:, 0], X[:, 1], c=t.flatten())
plt.xlabel('Petal length')
plt.ylabel('Petal width')
plt.title('Real Data', fontsize=16)
plt.show()
# Build neural network
model = NN()
model.add_layer(units=10, activation=np.tanh, initialization=np.random.normal)
model.add_layer(units=5, activation=np.tanh, initialization=np.random.normal)
model.add_layer(units=2, activation=np.tanh, initialization=np.random.normal)
model.add_layer(1, activation=model.sigmoid)
model.train(X, t, iterations=5000, learn_rate=0.0001) # 5000
prediction = model.predict(X)
# Plot error
plt.plot(model.history['epoch'],
model.history['error_train'],
label='Training error')
plt.xlabel('Epoch')
plt.ylabel('Error')
plt.legend()
def main(option):
logging.basicConfig(
stream=sys.stdout,
level=logging.DEBUG,
format=
'%(asctime)s - %(filename)s[line:%(lineno)d] - %(levelname)s: %(message)s'
)
dataset = TransitiveSentenceSimilarityDataset()
if option.dataset_cache is None:
glove = Glove(option.emb_file)
logging.info('Embeddings loaded')
dataset.load(option.dataset_file, glove)
else:
dataset.load_cache(option.dataset_cache)
logging.info('Dataset loaded')
embeddings = nn.Embedding(option.vocab_size, option.emb_dim, padding_idx=1)
if option.model == 'NTN':
model = NeuralTensorNetwork(embeddings, option.em_k)
elif option.model == 'LowRankNTN':
model = LowRankNeuralTensorNetwork(embeddings, option.em_k,
option.em_r)
elif option.model == 'RoleFactor':
model = RoleFactoredTensorModel(embeddings, option.em_k)
elif option.model == 'Predicate':
model = PredicateTensorModel(embeddings)
elif option.model == 'NN':
model = NN(embeddings, 2 * option.em_k, option.em_k)
elif option.model == 'EMC':
model = EMC(embeddings, 2 * option.em_k, option.em_k)
else:
logging.info('Unknown model type: ' + option.model)
exit(1)
checkpoint = torch.load(option.model_file, map_location='cpu')
if type(checkpoint) == dict:
if 'event_model_state_dict' in checkpoint:
state_dict = checkpoint['event_model_state_dict']
else:
state_dict = checkpoint['model_state_dict']
else:
state_dict = checkpoint
model.load_state_dict(state_dict)
logging.info(option.model_file + ' loaded')
# embeddings = nn.Embedding(option.vocab_size, option.emb_dim, padding_idx=1)
# embeddings.weight.data = torch.from_numpy(glove.embd).float()
# model = Averaging(embeddings)
if option.use_gpu:
model.cuda()
model.eval()
data_loader = torch.utils.data.DataLoader(
dataset,
collate_fn=TransitiveSentenceSimilarityDataset_collate_fn,
shuffle=False,
batch_size=len(dataset))
batch = next(iter(data_loader))
e1_subj_id, e1_subj_w, e1_verb_id, e1_verb_w, e1_obj_id, e1_obj_w, \
e2_subj_id, e2_subj_w, e2_verb_id, e2_verb_w, e2_obj_id, e2_obj_w, \
gold = batch
if option.use_gpu:
e1_subj_id = e1_subj_id.cuda()
e1_subj_w = e1_subj_w.cuda()
e1_verb_id = e1_verb_id.cuda()
e1_verb_w = e1_verb_w.cuda()
e1_obj_id = e1_obj_id.cuda()
e1_obj_w = e1_obj_w.cuda()
e2_subj_id = e2_subj_id.cuda()
e2_subj_w = e2_subj_w.cuda()
e2_verb_id = e2_verb_id.cuda()
e2_verb_w = e2_verb_w.cuda()
e2_obj_id = e2_obj_id.cuda()
e2_obj_w = e2_obj_w.cuda()
e1_emb = model(e1_subj_id, e1_subj_w, e1_verb_id, e1_verb_w, e1_obj_id,
e1_obj_w)
e2_emb = model(e2_subj_id, e2_subj_w, e2_verb_id, e2_verb_w, e2_obj_id,
e2_obj_w)
if option.distance_metric == 'cosine':
distance_func = cosine_distance
elif option.distance_metric == 'euclid':
distance_func = euclid_distance
pred = -distance_func(e1_emb, e2_emb)
if option.use_gpu:
pred = pred.cpu()
pred = pred.detach().numpy()
gold = gold.numpy()
spearman_correlation, spearman_p = scipy.stats.spearmanr(pred, gold)
if option.output_file.strip() != '':
output_file = open(option.output_file, 'w')
for score in pred:
output_file.write(str(score) + '\n')
output_file.close()
logging.info('Output saved to ' + option.output_file)
logging.info('Spearman correlation: %.4f' % (spearman_correlation, ))
def main():
train_loader = Data(train_path, shuffle=True, split=0.2)
pca = train_loader.dataloading(batch_size=batch_size,
pca=PCA(n_components=pca_components))
test_loader = Data(test_path, shuffle=False, test=True)
test_loader.dataloading(pca=pca)
model = NN(in_dim=pca_components + 1,
n_cls=3,
neurons=[256],
lr=lr,
hidden_activation='sigmoid',
load_weight=False,
save_weight=False)
info = {
'train_loss': [],
'train_acc': [],
'val_loss': [],
'val_acc': [],
'EPOCHS': EPOCHS
}
for epoch in range(EPOCHS):
cache = np.zeros(4)
itr = 0
for train_data in train_loader:
x_train, y_train, x_val, y_val = train_data
y_train = to_onehot(y_train, 3)
y_val = to_onehot(y_val, 3)
train_loss, train_acc = model.train(x_train, y_train)
val_loss, val_acc, pred = model.predict(x_val, y_val)
cache[0] += train_loss
cache[1] += train_acc
cache[2] += val_loss
cache[3] += val_acc
itr += 1
cache /= itr
info['train_loss'].append(cache[0])
info['train_acc'].append(cache[1])
info['val_loss'].append(cache[2])
info['val_acc'].append(cache[3])
print(
'EPOCH:{:05d}/{:05d} train_loss: {:.5f} train_acc: {:.4f} val_loss: {:.5f} val_acc: {:.4f}'
.format(epoch + 1, EPOCHS, *cache.tolist()))
plot_info(**info)
train_loader.set_batch_size(2000)
for train_data in train_loader:
x_train, y_train, x_val, y_val = train_data
y_train = to_onehot(y_train, 3)
plot_decision_region(x_train,
y_train,
model,
train_loader.CLASSES,
title='training')
for test_data in test_loader:
x_test, y_test = test_data
y_test = to_onehot(y_test, 3)
plot_decision_region(x_test,
y_test,
model,
train_loader.CLASSES,
title='testing')
test_loss, test_acc, pred = model.predict(x_test, y_test)
print('test_loss: {:.5f} test_acc: {:.4f}'.format(test_loss, test_acc))
# act_funcs=[tf.nn.relu for _ in range(layer)],
# batch_size=BATCH_SIZE,
# learning_rate=lr,
# test_samples=test_samples,
# dropout=True,
# batch_norm=False)
#
# fd = 'nn_d/{}l_{}n_{}lr/{}'.format(layer, node, lr, i)
# losses = train_model(model, train_samples, fd)
# loss_data[fd].append(losses)
# training on a generic neural network model with batch norm
model = NN(inputs=N_INPUTS,
hidden_size=[node for _ in range(layer)],
act_funcs=[tf.nn.leaky_relu for _ in range(layer)],
batch_size=BATCH_SIZE,
learning_rate=lr,
test_samples=test_samples,
dropout=False,
batch_norm=True)
fd = 'nn_bn/{}l_{}n_{}lr/{}'.format(layer, node, lr, i)
losses = train_model(model, train_samples, fd)
loss_data[fd].append(losses)
# save the losses to csv
for fd in loss_data.keys():
np.savetxt(os.path.join(fd, 'results.csv'),
np.asarray(loss_data[fd]),
delimiter=',')
from chainer.training import extensions
import numpy as np
import sys
from model import NN, CNN
# 各パラメータの定義
# バッチサイズは、データをまとめる数。
# エポックは、学習回数を表す。
batch_size = 100
epoch_size = 3
modelname = "mnist_NN.h5"
# 学習モデルの定義
model = NN()
if (len(sys.argv) > 1) and (sys.argv[1] == "CNN"):
model = CNN()
modelname = "mnist_CNN.h5"
# データセットの読み込み。
# 今回は、MNIST(エムニスト)という手書き文字データセットを使用。
# 学習データとテストデータで分ける。
# train, testのそれぞれには、１次元配列の画像情報(list)と、
# 画像の答え(int)の２つの情報が入っている。
# image size = 784 -> (1 x 28 x 28)の３次元に指定。
train, test = chainer.datasets.get_mnist(ndim=3)
# 学習データのイテレーションを作成
# バッチサイズ分のデータに区切る。
iterator = chainer.iterators.SerialIterator(train, batch_size)
import gym
from model import NN
import tensorflow as tf
import numpy as np
import random
import math
env = gym.make('Breakout-v0')
nn = NN()
nn2 = NN()
init_op = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init_op)
discount = 0.99
full_reward = []
total_reward = 0
def choose_action(Q):
action = np.argmax(Q)
if epsilon > random.random():
action = env.action_space.sample()
return action
print(env.action_space)
for i_episode in range(1, 2000):
observation = env.reset()
from mcts import MCTS
from zertz.ZertzGame import ZertzGame as Game
from model import NNetWrapper as NN
from config import Config, Config1, Config2
if __name__ == '__main__':
# Game settings
rings = 19
marbles = {'w': 10, 'g': 10, 'b': 10}
win_con = [{'w': 2}, {'g': 2}, {'b': 2}, {'w': 1, 'g': 1, 'b': 1}]
t = 5
# Setup
game = Game(rings, marbles, win_con, t)
config = Config()
nnet = NN(game, config)
# Option #1: Learn
trainer = Individual(game, nnet, config)
trainer.learn()
# Option #2: Human vs AI
#nnet.load_checkpoint(filename='checkpoint_32_10_0001_29.pth.tar')
#ai_agent = MCTS(game, nnet, config.c_puct, config.num_sims)
#hp = HumanPlay(game, ai_agent)
#hp.play()
# Option #3: AI vs AI
#config1 = Config1()
#config2 = Config2()
#nnet1 = NN(game, config1)
def __len__(self):
return self.n_samples
# Hyperparameters
batch_size = 8
learning_rate = 0.001
num_epochs = 1000
hidden_size = 8
output_size = len(tags)
input_size = len(X_train[0])
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
dataset = ChatDataset()
train_loader = DataLoader(dataset=dataset, batch_size=batch_size, shuffle=True, num_workers=2)
model = NN(input_size, hidden_size, output_size).to(device)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
start_t = time()
for epoch in range(num_epochs):
for (words, labels) in train_loader:
words = words.to(device)
labels = labels.to(device)
#forward
outputs = model(words)
loss = criterion(outputs, labels)
#backward and optimizer step
print('Early stopping')
break
last_loss_test = loss_test
print('epoch {}, loss - {}'.format(epoch, loss))
print('test loss - {}'.format(loss_test))
optimizer.zero_grad()
loss.backward()
optimizer.step()
if __name__ == '__main__':
data = pd.read_csv('datasets/anchor-weed.csv')
scaler.fit_transform(data[['price']])
np.random.seed(100)
nr_classes = 6
hidden_size = 12
data = pd.read_csv('datasets/anchor-weed-prepared.csv')
data = data.drop('Unnamed: 0', axis=1)
y = data['price+1'].values
X = data.drop('price+1', axis=1).values
model = NN([nr_classes, 8, 4, 1])
train(model, X, y, 30000, 0.01)
torch.save(model.state_dict(), 'model_state/state')
y_pred = model.forward(torch.from_numpy(X[275, :]).float())
y_pred = np.exp(y_pred.detach().numpy().reshape(-1, 1))
y_ = np.exp(y[275])
plot_results(model, X, y)
fisher_size = 1
loss_option = 0
fisher_true = False
fisher_block = False
fisher_diagonal = False
# experiment 1: CNN on permuted MNIST
learning_rate = 100
fisher_multiplier = 10
input_size = 50
output_size = 1
hidden_units = 25
# generate data
train_data,P = data.generate_data(num_tasks_test,samples_train,input_size)
test_data,_ = data.generate_data(num_tasks_test,samples_test,input_size)
# generate model
model = NN(input_size,output_size,hidden_units=hidden_units)
def main():
experiment(P,fisher_multiplier,learning_rate,model,train_data,test_data,num_tasks_train,num_tasks_test,iterations,batch_size,log_period_updates,fisher_size,loss_option=loss_option,fisher_true=fisher_true,fisher_block=fisher_block,fisher_diagonal=fisher_diagonal,verbose=True)
print(tf.trainable_variables())
return
if __name__ == "__main__":
main()
batch_size=1,
shuffle=False)
nb_teams = next(iter(train_loader))[0].size(1) // 2
print(nb_teams)
training_loss_history = []
test_loss_history = []
test_accuracy_history = []
########## Train the model ##########
model = NN(nb_teams=nb_teams,
learning_rate=CHOSEN_LEARNING_RATE,
hidden_layer_size1=CHOSEN_HIDDEN_LAYER_SIZES[0],
hidden_layer_size2=CHOSEN_HIDDEN_LAYER_SIZES[1],
d_ratio=CHOSEN_DROPOUT_RATE)
best_loss = (1000, 0)
best_accuracy = (0, 0)
for epoch in tqdm(range(CHOSEN_EPOCH)):
losses_training = []
for data, target in train_loader:
data = Variable(data)
target = Variable(target)
loss = model.step(data, target)
losses_training.append(loss)
# Mean of the losses of training predictions
import torch as tc
from model import NN
from champion_dictionary import *
net = NN()
net.load_state_dict(tc.load("model.wab"))
input_data = tc.rand((10), dtype=tc.float)
out = net(input_data)
print(input_data)
if out < 0.5:
print("Blue team")
print(out)
else:
print("Red team")
print(out)
def transform_image(image_str):
my_transforms = transforms.Compose([
transforms.Resize((448, 448)),
transforms.Grayscale(),
transforms.ToTensor(),
transforms.Normalize(mean = [0.9823], std = [0.0758]),
])
cv2_image = data_uri_to_cv2_img(image_str)
image = Image.fromarray(cv2_image)
# image = Image.open(StringIO(image_bytes))
return my_transforms(image).unsqueeze(0)
PATH = 'models/model_feature_predictor.ckpt'
model = NN.load_from_checkpoint(PATH)
pretrained_model = model_gan.model_gan
pretrained_model.load_state_dict(torch.load('models/model_sketch_simplification.pth'))
feature_map = {
1: "Cross, Upper left corner, outside rectangle",
2: "Large Rectangle",
3: "Diagonal Cross",
4: "Horizontal midline of 2",
5: "Vertical Midline",
6: "Small rectangle within 2 to the left",
7: "Small segment above 6",
8: "Four parallel lines within 2, upper left",
9: "Triangle above 2, upper right",
10: "Small vertical line within 2, below 9",
def train(config):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Initialize the device which to run the model on
device = torch.device(device)
# Initialize the dataset and data loader (note the +1)
dataset = IMDBDataset(train_or_test='train', seq_length=config.seq_length)
data_loader = DataLoader(dataset,
config.batch_size,
shuffle=True,
num_workers=4)
# Initialize the dataset and data loader (note the +1)
test_dataset = IMDBDataset(train_or_test='test',
seq_length=config.seq_length)
test_data_loader = DataLoader(test_dataset,
config.batch_size,
shuffle=True,
num_workers=4)
# Initialize the model that we are going to use
if not (config.recurrent_dropout_model):
model = NN(dataset.vocab_size, config.embed_dim, config.hidden_dim,
config.output_dim, config.n_layers, config.bidirectional,
config.dropout, 0).to(device)
else:
model = Model(dataset.vocab_size,
output_dim=config.output_dim).to(device)
if not os.path.exists(f'runs/{config.name}'):
os.makedirs(f'runs/{config.name}')
print(config.__dict__)
with open(f'runs/{config.name}/args.txt', 'w') as f:
json.dump(config.__dict__, f, indent=2)
# Setup the loss and optimizer
criterion = nn.CrossEntropyLoss().to(device)
# criterion = torch.nn.MSELoss().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=config.learning_rate)
lowest = 100
save = []
epochs = 0
while epochs < config.train_epochs:
accuracies = []
losses = []
print('Training')
for step, (batch_inputs, batch_targets) in enumerate(data_loader):
x = batch_inputs.long().to(device)
y_target = batch_targets.long().to(device)
predictions = model(x)
loss = criterion(predictions, y_target)
optimizer.zero_grad()
loss.backward()
# torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=config.max_norm)
optimizer.step()
accuracy = (torch.argmax(predictions,
dim=1) == y_target).cpu().numpy().mean()
loss = loss.item()
accuracies.append(accuracy)
losses.append(loss)
accuracy = np.array(accuracies).mean()
loss = np.array(losses).mean()
# Test on test set
print('Testing')
with torch.no_grad():
test_accuracies = []
test_losses = []
for step, (batch_inputs,
batch_targets) in enumerate(test_data_loader):
x = batch_inputs.long().to(device)
y_target = batch_targets.long().to(device)
predictions = model(x)
test_loss = criterion(predictions, y_target)
test_accuracy = (torch.argmax(
predictions, dim=1) == y_target).cpu().numpy().mean()
test_loss = test_loss.item()
test_accuracies.append(test_accuracy)
test_losses.append(test_loss)
test_accuracy = np.array(test_accuracies).mean()
test_loss = np.array(test_losses).mean()
if (test_loss < lowest):
lowest = test_loss
torch.save(model.state_dict(), f'runs/{config.name}/model.pt')
epochs += 1
print(
"[{}] Train epochs {:04d}/{:04d}, Train Accuracy = {:.2f}, Train Loss = {:.3f}, Test Accuracy = {:.2f}, Test Loss = {:.3f}"
.format(datetime.now().strftime("%Y-%m-%d %H:%M"), epochs,
config.train_epochs, accuracy, loss, test_accuracy,
test_loss))
print('Done training.')
return accuracy, lowest, save
# 必要なものをインポート import chainer from chainer import training, Variable from chainer import datasets, iterators, optimizers, serializers from chainer.training import extensions import numpy as np import sys from PIL import Image, ImageOps from model import NN, CNN # 各パラメータ model = NN() image_name = "" CNN_flag = False Invert = False # コマンドライン引数を設定 def set_args(): global CNN_flag, image_name, Invert if (len(sys.argv) > 1): if(sys.argv[1] == "CNN"): CNN_flag = True if (len(sys.argv) > 2): image_name = sys.argv[2] if (len(sys.argv) > 3): Invert = True
def __getitem__(self, index):
return self.x_train[index], self.y_train[index]
def __len__(self):
return len(self.x_train)
dataset = Data()
train_loader = DataLoader(dataset=dataset,
batch_size=32,
shuffle=True,
num_workers=0)
#Declaring network and optimizer
print("Initializing network")
network = NN()
optimizer = optim.Adam(network.parameters(), lr=0.0005)
criterion = nn.BCELoss(reduction='mean')
correct = 0
for epoch in range(5):
for i, data in enumerate(train_loader, 0):
# get the inputs
inputs, labels = data
# Forward pass: Compute predicted y by passing x to the model
y_pred = network(inputs)
# Compute and print loss
loss = criterion(y_pred, labels.view((-1, 1)))
print(f'Epoch {epoch + 1} | Batch: {i+1} | Loss: {loss.item():.4f}')
dataset = TransitiveSentenceSimilarityDataset(option.dataset_file, glove)
logging.info('Dataset loaded')
embeddings = nn.Embedding(option.vocab_size, option.emb_dim, padding_idx=1)
if option.model == 'NTN':
model = NeuralTensorNetwork(embeddings, option.em_k)
elif option.model == 'LowRankNTN':
model = LowRankNeuralTensorNetwork(embeddings, option.em_k,
option.em_r)
elif option.model == 'RoleFactor':
model = RoleFactoredTensorModel(embeddings, option.em_k)
elif option.model == 'Predicate':
model = PredicateTensorModel(embeddings)
elif option.model == 'NN':
model = NN(embeddings, 2 * option.em_k, option.em_k)
elif option.model == 'EMC':
model = EMC(embeddings, 2 * option.em_k, option.em_k)
else:
logging.info('Unknown model type: ' + option.model)
exit(1)
checkpoint = torch.load(option.model_file)
if type(checkpoint) == dict:
if 'event_model_state_dict' in checkpoint:
state_dict = checkpoint['event_model_state_dict']
else:
state_dict = checkpoint['model_state_dict']
else:
state_dict = checkpoint
model.load_state_dict(state_dict)
batch_size=32,
shuffle=True)
classes = train.classes
device = torch.device("cuda:0")
load_weights = False
#
if use_pretrained_resnet:
nn_model = models.resnet50(pretrained=True)
num_ftr = nn_model.fc.in_features
nn_model.fc = nn.Linear(num_ftr, 2)
nn_model.to(device)
else:
nn_model = NN()
if load_weights:
print("Found State Dict. Loading...")
with open(r'state/state_dict.pickle', 'rb') as file:
state_dict = torch.load(r'state/state_dict.pickle')
nn_model.load_state_dict(state_dict)
nn_model.to(device)
learning_rate = 0.0001
f1_scores = []
epochs = 1
for epoch in range(epochs):
predicted_cumulated, labels_cumulated = np.array([]), np.array([])
running_loss = 0
counter = 0
daily_return = reduce(lambda left, right: pd.concat([left, right], axis=1),
df_list)
daily_return = np.mean(daily_return, axis=1).fillna(0)
daily_return = daily_return.fillna(0)
data = DataSplit(instance.unprep_data, daily_return, 20100101, 20180630,
20181231, 20190630, 7)
# In[5]:
#######################################################################################
#### Model Framework ################################################################
#######################################################################################
tot_mod = Model(data) # -- PARENT CLASS of Models
benchmark_model = BenchPredict(data_model=data, lb=28,
la=1) # -- BENCHMARK CLASS
NN_model = NN(data_model=data, lb=28, la=1) # -- NEURAL NETWORK CLASS
RNN_model = RNN(data_model=data, lb=28, la=1) # -- RNN CLASS
ARMA_data = DataSplit(instance.unprep_data, instance.unprep_data['return'],
20100101, 20180630, 20181231, 20190630,
7) # -- ARMA DATA PROCESS
ARMA_model = Arma(data_model=ARMA_data) # -- ARMA CLASS
Embed_model = combine_model(RNN_model, ARMA_model)
# In[9]:
try:
get_ipython().system('jupyter nbconvert --to python constants.ipynb')
except:
pass
# In[ ]:
def test(config):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Initialize the device which to run the model on
device = torch.device(device)
with open(f'runs/{config.name}/args.txt', 'r') as f:
config.__dict__ = json.load(f)
# Initialize the dataset and data loader (note the +1)
test_dataset = IMDBDataset(train_or_test='test',
seq_length=config.seq_length)
test_data_loader = DataLoader(test_dataset,
batch_size=config.batch_size,
shuffle=True,
num_workers=4)
# Initialize the model that we are going to use
if not (config.recurrent_dropout_model):
model = NN(test_dataset.vocab_size, config.embed_dim,
config.hidden_dim, config.output_dim, config.n_layers,
config.bidirectional, config.dropout, 0).to(device)
model.load_state_dict(torch.load(f'runs/{config.name}/model.pt'))
else:
model = Model(test_dataset.vocab_size,
output_dim=config.output_dim).to(device)
model.load_state_dict(torch.load(f'runs/{config.name}/model.pt'))
# Setup the loss and optimizer
criterion = torch.nn.MSELoss().to(device)
lowest = 100
save = []
epochs = 0
num_steps = math.floor(25000 / config.batch_size)
d = {'target': [], 'mean0': [], 'std0': [], 'mean1': [], 'std1': []}
with torch.no_grad():
for step, (batch_inputs, batch_targets) in enumerate(test_data_loader):
x = batch_inputs.long().to(device)
y_target = batch_targets.long().to(device)
# print('\n*************************************************\n')
# print(test_dataset.convert_to_string(x[0].tolist()))
preds = torch.zeros((100, x.shape[0], config.output_dim))
for i in range(config.B):
preds[i, :, :] = model(x)
# print('\n')
if step % 1 == 0:
print(step, '/', num_steps)
mean = preds.mean(dim=0)
std = preds.std(dim=0)
d['target'].extend(batch_targets.tolist())
d['mean0'].extend(mean[:, 0].tolist())
d['std0'].extend(std[:, 0].tolist())
d['mean1'].extend(mean[:, 1].tolist())
d['std1'].extend(std[:, 1].tolist())
pd.DataFrame(d).to_csv(f'runs/{config.name}/results.csv')
return 'joe'
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
df = pd.read_csv("breast-cancer-wisconsin.data",
index_col=0,
na_values=['?'],
header=None)
df.columns = [1, 2, 3, 4, 5, 6, 7, 8, 9, "Diagnosis"]
df.index.name = "ID"
df = df.fillna(df.mean())
X = df[[1, 2, 3, 4, 5, 6, 7, 8, 9]].to_numpy()
Y = df[["Diagnosis"]].to_numpy() # - 2) / 2
nn_model = NN([9, 9, 1]) # 10, 10, 10, 10, 1])
nn_model.import_set(X, Y, normalization=True, ratio=[7, 2, 1])
theta, cost_hist = nn_model.train_model((3, 1), max_iter=10000)
training_score = nn_model.cost_function(theta, nn_model.sets['train'])
cv_score = nn_model.cost_function(theta, nn_model.sets['cv'])
test_score = nn_model.cost_function(theta, nn_model.sets['test'])
print(nn_model.f1_score(theta, nn_model.sets['train']))
print(nn_model.f1_score(theta, nn_model.sets['cv']))
print(nn_model.f1_score(theta, nn_model.sets['test']))
print("training cost:", training_score)
print("cross validation cost:", cv_score)
print("test cost:", test_score)
import cv2
from model import NN
import numpy as np
import torch
cap = cv2.VideoCapture(0)
i = 0
classify = 1
labels =[]
Model = NN(batch_size = 1)
Model.load_state_dict(torch.load("1"))
Model.eval()
tardict = {1 : 'Face Detected' , 0 : 'Undetected' }
while True:
i += 1
ret , frame = cap.read()
gray = cv2.cvtColor(frame , cv2.COLOR_RGB2GRAY)
gray = cv2.GaussianBlur(gray, (15,15), 0)
cv2.imshow('feed' , frame)
gray = torch.from_numpy(gray).view(1 , 1, 480 , 640).float()
output = torch.round(Model.forward(gray))
output = output.item()
print (tardict[output])
if output != 0:
input()
if cv2.waitKey(1) & 0xFF == ord('q') :
break
