def conv2d(input, weight, bias=None, stride=1, padding=0):
input = _check_tensor(input)
weight = _check_tensor(weight)
bias = _check_tensor(bias)
p = [input.op, weight.op]
if bias is None:
p.append(None)
else:
p.append(bias.op)
op = _operators.Conv2d(p, stride, padding)
return op.tensor
def abs(t):
t = _check_tensor(t)
op = _operators.Abs([t.op])
return op.tensor
def exp(t):
t = _check_tensor(t)
op = _operators.Exp([t.op])
return op.tensor
def elu(t, alpha=1.0):
t = _check_tensor(t)
op = _operators.ELU([t.op], alpha)
return op.tensor
def dropout(t, p=0.5, training=False):
t = _check_tensor(t)
op = _operators.Dropout([t.op], p, training)
return op.tensor
def add(lt, rt):
lt = _check_tensor(lt)
rt = _check_tensor(rt)
op = _operators.Add([lt.op, rt.op])
return op.tensor
def maximum(lt, rt):
lt = _check_tensor(lt)
rt = _check_tensor(rt)
op = _operators.Maximum([lt.op, rt.op])
return op.tensor
def log(t):
t = _check_tensor(t)
op = _operators.Log([t.op])
return op.tensor
def transpose(t):
t = _check_tensor(t)
op = _operators.Transpose([t.op])
return op.tensor
def sub(lt, rt):
lt = _check_tensor(lt)
rt = _check_tensor(rt)
op = _operators.Sub([lt.op, rt.op])
return op.tensor
def sqrt(t):
t = _check_tensor(t)
op = _operators.Sqrt([t.op])
return op.tensor
def sigmoid(t):
t = _check_tensor(t)
op = _operators.Sigmoid([t.op])
return op.tensor
def reshape(t, *shapes):
t = _check_tensor(t)
op = _operators.Reshape([t.op], *shapes)
return op.tensor
def relu(t):
t = _check_tensor(t)
op = _operators.ReLU([t.op])
return op.tensor
def neg(t):
t = _check_tensor(t)
op = _operators.Neg([t.op])
return op.tensor
def getitem(t, ind):
t = _check_tensor(t)
op = _operators.GetItem([t.op], ind)
return op.tensor
def leakyrelu(t, negative_slope=1e-2):
t = _check_tensor(t)
op = _operators.LeakyReLU([t.op], negative_slope)
return op.tensor
def cross_entropy(logits, target):
logits = _check_tensor(logits)
target = _check_tensor(target).data
op = _operators.CrossEntropy([logits.op], target)
return op.tensor
def matmul(lt, rt):
lt = _check_tensor(lt)
rt = _check_tensor(rt)
op = _operators.MatMul([lt.op, rt.op])
return op.tensor
def div(lt, rt):
lt = _check_tensor(lt)
rt = _check_tensor(rt)
op = _operators.Div([lt.op, rt.op])
return op.tensor
def max(t, axis=None, keepdims=False):
t = _check_tensor(t)
op = _operators.Max([t.op], axis, keepdims)
return op.tensor
def max_pool2d(input, kernel_size, stride, padding=0, return_indices=True):
input = _check_tensor(input)
op = _operators.MaxPool2d([input.op], kernel_size, stride, padding,
return_indices)
return op.tensor
