wassname/ETSformer
1
0
Fork 0
mirror of https://github.com/wassname/ETSformer.git synced 2026-06-27 16:59:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

multivar 2 single, an plot

Browse Source
This commit is contained in:
wassname
2022-11-28 14:10:49 +08:00
parent 244766827e
commit 0bf1606713
8 changed files with 1509 additions and 229 deletions
Download Patch File Download Diff File Expand all files Collapse all files
data_provider/data_loader.py
+14 -1
View File
@@ -229,6 +229,7 @@ class Dataset_Custom(Dataset):
cols = list(df_raw.columns)
cols.remove(self.target)
cols.remove('date')
self.cols = ['date'] + cols + [self.target]
df_raw = df_raw[['date'] + cols + [self.target]]
# print(cols)
num_train = int(len(df_raw) * 0.7)
@@ -244,6 +245,11 @@ class Dataset_Custom(Dataset):
df_data = df_raw[cols_data]
elif self.features == 'S':
df_data = df_raw[[self.target]]
elif self.features == 'M2S':
df_data = df_raw[cols + [self.target]]
self.n_dims = 1 # len(cols + [self.target])
else:
raise NotImplementedError(self.features)
if self.scale:
train_data = df_data[border1s[0]:border2s[0]]
@@ -265,8 +271,15 @@ class Dataset_Custom(Dataset):
data_stamp = data_stamp.transpose(1, 0)
self.data_x = data[border1:border2]
self.data_y = data[border1:border2]
self.dates = df_raw['date'][border1:border2]
# self.data_y = data[border1:border2]
# y is just the col we predict
self.data_y = data[border1:border2][:, [-1]]
self.data_stamp = data_stamp
# TODO check this is right. For example check that it makes the df
o = self.seq_len - self.label_len # + self.label_len
self.index = self.dates.iloc[o:].iloc[:len(self)]
def __getitem__(self, index):
s_begin = index
exp/exp_main.py
+52 -44
View File
@@ -5,6 +5,8 @@ from utils.tools import EarlyStopping, adjust_learning_rate
from utils.metrics import metric
from utils.Adam import Adam
from tqdm.auto import tqdm
import numpy as np
import torch
import torch.nn as nn
@@ -121,58 +123,64 @@ class Exp_Main(Exp_Basic):
model_optim = self._select_optimizer()
criterion = self._select_criterion()
for epoch in range(self.args.train_epochs):
iter_count = 0
train_loss = []
with tqdm(unit='epoch', total=self.args.train_epochs) as p:
for epoch in range(self.args.train_epochs):
iter_count = 0
train_loss = []
self.model.train()
epoch_time = time.time()
for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(train_loader):
iter_count += 1
model_optim.zero_grad()
batch_x = batch_x.float().to(self.device)
self.model.train()
epoch_time = time.time()
for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(tqdm(train_loader, leave=False, desc='train')):
iter_count += 1
model_optim.zero_grad()
batch_x = batch_x.float().to(self.device)
batch_y = batch_y.float().to(self.device)
batch_x_mark = batch_x_mark.float().to(self.device)
batch_y_mark = batch_y_mark.float().to(self.device)
batch_y = batch_y.float().to(self.device)
batch_x_mark = batch_x_mark.float().to(self.device)
batch_y_mark = batch_y_mark.float().to(self.device)
# decoder input
dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
# decoder input
dec_inp = torch.zeros_like(batch_y[:, -self.args.pred_len:, :]).float()
dec_inp = torch.cat([batch_y[:, :self.args.label_len, :], dec_inp], dim=1).float().to(self.device)
# encoder - decoder
outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
# encoder - decoder
outputs = self.model(batch_x, batch_x_mark, dec_inp, batch_y_mark)
f_dim = -1 if self.args.features == 'MS' else 0
batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
loss = criterion(outputs, batch_y)
train_loss.append(loss.item())
f_dim = -1 if self.args.features == 'MS' else 0
batch_y = batch_y[:, -self.args.pred_len:, f_dim:].to(self.device)
loss = criterion(outputs, batch_y)
train_loss.append(loss.item())
if (i + 1) % 100 == 0:
print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
speed = (time.time() - time_now) / iter_count
left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
iter_count = 0
time_now = time.time()
# if (i + 1) % 10 == 0:
# print("\titers: {0}, epoch: {1} | loss: {2:.7f}".format(i + 1, epoch + 1, loss.item()))
# speed = (time.time() - time_now) / iter_count
# left_time = speed * ((self.args.train_epochs - epoch) * train_steps - i)
# print('\tspeed: {:.4f}s/iter; left time: {:.4f}s'.format(speed, left_time))
# iter_count = 0
# time_now = time.time()
# p.desc = f'loss: {loss.item():.7f}'
loss.backward()
torch.nn.utils.clip_grad_norm(self.model.parameters(), 1.0)
model_optim.step()
loss.backward()
torch.nn.utils.clip_grad_norm(self.model.parameters(), 1.0)
model_optim.step()
print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
train_loss = np.average(train_loss)
vali_loss = self.vali(vali_data, vali_loader, criterion)
test_loss = self.vali(test_data, test_loader, criterion)
print("Epoch: {} cost time: {}".format(epoch + 1, time.time() - epoch_time))
train_loss = np.average(train_loss)
vali_loss = self.vali(vali_data, vali_loader, criterion)
test_loss = self.vali(test_data, test_loader, criterion)
print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
epoch + 1, train_steps, train_loss, vali_loss, test_loss))
early_stopping(vali_loss, self.model, path)
if early_stopping.early_stop:
print("Early stopping")
break
p.update()
p.desc = f'Train Loss: {test_loss:.7f} Vali Loss: {vali_loss:.7f}'
print("Epoch: {0}, Steps: {1} | Train Loss: {2:.7f} Vali Loss: {3:.7f} Test Loss: {4:.7f}".format(
epoch + 1, train_steps, train_loss, vali_loss, test_loss))
early_stopping(vali_loss, self.model, path)
if early_stopping.early_stop:
print("Early stopping")
break
adjust_learning_rate(model_optim, epoch + 1, self.args)
adjust_learning_rate(model_optim, epoch + 1, self.args)
best_model_path = path + '/' + 'checkpoint.pth'
self.model.load_state_dict(torch.load(best_model_path))
@@ -191,7 +199,7 @@ class Exp_Main(Exp_Basic):
self.model.eval()
with torch.no_grad():
for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(test_loader):
for i, (batch_x, batch_y, batch_x_mark, batch_y_mark) in enumerate(tqdm(test_loader, leave=False)):
batch_x = batch_x.float().to(self.device)
batch_y = batch_y.float().to(self.device)
@@ -237,4 +245,4 @@ class Exp_Main(Exp_Basic):
np.save(folder_path + 'pred.npy', preds)
np.save(folder_path + 'true.npy', trues)
return
return preds, trues
mjc_notes.md
+3 -3
View File
@@ -2,6 +2,6 @@
TODO:
- [ ] run on stocks data
- [ ] multivariate in, univariate out
- [ ] graph
- [x] run on stocks data
- [x] multivariate in, univariate out
- [x] graph
models/etsformer/encoder.py
+4
View File
@@ -183,6 +183,10 @@ class EncoderLayer(nn.Module):
self.dropout2 = nn.Dropout(dropout)
def forward(self, res, level, attn_mask=None):
# when c_in!=c_out assume target columns are at end of channels
level = level[:, :, -self.c_out:]
season, season_attn = self._season_block(res)
res = res - season[:, :-self.pred_len]
growth, growth_attn = self._growth_block(res)
models/etsformer/model.py
+1
View File
@@ -73,6 +73,7 @@ class ETSformer(nn.Module):
if self.training:
x_enc = self.transform.transform(x_enc)
res = self.enc_embedding(x_enc)
level, growths, seasons, season_attns, growth_attns = self.encoder(res, x_enc, attn_mask=enc_self_mask)
growth, season, growth_dampings = self.decoder(growths, seasons)
notebooks/001_001_run.ipynb
+1435 -181
View File
File diff suppressed because one or more lines are too long
notebooks/results/Exchange_ETSformer_custom_ftM_sl128_pl48_dm512_nh8_el2_dl2_df2048_K0_lr0.001_'Exp'_0/test_metrics.npy
BIN
View File
Binary file not shown.
notebooks/results/Exchange_ETSformer_custom_ftM_sl128_pl48_dm512_nh8_el2_dl2_df2048_K0_lr0.001_'Exp'_0/val_metrics.npy
BIN
View File
Binary file not shown.