PK ! G jrpytorch/__init__.py__version__ = '0.1.0'
from . import datasets
from . import visualiser
import torch
import matplotlib.pyplot as plt
import numpy as np
def vis_binary_classification(model, loss_fn, optimiser, X, y, epochs=1000):
x_tensor = torch.tensor(X).float()
y_tensor = torch.tensor(y.reshape(-1, 1)).float()
plt.ion()
min = np.min(X)-0.2
max = np.max(X)+0.2
x1 = np.linspace(min, max, 100)
grid = np.array([(x, y) for x in x1 for y in x1])
fig = plt.figure()
ax = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
real_x = x1
real_y = x1
dx = (real_x[1]-real_x[0])/2.
dy = (real_y[1]-real_y[0])/2.
extent = [real_x[0]-dx, real_x[-1]+dx, real_y[0]-dy, real_y[-1]+dy]
ax.set_xlim([min, max])
ax.set_ylim([min, max])
ax2.set_xlim([0, epochs])
line1, = ax2.plot([], [], 'r-')
# first image
output = model(torch.from_numpy(grid).float()).detach().numpy()
img = ax.imshow(output.reshape(100, 100), extent=extent)
ax.scatter(X[:, 0], X[:, 1], c=y, edgecolor='black')
losses = []
for epoch in range(epochs):
optimiser.zero_grad()
output = model(x_tensor)
loss = loss_fn(output, y_tensor)
loss.backward()
optimiser.step()
losses.append(loss.item())
if epoch == 0:
ax2.set_ylim([0, loss.item()*1.1])
with torch.no_grad():
# update plot
output = model(torch.from_numpy(grid).float())
arr_output = output.detach().numpy().reshape(100, 100)
img.set_data(arr_output)
ax.set_title('Epoch: %d Loss %.3f' % (epoch + 1, loss.item()))
line1.set_data(list(range(epoch+1)), losses)
fig.canvas.draw()
fig.canvas.flush_events()
plt.ioff()
PK ! OAشy y jrpytorch/datasets/__init__.pyfrom sklearn.datasets import make_circles
def load_circles():
return make_circles(200, random_state=1, noise=0.05)
PK ! F jrpytorch/visualiser.pyfrom datetime import datetime
import visdom
import torch
class Classification_Visualiser:
def __init__(self, env_name=None, host='http://localhost', port=8097):
if env_name is None:
env_name = str(datetime.now().strftime('%d-%m-%Hh%M'))
self.env_name = env_name
self.vis = visdom.Visdom(server=host, port=port, env=self.env_name)
self.loss_win = None
self.acc_win = None
self.text_win = None
def plot_loss(self, loss, epoch):
self.loss_win = self.vis.line(
[loss], [epoch],
win=self.loss_win, update='append' if self.loss_win else None,
opts={
'xlabel': 'Epoch',
'ylabel': 'Loss',
'legend': ['train', 'test'],
'layoutopts': {
'legend': {
'train': 0,
'test': 1
}
}
}
)
def plot_accuracy(self, acc, epoch):
self.acc_win = self.vis.line(
[acc], [epoch],
win=self.acc_win, update='append' if self.acc_win else None,
opts={
'xlabel': 'Epoch',
'ylabel': 'Accuracy',
'legend': ['train', 'test'],
'layoutopts': {
'legend': {
'train': 0,
'test': 1
}
}
}
)
def show_text(self, text):
self.text_win = self.vis.text(
text, win=self.text_win
)
def train_classifier(model, criterion, optimiser, epochs, visualiser, loaders, sizes, device='cpu'):
model = model.to(device)
losses = {
'train': None,
'test': None
}
acc = {
'train': None,
'test': None
}
for epoch in range(epochs):
text = ' Epoch %d / %d
' % (epoch + 1, epochs)
for stage in ['train', 'test']:
if stage == 'train':
model.train()
else:
model.eval()
running_loss = .0
running_correct = 0
for input, label in loaders[stage]:
input, label = input.to(device), label.to(device)
optimiser.zero_grad()
with torch.set_grad_enabled(stage == 'train'):
output = model(input)
_, pred = torch.max(output, 1)
loss = criterion(output, label)
if stage == 'train':
loss.backward()
optimiser.step()
running_loss += loss.item() * input.size(0)
running_correct += torch.sum(pred == label.data)
stage_loss = running_loss/sizes[stage]
stage_accuracy = running_correct.double() / sizes[stage]
losses[stage] = stage_loss
acc[stage] = stage_accuracy.cpu().numpy()
text += ' %s Loss: %.3f Acc: %.3f
' % (stage,stage_loss, stage_accuracy)
visualiser.plot_loss(list(losses.values()), [epoch, epoch])
visualiser.plot_accuracy(list(acc.values()), [epoch, epoch])
visualiser.show_text(text)
PK !H=BT T jrpytorch-0.1.2.dist-info/WHEEL
A
н#Z;/"d&F[xzw@ Zpy3Fv]n0H*J>mlc A PK !HZ! V " jrpytorch-0.1.2.dist-info/METADATAN <GMZRFFѨ.k
tn4^4d <:i4a%8nVI #z;pniM_D+Y B̯z1;PRT(mPm:+wATF'K+\ t/sLaښƂR@7}pym ʓ ~o +]k|+QtNG1 ";Il@"CIER87}Hc`& 3lj&]ľPK !HgR- jrpytorch-0.1.2.dist-info/RECORD}9v@ >guP`,RvB ˠ3 (A<}*<>.Svi$}.CO) Avw[ިA(L}:f
9^vl ow+>N0'%" Cތ%eU7-%Vl( 2O7CZc. ^8h2|xEC R IVrY..f`
L{7e;$_::*ܪNry<BZ-͗Tɗ qI36U{NhtmQ!Ziv2]MpUyPK ! G jrpytorch/__init__.pyPK ! OAشy y . jrpytorch/datasets/__init__.pyPK ! F jrpytorch/visualiser.pyPK !H=BT T jrpytorch-0.1.2.dist-info/WHEELPK !HZ! V " jrpytorch-0.1.2.dist-info/METADATAPK !HgR- jrpytorch-0.1.2.dist-info/RECORDPK n