aparler/PyNorch
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

softmax v1 need fix some erros yet

Browse source
This commit is contained in:
lucasdelimanogueira 2024-05-20 20:16:41 -03:00
parent cd57c39ff0
commit 5e72fe57d3
10 changed files with 106 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

BIN
build/tensor.o
View file

Binary file not shown.

BIN
norch/__pycache__/tensor.cpython-38.pyc
View file

Binary file not shown.

28
norch/csrc/tensor.cpp
View file

@ -225,14 +225,20 @@ extern "C" {
if (keepdim) {
shape = (int*) malloc((tensor->ndim) * sizeof(int));
for (int i = 0; i < tensor->ndim; i++) {
shape[i] = tensor->shape[i];
if (axis == -1) {
shape[i] = 1;
} else {
shape[i] = tensor->shape[i];
}
}
shape[axis] = 1;
ndim = tensor->ndim;
Tensor* new_tensor = create_tensor(result_data, shape, ndim, device);
return reshape_tensor(new_tensor, shape, ndim);
}
return create_tensor(result_data, shape, ndim, device);
}
}
@ -289,11 +295,16 @@ extern "C" {
if (keepdim) {
shape = (int*) malloc((tensor->ndim) * sizeof(int));
for (int i = 0; i < tensor->ndim; i++) {
shape[i] = tensor->shape[i];
if (axis == -1) {
shape[i] = 1;
} else {
shape[i] = tensor->shape[i];
}
}
shape[axis] = 1;
ndim = tensor->ndim;
Tensor* new_tensor = create_tensor(result_data, shape, ndim, device);
return reshape_tensor(new_tensor, shape, ndim);
}
return create_tensor(result_data, shape, ndim, device);
@ -353,11 +364,16 @@ extern "C" {
if (keepdim) {
shape = (int*) malloc((tensor->ndim) * sizeof(int));
for (int i = 0; i < tensor->ndim; i++) {
shape[i] = tensor->shape[i];
if (axis == -1) {
shape[i] = 1;
} else {
shape[i] = tensor->shape[i];
}
}
shape[axis] = 1;
ndim = tensor->ndim;
Tensor* new_tensor = create_tensor(result_data, shape, ndim, device);
return reshape_tensor(new_tensor, shape, ndim);
}
return create_tensor(result_data, shape, ndim, device);

BIN
norch/libtensor.so
View file

Binary file not shown.

BIN
norch/nn/__pycache__/activation.cpython-38.pyc
View file

Binary file not shown.

11
norch/nn/activation.py
View file

@ -18,4 +18,13 @@ class Sigmoid(Activation):
super().__init__()
def forward(self, x):
return F.sigmoid(x)
return F.sigmoid(x)
class Softmax(Activation):
def __init__(self, dim):
super(Softmax, self).__init__()
self.dim = dim
def forward(self, x):
return F.softmax(x, self.dim)

13
norch/nn/functional.py
View file

@ -3,3 +3,16 @@ import math
def sigmoid(x):
return 1.0 / (1.0 + (math.e) ** (-x))
def softmax(x, dim=None):
e = math.e ** x
s = e.sum(axis=dim, keepdim=True)
#l = s.log()
print('x', x)
print('\n\n')
#print('log', l)
print('\n\n')
#print('x-log', x - l)
#return math.e ** (x - sum.log())

15
norch/tensor.py
View file

@ -697,7 +697,7 @@ class Tensor:
elif self.ndim == other.ndim - 1:
self = self.reshape([1] + self.shape)
# Call equal_broadcasted_tensor if broadcasting is needed
Tensor._C.equal_broadcasted_tensor.argtypes = [ctypes.POINTER(CTensor), ctypes.POINTER(CTensor)]
Tensor._C.equal_broadcasted_tensor.restype = ctypes.POINTER(CTensor)
@ -748,7 +748,9 @@ class Tensor:
return result_data
def sum(self, axis=-1, keepdim=False):
def sum(self, axis=None, keepdim=False):
if axis == None:
axis = -1
Tensor._C.sum_tensor.argtypes = [ctypes.POINTER(CTensor), ctypes.c_int, ctypes.c_bool]
Tensor._C.sum_tensor.restype = ctypes.POINTER(CTensor)
@ -783,10 +785,13 @@ class Tensor:
return result_data
def max(self, axis=-1, keepdim=False):
def max(self, axis=None, keepdim=False):
if axis == None:
axis = -1
Tensor._C.max_tensor.argtypes = [ctypes.POINTER(CTensor), ctypes.c_int, ctypes.c_bool]
Tensor._C.max_tensor.restype = ctypes.POINTER(CTensor)
print(axis, keepdim)
result_tensor_ptr = Tensor._C.max_tensor(self.tensor, axis, keepdim)
result_data = Tensor()
@ -818,7 +823,9 @@ class Tensor:
return result_data
def min(self, axis=-1, keepdim=False):
def min(self, axis=None, keepdim=False):
if axis == None:
axis = -1
Tensor._C.min_tensor.argtypes = [ctypes.POINTER(CTensor), ctypes.c_int, ctypes.c_bool]
Tensor._C.min_tensor.restype = ctypes.POINTER(CTensor)

15
test.py
View file

@ -6,11 +6,16 @@ import norch.optim as optim
import random
random.seed(1)
torch_tensor = norch.Tensor([[[10, 10], [-4, -4]], [[5., 6], [7, 8]]], requires_grad=True)#.to(self.device)
torch_tensor2 = norch.Tensor([[[10.0,], [-4.0,]],[[6.0,], [8.0,]]])
torch_tensor3 = torch_tensor.max(axis=2, keepdim=True)
print(torch_tensor3)
print(torch_tensor.equal(torch_tensor3))
torch_tensor = norch.Tensor([[[2, 2], [-1, -1]], [[1., 2], [3, 3]]], requires_grad=True)#.to(self.device)
b = norch.nn.functional.softmax(torch_tensor)
"""a = norch.Tensor([[[4.186502456665039]]])
b = norch.Tensor([[[2.0, 2.0,],[-1.0, -1.0,]],[[1.0, 2.0,],[3.0, 3.0,]]])
print(b)
print(a.shape)
print(b-a)"""
"""print(torch_tensor.shape)
print('\n\n')

43
tests/test_nn.py
View file

@ -60,11 +60,11 @@ class TestNNModuleActivationFn(unittest.TestCase):
sigmoid_fn_torch = torch.nn.Sigmoid()
# Test case 1: Positive input
x = norch.Tensor([1, 2, 3]).to(self.device)
x = norch.Tensor([[1, 2, 3]]).to(self.device)
sigmoid_norch = sigmoid_fn_norch.forward(x)
sigmoid_torch_result = utils.to_torch(sigmoid_norch).to(self.device)
x = torch.tensor([1, 2, 3]).to(self.device)
x = torch.tensor([[1, 2, 3]]).to(self.device)
sigmoid_torch_expected = sigmoid_fn_torch.forward(x)
self.assertTrue(utils.compare_torch(sigmoid_torch_result, sigmoid_torch_expected))
@ -87,4 +87,41 @@ class TestNNModuleActivationFn(unittest.TestCase):
x = torch.tensor([0, 0, 0]).to(self.device)
sigmoid_torch_expected = sigmoid_fn_torch.forward(x)
self.assertTrue(utils.compare_torch(sigmoid_torch_result, sigmoid_torch_expected))
self.assertTrue(utils.compare_torch(sigmoid_torch_result, sigmoid_torch_expected))
def test_softmax_activation(self):
"""
Test Softmax activation function
"""
# Test different axes
axes = [None, 0, 1, -1]
# Define the input tensors for different test cases
test_cases = [
(norch.Tensor([[1., 2, 3], [4, 5, 6]]), torch.tensor([[1., 2, 3], [4, 5, 6]])),
(norch.Tensor([[1., -1, 0], [2, -2, 0]]), torch.tensor([[1., -1, 0], [2, -2, 0]])),
(norch.Tensor([[0., 0, 0], [0, 0, 0]]), torch.tensor([[0., 0, 0], [0, 0, 0]]))
]
for dim in axes:
softmax_fn_norch = norch.nn.Softmax(dim=dim)
softmax_fn_torch = torch.nn.Softmax(dim=dim)
for norch_input, torch_input in test_cases:
# Move tensors to the correct device
norch_input = norch_input.to(self.device)
torch_input = torch_input.to(self.device)
# Forward pass using norch
softmax_norch = softmax_fn_norch.forward(norch_input)
softmax_torch_result = utils.to_torch(softmax_norch).to(self.device)
# Forward pass using torch
softmax_torch_expected = softmax_fn_torch.forward(torch_input)
# Compare the results
print(softmax_torch_result)
print(softmax_torch_expected)
self.assertTrue(utils.compare_torch(softmax_torch_result, softmax_torch_expected))