if __name__ == "__main__":
from q71 import SingleLayerNN
from q73 import load_dataloader
from q75 import train_model
batch_size_list = [2**i for i in range(20)]
for batch_size in batch_size_list:
dataloader_train = load_dataloader("train.data", batch_size=batch_size)
dataloader_valid = load_dataloader("valid.data", batch_size=512)
model = SingleLayerNN(300, 4)
model, *_ = train_model(model,
dataloader_train,
dataloader_valid,
lr=1e-2,
epochs=100,
take_time=True)
conc = self.dropout(conc)
out = self.out(conc)
return out
if __name__ == "__main__":
import torch
from torch import nn
from q73 import load_dataloader
from q75 import train_model
n_epochs = 6
batch_size = 512
load_corpus()
train_loader = load_dataloader("train2.data", batch_size=batch_size)
valid_loader = load_dataloader("valid2.data", batch_size=batch_size)
model = BiLSTM()
loss_fn = nn.CrossEntropyLoss(reduction="sum")
optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr=0.001)
model.cuda()
train_loss = []
valid_loss = []
for epoch in range(n_epochs):
# 訓練モード
model.train()
avg_loss = 0.
for i, (x_batch, y_batch) in enumerate(train_loader):
y_pred = model(x_batch)