95 lines
3.0 KiB
Python
95 lines
3.0 KiB
Python
import numpy as np
|
|
import torch
|
|
from PIL.Image import Image
|
|
|
|
import DataLoader
|
|
from DataLoader import get_image_loader
|
|
from Net import ImageNN
|
|
|
|
# 01.05.22 -- 0.5h
|
|
from netio import save_model, load_model, eval_evalset
|
|
|
|
|
|
def get_train_device():
|
|
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
|
print(f'Device to train net: {device}')
|
|
return torch.device(device)
|
|
|
|
|
|
def train_model():
|
|
# Set a known random seed for reproducibility
|
|
np.random.seed(0)
|
|
torch.manual_seed(0)
|
|
device = get_train_device()
|
|
|
|
# Load datasets
|
|
train_loader, test_loader = get_image_loader("training/", precision=np.float32)
|
|
nn = ImageNN(n_in_channels=3, precision=np.float32) # todo net params
|
|
nn.train() # init with train mode
|
|
nn.to(device) # send net to device available
|
|
|
|
optimizer = torch.optim.AdamW(nn.parameters(), lr=0.1, weight_decay=1e-5) # todo adjust parameters and lr
|
|
loss_function = torch.nn.MSELoss()
|
|
loss_function.to(device)
|
|
n_epochs = 7 # todo epcchs here
|
|
|
|
train_sample_size = len(train_loader)
|
|
losses = []
|
|
best_eval_loss = np.inf
|
|
for epoch in range(n_epochs):
|
|
print(f"Epoch {epoch}/{n_epochs}\n")
|
|
i = 0
|
|
for input_tensor, target_tensor in train_loader:
|
|
output = nn(input_tensor.to(device)) # get model output (forward pass)
|
|
|
|
loss = loss_function(output.to(device), target_tensor.to(device)) # compute loss given model output and true target
|
|
loss.backward() # compute gradients (backward pass)
|
|
optimizer.step() # perform gradient descent update step
|
|
optimizer.zero_grad() # reset gradients
|
|
losses.append(loss.item())
|
|
|
|
i += train_loader.batch_size
|
|
print(
|
|
f'\rTraining epoch {epoch} [{i}/{train_sample_size * train_loader.batch_size}] (curr loss: {loss.item():.3})',
|
|
end='')
|
|
|
|
# eval model every 3000th sample
|
|
if i % 3000 == 0:
|
|
print(f"\nEvaluating model")
|
|
eval_loss = eval_model(nn, test_loader, loss_function, device)
|
|
print(f"Evalution loss={eval_loss}")
|
|
if eval_loss < best_eval_loss:
|
|
best_eval_loss = eval_loss
|
|
save_model(nn)
|
|
|
|
nn.train()
|
|
|
|
# evaluate model with submission pkl file
|
|
eval_evalset()
|
|
|
|
|
|
# func to evaluate our trained model
|
|
def eval_model(model: torch.nn.Module, dataloader: torch.utils.data.DataLoader, loss_fn, device: torch.device):
|
|
# switch to eval mode
|
|
model.eval()
|
|
loss = .0
|
|
# disable gradient calculations
|
|
with torch.no_grad():
|
|
i = 0
|
|
for input, target in dataloader:
|
|
input = input.to(device)
|
|
target = target.to(device)
|
|
|
|
out = model(input)
|
|
loss += loss_fn(out, target).item()
|
|
print(f'\rEval prog[{i}/{len(dataloader) * dataloader.batch_size}]', end='')
|
|
i += dataloader.batch_size
|
|
print()
|
|
loss /= len(dataloader)
|
|
return loss
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
train_model()
|