标签：loss train metric 对比 self valid LR test FM
%%time
from utils.utils import create_dataset, Trainer
from layer.layer import Embedding, FeaturesEmbedding, EmbeddingsInteraction, MultiLayerPerceptron
import torch
import torch.nn as nn
import torch.optim as optim
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print('Training on [{}].'.format(device))
Training on [cpu].
CPU times: user 2.59 s, sys: 1.43 s, total: 4.03 s
Wall time: 1.95 s
%%time
dataset = create_dataset('lrmovielens100k', read_part=False, sample_num=1000000, device=device)
field_dims, (train_X, train_y), (valid_X, valid_y), (test_X, test_y) = dataset.train_valid_test_split()
CPU times: user 933 ms, sys: 30.5 ms, total: 964 ms
Wall time: 1.38 s
class FactorizationMachine(nn.Module):
def __init__(self, field_dims, embed_dim=4):
super(FactorizationMachine, self).__init__()
self.embed1 = FeaturesEmbedding(field_dims, 1)
self.embed2 = FeaturesEmbedding(field_dims, embed_dim)
self.bias = nn.Parameter(torch.zeros((1, )))
def forward(self, x):
# x shape: (batch_size, num_fields)
# embed(x) shape: (batch_size, num_fields, embed_dim)
square_sum = self.embed2(x).sum(dim=1).pow(2).sum(dim=1)
sum_square = self.embed2(x).pow(2).sum(dim=1).sum(dim=1)
output = self.embed1(x).squeeze(-1).sum(dim=1) + self.bias + (square_sum - sum_square) / 2
output = torch.sigmoid(output).unsqueeze(-1)
return output
%%time
EMBEDDING_DIM = 8
LEARNING_RATE = 1e-4
REGULARIZATION = 1e-6
BATCH_SIZE = 4096
EPOCH = 600
TRIAL = 100
fm = FactorizationMachine(field_dims, EMBEDDING_DIM).to(device)
optimizer = optim.Adam(fm.parameters(), lr=LEARNING_RATE, weight_decay=REGULARIZATION)
criterion = nn.BCELoss()
trainer = Trainer(fm, optimizer, criterion, BATCH_SIZE)
trainer.train(train_X, train_y, epoch=EPOCH, trials=TRIAL, valid_X=valid_X, valid_y=valid_y)
test_loss, test_metric = trainer.test(test_X, test_y)
print('test_loss: {:.5f} | test_metric: {:.5f}'.format(test_loss, test_metric))
68%|█████████████████████████████████████████▊ | 411/600 [03:54<01:47, 1.75it/s]
train_loss: 0.66286 | train_metric: 0.60023
valid_loss: 0.66593 | valid_metric: 0.59360
test_loss: 0.66421 | test_metric: 0.59830
CPU times: user 15min 34s, sys: 1.09 s, total: 15min 35s
Wall time: 3min 55s
from sklearn import metrics
test_loss, test_metric = trainer.test(test_X, test_y, metricer=metrics.precision_score)
print('test_loss: {:.5f} | test_metric: {:.5f}'.format(test_loss, test_metric))
test_loss: 0.66421 | test_metric: 0.61538
class LogisticRegression(nn.Module):
def __init__(self, field_dims):
super(LogisticRegression, self).__init__()
self.bias = nn.Parameter(torch.zeros((1, )))
self.embed = FeaturesEmbedding(field_dims, 1)
def forward(self, x):
# x shape: (batch_size, num_fields)
output = self.embed(x).sum(dim=1) + self.bias
output = torch.sigmoid(output)
return output
%%time
LEARNING_RATE = 1e-4
REGULARIZATION = 1e-6
BATCH_SIZE = 4096
EPOCH = 600
TRIAL = 100
lr = LogisticRegression(field_dims).to(device)
optimizer = optim.Adam(lr.parameters(), lr=LEARNING_RATE, weight_decay=REGULARIZATION)
criterion = nn.BCELoss()
trainer = Trainer(lr, optimizer, criterion, BATCH_SIZE)
trainer.train(train_X, train_y, epoch=EPOCH, trials=TRIAL, valid_X=valid_X, valid_y=valid_y)
test_loss, test_metric = trainer.test(test_X, test_y)
print('test_loss: {:.5f} | test_metric: {:.5f}'.format(test_loss, test_metric))
38%|███████████████████████▍ | 230/600 [00:21<00:34, 10.76it/s]
train_loss: 0.67392 | train_metric: 0.57983
valid_loss: 0.67499 | valid_metric: 0.58050
test_loss: 0.67422 | test_metric: 0.57600
CPU times: user 1min 25s, sys: 239 ms, total: 1min 25s
Wall time: 21.6 s
from sklearn import metrics
test_loss, test_metric = trainer.test(test_X, test_y, metricer=metrics.precision_score)
print('test_loss: {:.5f} | test_metric: {:.5f}'.format(test_loss, test_metric))
test_loss: 0.67422 | test_metric: 0.58674
标签：loss,
train,
metric,
对比,
self,
valid,
LR,
test,
FM
From： https://www.cnblogs.com/RanX2018/p/17275802.html