PyTorch搭建多项式回归模型（三）

def count(firstval=0, step=1):
 x = firstval
 while 1:
 yield x
 x += step

from itertools import count
import torch
import torch.autograd
import torch.nn.functional as F
 
POLY_DEGREE = 3

def make_features(x):
 """Builds features i.e. a matrix with columns [x, x^2, x^3, x^4]."""
 x = x.unsqueeze(1)
 return torch.cat([x ** i for i in range(1, POLY_DEGREE+1)], 1)

W_target = torch.randn(POLY_DEGREE, 1)
b_target = torch.randn(1)
 
def f(x):
 """Approximated function."""
 return x.mm(W_target) + b_target.item()

def get_batch(batch_size=32):
 """Builds a batch i.e. (x, f(x)) pair."""
 random = torch.randn(batch_size)
 x = make_features(random)
 y = f(x)
 return x, y

# Define model
fc = torch.nn.Linear(W_target.size(0), 1)

for batch_idx in count(1):
 # Get data
 batch_x, batch_y = get_batch()
 
 # Reset gradients
 fc.zero_grad()
 
 # Forward pass
 output = F.smooth_l1_loss(fc(batch_x), batch_y)
 loss = output.item()
 
 # Backward pass
 output.backward()
 
 # Apply gradients
 for param in fc.parameters():
 param.data.add_(-0.1 * param.grad.data)
 
 # Stop criterion
 if loss < 1e-3:
 break

def poly_desc(W, b):
 """Creates a string description of a polynomial."""
 result = 'y = '
 for i, w in enumerate(W):
 result += '{:+.2f} x^{} '.format(w, len(W) - i)
 result += '{:+.2f}'.format(b[0])
 return result
 
print('Loss: {:.6f} after {} batches'.format(loss, batch_idx))
print('==> Learned function:\t' + poly_desc(fc.weight.view(-1), fc.bias))
print('==> Actual function:\t' + poly_desc(W_target.view(-1), b_target))