def concat(tensors,axis=0):
"""Concatenate the tensors along the given axis"""
tensors = [t if isinstance(t, engine.Tensor) else engine.Tensor(t) for t in tensors]
out = engine.Tensor(np.concatenate([t.data for t in tensors], axis=axis),_children=tensors, op='concatenate')
def _backward():
st = 0
for t in enumerate(tensors):
indexing = [slice(st, st + d) if i == axis else slice(0,d) for i, d in enumerate(t.shape)]
t.grad += out.grad[indexing]
st += t.shape[axis]
out._backward = _backward
return out
def assign(x, indexing, y):
y = y if isinstance(y, engine.Tensor) else engine.Tensor(y)
data = x.data.copy()
data[indexing] = y.data
out = engine.Tensor(data, _children=(x, y), op='set item',)
def _backward():
y.grad += out.grad[indexing]
gradient = out.grad.copy()
x.grad += gradient
x.grad[indexing] = 0.
out._backward = _backward
return out
def glorot_uniform(fan_in, fan_out, **kwargs):
limit = math.sqrt(6 / (fan_in + fan_out))
return engine.Tensor(np.random.uniform(-limit, limit, **kwargs))
def random(*args,**kwargs):
return engine.Tensor(np.random.uniform(*args,**kwargs))
def arange(*args, **kwargs):
return engine.Tensor(np.arange(*args, **kwargs))
def zeros(*args, **kwargs):
return engine.Tensor(np.zeros(*args,**kwargs))\
def empty(*args,**kwargs):
return engine.Tensor(np.empty(*args,**kwargs))