%matplotlib inline
from fastai.basics import *
Get the 'pickled' MNIST dataset from http://deeplearning.net/data/mnist/mnist.pkl.gz. We're going to treat it as a standard flat dataset with fully connected layers, rather than using a CNN.
!mkdir data/mnist
!rm data/mnist.pkl.gz
!wget http://deeplearning.net/data/mnist/mnist.pkl.gz -O data/mnist/mnist.pkl.gz
--2019-01-03 20:09:52-- http://deeplearning.net/data/mnist/mnist.pkl.gz Resolving deeplearning.net (deeplearning.net)... 132.204.26.28 Connecting to deeplearning.net (deeplearning.net)|132.204.26.28|:80... connected. HTTP request sent, awaiting response... 200 OK Length: 16168813 (15M) [application/x-gzip] Saving to: ‘data/mnist/mnist.pkl.gz’ data/mnist/mnist.pk 100%[===================>] 15.42M 4.11MB/s in 4.1s 2019-01-03 20:09:56 (3.74 MB/s) - ‘data/mnist/mnist.pkl.gz’ saved [16168813/16168813]
path = Path('data/mnist')
path.ls()
[PosixPath('data/mnist/mnist.pkl.gz')]
with gzip.open(path/'mnist.pkl.gz', 'rb') as f:
((x_train, y_train), (x_valid, y_valid), _) = pickle.load(f, encoding='latin-1')
plt.imshow(x_train[0].reshape((28,28)), cmap='gray')
x_train.shape
(50000, 784)
x_train,y_train,x_valid,y_valid = map(torch.tensor, (x_train,y_train,x_valid,y_valid))
n,c = x_train.shape
x_train.shape, y_train.min(), y_train.max()
(torch.Size([50000, 784]), tensor(0), tensor(9))
In lesson2-sgd we did these things ourselves:
x = torch.ones(n,2)
def mse(y_hat, y): return ((y_hat-y)**2).mean()
y_hat = x@a
Now instead we'll use PyTorch's functions to do it for us, and also to handle mini-batches (which we didn't do last time, since our dataset was so small).
bs = 64
train_ds = TensorDataset(x_train, y_train) # PyTorch dataset wrapping tensors.
valid_ds = TensorDataset(x_valid, y_valid)
data = DataBunch.create(train_ds, valid_ds, bs=bs)
x,y = next(iter(data.train_dl))
x.shape, y.shape
(torch.Size([64, 784]), torch.Size([64]))
class Mnist_Logistic(nn.Module):
def __init__(self):
super().__init__()
self.lin = nn.Linear(784, 10, bias=True)
def forward(self, xb): return self.lin(xb) # xb is a batch of X
model = Mnist_Logistic().cuda()
model
Mnist_Logistic( (lin): Linear(in_features=784, out_features=10, bias=True) )
model.lin
Linear(in_features=784, out_features=10, bias=True)
model(x).shape
torch.Size([64, 10])
# Sanity check
for param in model.parameters():
print(type(param.data), param.size())
<class 'torch.Tensor'> torch.Size([10, 784]) <class 'torch.Tensor'> torch.Size([10])
[p.shape for p in model.parameters()]
[torch.Size([10, 784]), torch.Size([10])]
lr = 2e-2
loss_func = nn.CrossEntropyLoss() # combines Log Softmax and NLL loss
def update(x, y, lr):
wd = 1e-5 # new
y_hat = model(x) # was y_hat = x@a
# weight decay
w2 = 0.
for p in model.parameters(): w2 += (p**2).sum()
# add to regular loss
loss = loss_func(y_hat, y) + w2*wd # was mse(y, y_hat)
loss.backward()
with torch.no_grad():
for p in model.parameters():
p.sub_(lr * p.grad) # was a.sub_(lr * a.grad)
p.grad.zero_() # was a.grad.zero_()
return loss.item()
losses = [update(x, y, lr) for x, y in data.train_dl]
plt.plot(losses)
[<matplotlib.lines.Line2D at 0x7fc930369cc0>]
class Mnist_NN(nn.Module):
def __init__(self):
super().__init__()
self.lin1 = nn.Linear(784, 50, bias=True)
self.lin2 = nn.Linear(50, 10, bias=True)
def forward(self, xb):
x = self.lin1(xb)
x = F.relu(x) # add non-linearities
return self.lin2(x)
model = Mnist_NN().cuda()
losses = [update(x, y, lr) for x, y in data.train_dl]
plt.plot(losses)
[<matplotlib.lines.Line2D at 0x7fc916e14f98>]
model = Mnist_NN().cuda()
def update(x, y, lr):
opt = optim.Adam(model.parameters(), lr)
y_hat = model(x)
loss = loss_func(y_hat, y)
loss.backward()
opt.step()
opt.zero_grad()
return loss.item()
losses = [update(x, y, 1e-3) for x, y in data.train_dl]
plt.plot(losses)
[<matplotlib.lines.Line2D at 0x7fc930326668>]
learn = Learner(data, Mnist_NN(), loss_func=loss_func, metrics=accuracy)
learn.lr_find()
learn.recorder.plot()
LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.
learn.fit_one_cycle(1, 1e-2)
epoch | train_loss | valid_loss | accuracy |
---|---|---|---|
1 | 0.142047 | 0.127680 | 0.963500 |
learn.recorder.plot_lr(show_moms=True)
learn.recorder.plot_losses()