def predict_proba(self, X):
loader = _build_data_loader(self.batch_size, X)
for idx, x in enumerate(loader):
if idx == 0:
y_hat = self(x.float()).data.numpy()
else:
y_hat = np.vstack([y_hat, self(x.float()).data.numpy()])
return y_hat
def fit(self,
X,
y,
epochs=10,
loss='mse',
learning_rate=1e-3,
batch_size=10,
solver='adam',
verbose=False):
self.batch_size = batch_size
loader = _build_data_loader(self.batch_size, X, y)
opt = OPTIMIZERS[solver](self.parameters(), lr=learning_rate)
crit = LOSS_FUNCS[loss]
for _ in range(epochs):
for x, y in loader:
opt.zero_grad()
y_hat = self(x.float())
loss = crit(y_hat, y.float())
loss.backward()
opt.step()
def fit(self,
X,
y,
epochs=10,
loss='crossentropy',
learning_rate=1e-3,
batch_size=10,
optim='adam',
verbose=False):
self.batch_size = batch_size
loader = _build_data_loader(self.batch_size, X, y)
opt = OPTIMIZERS[optim](self.parameters(), lr=learning_rate)
crit = LOSS_FUNCS[loss]
for e in range(epochs):
if verbose:
print('\rEpoch {}'.format(e + 1), end='')
for x, y in loader:
opt.zero_grad()
y_hat = self(x.float())
loss = crit(y_hat, y.float())
loss.backward()
opt.step()