def trainingStep(net, NTRAIN, min_difficulty = 1.0, max_difficulty = 1.0, min_sparseness = 0, max_sparseness = 0, min_imbalance = 0, max_imbalance = 0, feature_variation = True, class_variation = True, BS = 200): FEATURES = net.FEATURES CLASSES = net.CLASSES net.zero_grad() batch_mem = [] batch_test = [] batch_label = [] class_count = [] for i in range(BS): if feature_variation: feat = np.random.randint(2.5*FEATURES) + FEATURES//2 else: feat = FEATURES if class_variation: classes = np.random.randint(CLASSES-2) + 2 else: classes = CLASSES xd,yd = problemGenerator(N=NTRAIN+100, FEATURES=feat, CLASSES=classes, sigma = np.random.rand()*(max_difficulty - min_difficulty) + min_difficulty, sparseness = np.random.rand()*(max_sparseness - min_sparseness) + min_sparseness, imbalance = np.random.rand()*(max_imbalance - min_imbalance) + min_imbalance) if classes<CLASSES: yd = np.pad(yd, ( (0,0), (0,CLASSES-classes)), 'constant', constant_values=0) xd = normalizeAndProject(xd, NTRAIN, FEATURES) trainset = np.hstack([xd[0:NTRAIN],yd[0:NTRAIN]]) testset = xd[NTRAIN:] labelset = yd[NTRAIN:] batch_mem.append(trainset) batch_test.append(testset) batch_label.append(labelset) class_count.append(classes) batch_mem = tovar(np.array(batch_mem).transpose(0,2,1).reshape(BS,1,FEATURES+CLASSES,NTRAIN)) batch_test = tovar(np.array(batch_test).transpose(0,2,1).reshape(BS,1,FEATURES,100)) batch_label = tovar(np.array(batch_label).transpose(0,2,1)) class_count = torch.cuda.FloatTensor(np.array(class_count)) net.zero_grad() p = net.forward(batch_mem, batch_test, class_count) loss = -torch.sum(p*batch_label,1).mean() loss.backward() net.adam.step() err = loss.cpu().data.numpy()[0] return err
def trainingStep(net, NTRAIN, NTEST, data_x, data_y, BS=200):
FEATURES = net.FEATURES
CLASSES = net.CLASSES
net.zero_grad()
err = []
for i in range(BS):
j = np.random.randint(len(data_x))
feat = data_x[j].shape[1]
classes = np.unique(data_y[j]).shape[0]
xd = data_x[j].copy()
# Data augmentation
f_idx = np.random.permutation(feat)
xd = xd[:, f_idx]
c_idx = np.random.permutation(classes)
yd = np.zeros((data_y[j].shape[0], classes))
yd[np.arange(data_y[j].shape[0]),
c_idx[data_y[j][np.arange(data_y[j].shape[0])]]] = 1
idx = np.random.permutation(xd.shape[0])
xd = xd[idx]
yd = yd[idx]
if classes < CLASSES:
yd = np.pad(yd, ((0, 0), (0, CLASSES - classes)),
'constant',
constant_values=0)
xd = normalizeAndProject(xd, NTRAIN, FEATURES)
yd = np.argmax(yd, axis=1)
trainset_x = tovar(xd[0:NTRAIN])
trainset_y = toivar(yd[0:NTRAIN])
testset = tovar(xd[NTRAIN:NTRAIN + NTEST])
labelset = toivar(yd[NTRAIN:NTRAIN + NTEST])
idx = torch.arange(NTEST).cuda().long()
p = net.fullpass(trainset_x, trainset_y, testset, classes)
loss = -torch.mean(p[idx, labelset[idx]])
loss.backward()
err.append(loss.cpu().detach().item())
net.adam.step()
return np.mean(err)
def plotDecisionBoundary(x, y, net):
x = (x - np.mean(x, axis=0, keepdims=True)) / (
1e-16 + np.std(x, axis=0, keepdims=True))
x = x.reshape((1, 1, x.shape[0], x.shape[1])).transpose(0, 1, 3, 2)
y = y.reshape((1, 1, y.shape[0], y.shape[1])).transpose(0, 1, 3, 2)
trainset = np.concatenate([x, y], axis=2)
xx, yy = np.meshgrid(np.arange(-3.0, 3.05, 0.05),
np.arange(-3.0, 3.05, 0.05))
XR = xx.shape[0]
xx = xx.reshape((1, 1, 1, XR * XR))
yy = yy.reshape((1, 1, 1, XR * XR))
testset = np.concatenate([xx, yy], axis=2)
p = np.exp(
net.forward(tovar(trainset), tovar(testset),
torch.cuda.FloatTensor(np.array([4]))).cpu().data.numpy())
p = p.reshape((4, XR, XR)).transpose(1, 2, 0)
xx = xx.reshape((XR, XR))
yy = yy.reshape((XR, XR))
colors = np.array([[0.7, 0.2, 0.2], [0.2, 0.7, 0.2], [0.2, 0.2, 0.7],
[0.7, 0.2, 0.7]])
im = np.zeros((XR, XR, 3))
for j in range(4):
im += p[:, :, j].reshape((XR, XR, 1)) * np.array(colors[j]).reshape(
(1, 1, 3))
yl = np.argmax(y, axis=2)[0, 0]
plt.imshow(im, extent=[-3, 3, 3, -3])
for j in range(4):
plt.scatter(x[0, 0, 0, yl == j],
x[0, 0, 1, yl == j],
c=colors[j],
edgecolors='k',
lw=1,
s=10)
plt.xticks([])
plt.yticks([])
plt.xlim(-3, 3)
plt.ylim(-3, 3)
def trainingStep(net,
NTRAIN,
min_difficulty=1.0,
max_difficulty=1.0,
min_sparseness=0,
max_sparseness=0,
min_imbalance=0,
max_imbalance=0,
feature_variation=True,
class_variation=True,
BS=20):
FEATURES = net.FEATURES
CLASSES = net.CLASSES
net.zero_grad()
err = []
for i in range(BS):
if feature_variation:
feat = np.random.randint(2.5 * FEATURES) + FEATURES // 2
else:
feat = FEATURES
if class_variation:
classes = np.random.randint(CLASSES - 2) + 2
else:
classes = CLASSES
xd, yd = problemGenerator(
N=NTRAIN + 100,
FEATURES=feat,
CLASSES=classes,
sigma=np.random.rand() * (max_difficulty - min_difficulty) +
min_difficulty,
sparseness=np.random.rand() * (max_sparseness - min_sparseness) +
min_sparseness,
imbalance=np.random.rand() * (max_imbalance - min_imbalance) +
min_imbalance)
if classes < CLASSES:
yd = np.pad(yd, ((0, 0), (0, CLASSES - classes)),
'constant',
constant_values=0)
xd = normalizeAndProject(xd, NTRAIN, FEATURES)
yd = np.argmax(yd, axis=1)
trainset_x = tovar(xd[0:NTRAIN])
trainset_y = toivar(yd[0:NTRAIN])
testset = tovar(xd[NTRAIN:])
labelset = toivar(yd[NTRAIN:])
idx = torch.arange(100).cuda().long()
p = net.fullpass(trainset_x, trainset_y, testset, classes)
loss = -torch.mean(p[idx, labelset[idx]])
loss.backward()
err.append(loss.cpu().detach().item())
net.adam.step()
return np.mean(err)
def trainingStep(net, NTRAIN, NTEST, data_x, data_y, BS=200):
FEATURES = net.FEATURES
CLASSES = net.CLASSES
net.zero_grad()
batch_mem = []
batch_test = []
batch_label = []
class_count = []
for i in range(BS):
j = np.random.randint(len(data_x))
feat = data_x[j].shape[1]
classes = np.unique(data_y[j]).shape[0]
xd = data_x[j].copy()
# Data augmentation
f_idx = np.random.permutation(feat)
xd = xd[:, f_idx]
c_idx = np.random.permutation(classes)
yd = np.zeros((data_y[j].shape[0], classes))
yd[np.arange(data_y[j].shape[0]),
c_idx[data_y[j][np.arange(data_y[j].shape[0])]]] = 1
idx = np.random.permutation(xd.shape[0])
xd = xd[idx]
yd = yd[idx]
if classes < CLASSES:
yd = np.pad(yd, ((0, 0), (0, CLASSES - classes)),
'constant',
constant_values=0)
xd = normalizeAndProject(xd, NTRAIN, FEATURES)
trainset = np.hstack([xd[0:NTRAIN], yd[0:NTRAIN]])
testset = xd[NTRAIN:NTRAIN + NTEST]
labelset = yd[NTRAIN:NTRAIN + NTEST]
batch_mem.append(trainset)
batch_test.append(testset)
batch_label.append(labelset)
class_count.append(classes)
batch_mem = tovar(
np.array(batch_mem).transpose(0, 2,
1).reshape(BS, 1, FEATURES + CLASSES,
NTRAIN))
batch_test = tovar(
np.array(batch_test).transpose(0, 2,
1).reshape(BS, 1, FEATURES, NTEST))
batch_label = tovar(np.array(batch_label).transpose(0, 2, 1))
class_count = torch.cuda.FloatTensor(np.array(class_count))
net.zero_grad()
p = net.forward(batch_mem, batch_test, class_count)
loss = -torch.sum(p * batch_label, 1).mean()
loss.backward()
net.adam.step()
err = loss.cpu().data.numpy()[0]
return err
