def sum(t: Tensor, axis=None, keepdims=False):
"""
Sums all the elements in tensor along given axis
## Parameters:
t: `Tensor`
axis: `int` - defaults to None
keepdims: `bool` - defaults to False
## Example usage
```python
from beacon.tensor import Tensor
from beacon.tensor import functions as fn
t = Tensor([1, 2, 3])
x = fn.sum(t)
```
"""
data = np.sum(t.data, axis=axis, keepdims=keepdims)
requires_grad = t.requires_grad and not Tensor.NO_GRAD
nodes = []
if requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t, df=lambda x: _match_shape(x, t.data.shape, axis, keepdims)[0]))
return Tensor(data=data, requires_grad=requires_grad, nodes=nodes)
def divide(t1: Tensor, t2: Tensor):
"""
Divides two tensors.
## Parameters:
t1: `Tensor` - first tensor
t2: `Tensor` - second tensor
## Example usage
```python
from beacon.tensor import Tensor
from beacon.tensor import functions as fn
t1 = Tensor([1, 2, 3])
t2 = Tensor([4, 5, 6])
x = fn.divide(t1, t2)
```
"""
data = t1.data / t2.data
requires_grad = (t1.requires_grad or t2.requires_grad) and not Tensor.NO_GRAD
nodes = []
if t1.requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t1, df=lambda x: _broadcast(t1.grad.data, x /t2.data)))
if t2.requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t2, df=lambda x: _broadcast(t2.grad.data, -x * t1.data/ t2.data**2 )))
return Tensor(data=data, requires_grad=requires_grad, nodes=nodes)
def forward(self, x):
if self.train_mode:
activation_mask = Tensor(data=np.random.rand(*(x.shape)) /
(1 - self.dropout_rate),
requires_grad=True)
x = fn.mul(x, activation_mask > self.dropout_rate)
return x
def tan(t: Tensor):
"""
Applies tan function to all the elements of the input tensor.
## Parameters:
t: `Tensor` - input tensor
## Example usage
```python
from beacon.tensor import Tensor
from beacon.tensor import functions as fn
t = Tensor([1, 2, 3])
x = fn.tan(t)
```
"""
data = np.tan(t.data)
requires_grad = t.requires_grad and not Tensor.NO_GRAD
nodes = []
if requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t, df=lambda x: x / np.cos(t.data)**2))
return Tensor(data=data, requires_grad=requires_grad, nodes=nodes)
def neg(t: Tensor):
"""
Unary negation of tensor elements.
## Parameters:
t: `Tensor` - input tensor
## Example usage
```python
from beacon.tensor import Tensor
from beacon.tensor import functions as fn
t = Tensor([1, 2, 3])
x = -t
```
"""
data = -t.data
requires_grad = t.requires_grad and not Tensor.NO_GRAD
nodes = []
if requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t, df=lambda x: -x))
return Tensor(data=data, requires_grad=requires_grad, nodes=nodes)
def to_tensor(x):
"""
Convert input parameter to tensor if it isn't already.
## Parameters
x: `Tensor-like` - input parameter
## Example usage
```python
from beacon.tensor import functinos as fn
t = fn.to_tensor(10.0)
```
"""
return Tensor._to_tensor(x)
def reshape(t: Tensor, shape):
"""
Reshapes tensor.
## Parameters:
t: `Tensor` - input tensor
shape: `tuple` - new shape
## Example usage
```python
from beacon.tensor import Tensor
from beacon.tensor import functions as fn
t = Tensor([1, 2, 3], [4, 5, 6])
x = fn.reshape(t, shape=(1, 6))
```
"""
data = np.reshape(t.data, newshape=shape)
requires_grad = t.requires_grad and not Tensor.NO_GRAD
nodes = []
if requires_grad:
nodes.append(Tensor.ComputationalGraphNode(tensor=t, df=lambda x: np.reshape(x, np.shape(t.data))))
return Tensor(data=data, requires_grad=requires_grad, nodes=nodes)