tensor
import torch
tensor1d = torch.tensor([1, -1, 15, 1000])
print(tensor1d)
tensor([ 1, -1, 15, 1000])
tensor2d = torch.tensor([[-10, 11, 1.5],
[ 5.6, 88, 10.0],
[ 2.3, 36, -10.0],
[ 2.0, 12, 0.2]])
print(tensor2d)
tensor([[-10.0000, 11.0000, 1.5000],
[ 5.6000, 88.0000, 10.0000],
[ 2.3000, 36.0000, -10.0000],
[ 2.0000, 12.0000, 0.2000]])
tensor3d = torch.tensor([[[-10, 11, 1.50],
[ 5.6, 88, 10.0],
[ 2.3, 36, -10.0],
[ 2.0, 12, 0.2]],
[[-1.0, 11, 1.50],
[ 0.6, 88, 1.00],
[ 0.3, 3.6, -10.0],
[ 2.1, 1.2, 4.20]]])
print(tensor3d)
tensor([[[-10.0000, 11.0000, 1.5000],
[ 5.6000, 88.0000, 10.0000],
[ 2.3000, 36.0000, -10.0000],
[ 2.0000, 12.0000, 0.2000]],
[[ -1.0000, 11.0000, 1.5000],
[ 0.6000, 88.0000, 1.0000],
[ 0.3000, 3.6000, -10.0000],
[ 2.1000, 1.2000, 4.2000]]])
tensor2d.shape
torch.Size([4, 3])
tensor2d.size()
torch.Size([4, 3])
tensor1d.dtype
torch.int64
tensor3d.dtype
torch.float32
zeros1d = torch.zeros(1)
print(zeros1d)
tensor([0.])
zeros1d = torch.zeros(5)
print(zeros1d)
tensor([0., 0., 0., 0., 0.])
zeros2d = torch.zeros(2, 3)
print(zeros2d)
tensor([[0., 0., 0.],
[0., 0., 0.]])
zeros5d = torch.zeros(2, 3, 8, 6, 5)
print(zeros5d.shape)
torch.Size([2, 3, 8, 6, 5])
print(zeros5d.numel())
1440
zeros2d = torch.zeros(2, 2, dtype=torch.float64)
print(zeros2d)
tensor([[0., 0.],
[0., 0.]], dtype=torch.float64)
ones1d = torch.ones(4)
print(ones1d)
tensor([1., 1., 1., 1.])
ones2d = torch.ones(4, 8)
print(ones2d)
tensor([[1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1., 1., 1.]])
ones3d = torch.ones(1, 2, 1, dtype=torch.int16)
print(ones3d)
tensor([[[1],
[1]]], dtype=torch.int16)
full2d = torch.full((3,4), 5)
print(full2d)
tensor([[5, 5, 5, 5],
[5, 5, 5, 5],
[5, 5, 5, 5]])
eye_tensor = torch.eye(5)
print(eye_tensor)
tensor([[1., 0., 0., 0., 0.],
[0., 1., 0., 0., 0.],
[0., 0., 1., 0., 0.],
[0., 0., 0., 1., 0.],
[0., 0., 0., 0., 1.]])
arange_tensor = torch.arange(0, 10, 1)
print(arange_tensor)
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
arange_tensor = torch.arange(1.5, 2.5, 0.1)
print(arange_tensor)
tensor([1.5000, 1.6000, 1.7000, 1.8000, 1.9000, 2.0000, 2.1000, 2.2000, 2.3000,
2.4000])
linspace_tensor = torch.linspace(0, 1, 100)
print(linspace_tensor)
tensor([0.0000, 0.0101, 0.0202, 0.0303, 0.0404, 0.0505, 0.0606, 0.0707, 0.0808,
0.0909, 0.1010, 0.1111, 0.1212, 0.1313, 0.1414, 0.1515, 0.1616, 0.1717,
0.1818, 0.1919, 0.2020, 0.2121, 0.2222, 0.2323, 0.2424, 0.2525, 0.2626,
0.2727, 0.2828, 0.2929, 0.3030, 0.3131, 0.3232, 0.3333, 0.3434, 0.3535,
0.3636, 0.3737, 0.3838, 0.3939, 0.4040, 0.4141, 0.4242, 0.4343, 0.4444,
0.4545, 0.4646, 0.4747, 0.4848, 0.4949, 0.5051, 0.5152, 0.5253, 0.5354,
0.5455, 0.5556, 0.5657, 0.5758, 0.5859, 0.5960, 0.6061, 0.6162, 0.6263,
0.6364, 0.6465, 0.6566, 0.6667, 0.6768, 0.6869, 0.6970, 0.7071, 0.7172,
0.7273, 0.7374, 0.7475, 0.7576, 0.7677, 0.7778, 0.7879, 0.7980, 0.8081,
0.8182, 0.8283, 0.8384, 0.8485, 0.8586, 0.8687, 0.8788, 0.8889, 0.8990,
0.9091, 0.9192, 0.9293, 0.9394, 0.9495, 0.9596, 0.9697, 0.9798, 0.9899,
1.0000])
a = torch.rand(2,2)
b = torch.rand(2,2)
print(a)
print(b)
c = a + b
print(c)
tensor([[0.4594, 0.7425],
[0.1502, 0.2436]])
tensor([[0.6796, 0.4534],
[0.4514, 0.6632]])
tensor([[1.1390, 1.1959],
[0.6016, 0.9068]])
a = torch.rand(1, 2, 3)
print(a)
b = 2.5
c = a + b
print(c)
tensor([[[0.6750, 0.2402, 0.8859],
[0.5521, 0.3160, 0.9857]]])
tensor([[[3.1750, 2.7402, 3.3859],
[3.0521, 2.8160, 3.4857]]])
out=input+alpha×other
alpha = 0.3
c = a + alpha * b
print(c)
tensor([[[1.4250, 0.9902, 1.6359],
[1.3021, 1.0660, 1.7357]]])
d = torch.add(a, b, alpha=alpha)
print(d)
tensor([[[1.4250, 0.9902, 1.6359],
[1.3021, 1.0660, 1.7357]]])
a = torch.randint(10, (4,2))
print(a)
b = torch.randint(10, (2,))
print(b)
c = a * b
print(c)
tensor([[9, 8],
[9, 6],
[8, 1],
[1, 7]])
tensor([9, 8])
tensor([[81, 64],
[81, 48],
[72, 8],
[ 9, 56]])
a = torch.randn(4,2)
b = torch.randn(2,4)
print(a)
print(b)
c = torch.matmul(a, b)
print(c)
tensor([[-0.2558, 1.5415],
[-1.1941, 0.7168],
[-1.8735, -0.6500],
[-2.8587, -1.6622]])
tensor([[ 0.4040, 0.3098, -0.2444, -0.2476],
[-0.5253, -0.0618, 0.2590, 0.8926]])
tensor([[-0.9131, -0.1745, 0.4617, 1.4393],
[-0.8589, -0.4142, 0.4775, 0.9354],
[-0.4154, -0.5403, 0.2896, -0.1164],
[-0.2817, -0.7829, 0.2683, -0.7760]])
c = a @ b
print(c)
tensor([[-0.9131, -0.1745, 0.4617, 1.4393],
[-0.8589, -0.4142, 0.4775, 0.9354],
[-0.4154, -0.5403, 0.2896, -0.1164],
[-0.2817, -0.7829, 0.2683, -0.7760]])
a = torch.randn(5)
print(a)
torch.div(a, 0.5)
tensor([ 2.0433, -1.1218, -0.0541, 0.3179, 0.5792])
tensor([ 4.0866, -2.2436, -0.1082, 0.6359, 1.1585])
a = torch.randn(4, 4)
print(a)
b = torch.randn(4, 4)
print(b)
c = torch.div(a, b)
print(c)
tensor([[ 0.6657, 0.5462, -0.1865, 0.0868],
[ 0.8220, 1.1413, -0.1527, -1.1667],
[-0.6886, 0.3191, 0.7198, 0.7958],
[-0.3279, -0.1756, -2.1059, 1.9832]])
tensor([[-1.1737, 1.6681, 0.3638, -0.4306],
[ 0.3423, 0.1881, 0.0218, 0.4910],
[-1.8491, 0.5764, -1.3281, 1.4727],
[-1.7136, 0.1062, -0.4730, -0.6466]])
tensor([[-0.5672, 0.3275, -0.5127, -0.2015],
[ 2.4012, 6.0676, -6.9959, -2.3761],
[ 0.3724, 0.5536, -0.5420, 0.5404],
[ 0.1913, -1.6533, 4.4519, -3.0671]])
c = a/b
print(c)
tensor([[-0.5672, 0.3275, -0.5127, -0.2015],
[ 2.4012, 6.0676, -6.9959, -2.3761],
[ 0.3724, 0.5536, -0.5420, 0.5404],
[ 0.1913, -1.6533, 4.4519, -3.0671]])
#a/b = a @ inv(b)
a = torch.randn(4, 4)
print(a)
b = torch.inverse(a)
print(b)
tensor([[-0.3815, 0.2012, 0.4774, 1.9964],
[ 0.4370, -0.8975, -0.3551, -2.2378],
[-1.0331, -0.2406, -0.4446, -0.8757],
[-0.4407, 0.4014, -0.8402, 0.2170]])
tensor([[-1.4164e-03, 3.9795e-01, -9.3894e-01, 3.2788e-01],
[-1.8456e+00, -1.7272e+00, 1.1230e-01, -3.7817e-01],
[-6.6258e-01, -9.1275e-01, 4.6796e-01, -1.4285e+00],
[ 8.4510e-01, 4.6842e-01, -3.0267e-01, 4.4238e-01]])
Linear Regression
\begin{equation} y = a + bx \end{equation} $a$ is bias
$b$ is weights
import numpy as np
import torch
make fake data
$ 2 * x + 1 + \mbox{noise} = y$
np.random.seed(42)
x = np.random.rand(100, 1) #make random vector with size 100
y = 1 + (2 * x) + .1 * np.random.randn(100, 1) #make y depends on x with noise by equation y = 1 + 2 * x + (noise)
x.shape
(100, 1)
y.shape
(100, 1)
# Shuffles the indices
idx = np.arange(100)
np.random.shuffle(idx)
# Uses first 80 random indices for train
train_idx = idx[:80]
# Uses the remaining indices for validation
val_idx = idx[80:]
# Generates train and validation sets
x_train, y_train = x[train_idx], y[train_idx]
x_val, y_val = x[val_idx], y[val_idx]
x_train.shape
(80, 1)
y_train.shape
(80, 1)
import matplotlib.pyplot as plt
plt.plot(x_train, y_train, 'o')
plt.title('Generated Data (Train)')
plt.grid()
plt.xlabel('x_train')
plt.ylabel('y_train')
Text(0, 0.5, 'y_train')
gradient descent
we calculate loss and then by gradient we update our parameters
we want find $a$ and $b$
init parameters
np.random.seed(42)
a = np.random.randn(1)
b = np.random.randn(1)
print(a, b)
# Computes our model's predicted output
y = a + b * x_train
print(y.shape)
[0.49671415] [-0.1382643]
(80, 1)
loss function
\begin{equation} \mbox{MSE} = \frac{1}{N} \sum_{i = 1}^{N} (y_{i} - \hat{y}_{i})^{2} \end{equation}
\begin{equation} \mbox{MSE} = \frac{1}{N} \sum_{i = 1}^{N} (y_{i} - (a + b*x_{i}))^{2} \end{equation}
error = (y_train - y)
loss = (error ** 2).mean()
print(loss)
2.7421577700550976
gradient
\begin{equation} \frac{\partial \mbox{MSE}}{\partial a} = \frac{\partial \mbox{MSE}}{\partial \hat{y}{i}} * \frac{\partial \hat{y}{i}}{\partial a} = \cdots = -2 * \frac{1}{N} \sum_{i = 1}^{N}(y_{i} - \hat{y}_{i}) \end{equation}
\begin{equation} \frac{\partial \mbox{MSE}}{\partial b} = \frac{\partial \mbox{MSE}}{\partial \hat{y}{i}} * \frac{\partial \hat{y}{i}}{\partial b} = \cdots = -2 * \frac{1}{N} \sum_{i = 1}^{N} x_{i} (y_{i} - \hat{y}_{i}) \end{equation}
# Computes gradients for both "a" and "b" parameters
a_grad = -2 * error.mean()
b_grad = -2 * (x_train * error).mean()
print('a_grad: ', a_grad)
print('b_grad: ', b_grad)
a_grad: -3.044811379650508
b_grad: -1.8337537171510832
update parameters
\begin{equation} a = a - \mu \frac{\partial \mbox{MSE}}{\partial a} \end{equation} \begin{equation} b = b - \mu \frac{\partial \mbox{MSE}}{\partial b} \end{equation}
lr = 1e-1
# Updates parameters using gradients and the learning rate
a = a - lr * a_grad
b = b - lr * b_grad
print('a: ', a)
print('b: ', b)
a: [0.80119529]
b: [0.04511107]
Do it for 100 epochs
# Initializes parameters "a" and "b" randomly
np.random.seed(42)
a = np.random.randn(1)
b = np.random.randn(1)
print('Initial value of a: ', a)
print('Initial value of b: ', b)
# Sets learning rate
lr = 1e-1
# Defines number of epochs
n_epochs = 1000
for epoch in range(n_epochs):
# Computes our model's predicted output
yhat = a + b * x_train
# How wrong is our model? That's the error!
error = (y_train - yhat)
# It is a regression, so it computes mean squared error (MSE)
loss = (error ** 2).mean()
# Computes gradients for both "a" and "b" parameters
a_grad = -2 * error.mean()
b_grad = -2 * (x_train * error).mean()
# Updates parameters using gradients and the learning rate
a = a - lr * a_grad
b = b - lr * b_grad
print('Final value of a: ', a)
print('Final value of b: ', b)
print('Final Error : ',(((a + b * x_train) - y_train)**2).mean())
Initial value of a: [0.49671415]
Initial value of b: [-0.1382643]
Final value of a: [1.02354094]
Final value of b: [1.96896411]
Final Error : 0.008044657695505126
plt.plot(x_val, y_val, 'ro')
plt.title('Validation data and reult of the trained regression model')
plt.grid()
plt.xlabel('x_val')
plt.ylabel('y_val')
# Final regression model
yhat = a + b * x_val
plt.plot(x_val, yhat, 'g')
[<matplotlib.lines.Line2D at 0x7f37708a1040>]
make life easier by pytorch !!!
# tensor
tensor = torch.tensor([[[1, 2, 8, 2],
[3, 0, 9, 9],
[7, 2, 0, 6]],
[[0, 9, 8, 2],
[1, 8, 5, 5],
[2, 8, 1, 7]]])
tensor
tensor([[[1, 2, 8, 2],
[3, 0, 9, 9],
[7, 2, 0, 6]],
[[0, 9, 8, 2],
[1, 8, 5, 5],
[2, 8, 1, 7]]])
tensor.shape
torch.Size([2, 3, 4])
type(tensor)
torch.Tensor
# torch.tensor([[[1, 2, 8, 2],
# [3, 0, 9, 9],
# [7, 2, 0, 6]],
# [[0, 9, 8, 2],
# [1, 8, 5, 5],
# [2, 8, 1, 7]]], device=torch.device('cuda:0'))
# tensor.shape
torch.cuda.is_available()
True
#torch.cuda.get_device_properties('cuda:0')
#convert numpy array to tensor and
np_array = np.random.randint(0, 10, (2, 3))
print(np_array)
print(50*'-')
tensor = torch.from_numpy(np_array)
print(tensor)
[[7 4 6]
[9 2 6]]
--------------------------------------------------
tensor([[7, 4, 6],
[9, 2, 6]])
#convert tensor to numpy array
tensor = torch.from_numpy(np_array)
print(tensor)
print(50*'-')
np_array = tensor.numpy()
print(np_array)
tensor([[7, 4, 6],
[9, 2, 6]])
--------------------------------------------------
[[7 4 6]
[9 2 6]]
# TypeError: can't convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
# a = torch.randint(10, (2, 3), device=torch.device('cuda'))
# print(a)
# print(50*'-')
# # Method 1
# b = a.to('cpu').numpy()
# print(b)
# print(50*'-')
# # Method 2
# c = a.cpu().numpy()
# print(c)
use tensor
# Converts numpy arrays to tensor
x_train = torch.from_numpy(x_train)
y_train = torch.from_numpy(y_train)
# Initializes parameters "a" and "b" randomly
torch.manual_seed(42)
a = torch.randn(1)
b = torch.randn(1)
print('Initial value of a: ', a)
print('Initial value of b: ', b)
Initial value of a: tensor([0.3367])
Initial value of b: tensor([0.1288])
# Sets learning rate
lr = 1e-1
# Defines number of epochs
n_epochs = 1000
for epoch in range(n_epochs):
# Computes our model's predicted output
yhat = a + b * x_train
# How wrong is our model? That's the error!
error = (y_train - yhat)
# It is a regression, so it computes mean squared error (MSE)
loss = (error ** 2).mean()
# Computes gradients for both "a" and "b" parameters
a_grad = -2 * error.mean()
b_grad = -2 * (x_train * error).mean()
# Updates parameters using gradients and the learning rate
a = a - lr * a_grad
b = b - lr * b_grad
print('Final value of a: ', a)
print('Final value of b: ', b)
Final value of a: tensor([1.0235])
Final value of b: tensor([1.9690])
pytorch
x = torch.randn(1, requires_grad=True)
print('x: ', x)
y = x**2
print('y: ', y)
y.backward()
x: tensor([0.2345], requires_grad=True)
y: tensor([0.0550], grad_fn=<PowBackward0>)
x.grad
tensor([0.4689])
# optimizer
torch.manual_seed(42)
a = torch.randn(1, requires_grad=True)
b = torch.randn(1, requires_grad=True)
optimizer = torch.optim.SGD([a, b], lr=0.01)
print(optimizer)
# update parameters
# optimizer.step()
SGD (
Parameter Group 0
dampening: 0
differentiable: False
foreach: None
lr: 0.01
maximize: False
momentum: 0
nesterov: False
weight_decay: 0
)
np.random.seed(42)
x = np.random.rand(100, 1) #make random vector with size 100
y = 1 + 2 * x + .1 * np.random.randn(100, 1) #make y depends on x with noise by equation y = 1 + 2 * x + (noise)
# Shuffles the indices
idx = np.arange(100)
np.random.shuffle(idx)
# Uses first 80 random indices for train
train_idx = idx[:80]
# Uses the remaining indices for validation
val_idx = idx[80:]
# Generates train and validation sets
x_train, y_train = x[train_idx], y[train_idx]
x_val, y_val = x[val_idx], y[val_idx]
# Converts numpy arrays to tensor
x_train = torch.from_numpy(x_train)
y_train = torch.from_numpy(y_train)
# Initializes parameters "a" and "b" randomly
torch.manual_seed(42)
a = torch.randn(1, requires_grad=True)
b = torch.randn(1, requires_grad=True)
print('Initial value of a: ', a)
print('Initial value of b: ', b)
Initial value of a: tensor([0.3367], requires_grad=True)
Initial value of b: tensor([0.1288], requires_grad=True)
# Sets learning rate
lr = 1e-1
# Defines optimizer
optimizer = torch.optim.SGD([a, b], lr=lr)
# Defines number of epochs
n_epochs = 1000
for epoch in range(n_epochs):
# Computes our model's predicted output
yhat = a + b * x_train
# How wrong is our model? That's the error!
error = (y_train - yhat)
# It is a regression, so it computes mean squared error (MSE)
loss = (error ** 2).mean()
# Computes gradients for both "a" and "b" parameters
# a_grad = -2 * error.mean()
# b_grad = -2 * (x_train * error).mean()
loss.backward() # simple!
# Updates parameters using gradients and the learning rate
# a = a - lr * a_grad
# b = b - lr * b_grad
optimizer.step() # simple!
# Resets gradients!
optimizer.zero_grad()
print('Final value of a: ', a)
print('Final value of b: ', b)
Final value of a: tensor([1.0235], requires_grad=True)
Final value of b: tensor([1.9690], requires_grad=True)
# Resets gradients!
x = torch.tensor(1., requires_grad=True)
# First try
y = x**3
y.backward()
print('Gradient: ', x.grad)
Gradient: tensor(3.)
# Second try
y = x**3
y.backward()
print('Gradient: ', x.grad)
Gradient: tensor(6.)
pytorch loss function
loss = torch.nn.MSELoss()
input = torch.randn(3, 5, requires_grad=True)
target = torch.randn(3, 5)
output = loss(input, target)
print(output)
tensor(0.9842, grad_fn=<MseLossBackward0>)
np.random.seed(42)
x = np.random.rand(100, 1) #make random vector with size 100
y = 1 + 2 * x + .1 * np.random.randn(100, 1) #make y depends on x with noise by equation y = 1 + 2 * x + (noise)
# Shuffles the indices
idx = np.arange(100)
np.random.shuffle(idx)
# Uses first 80 random indices for train
train_idx = idx[:80]
# Uses the remaining indices for validation
val_idx = idx[80:]
# Generates train and validation sets
x_train, y_train = x[train_idx], y[train_idx]
x_val, y_val = x[val_idx], y[val_idx]
# Converts numpy arrays to tensor
x_train = torch.from_numpy(x_train)
y_train = torch.from_numpy(y_train)
# Initializes parameters "a" and "b" randomly
torch.manual_seed(42)
a = torch.randn(1, requires_grad=True)
b = torch.randn(1, requires_grad=True)
print('Initial value of a: ', a)
print('Initial value of b: ', b)
Initial value of a: tensor([0.3367], requires_grad=True)
Initial value of b: tensor([0.1288], requires_grad=True)
# Sets learning rate
lr = 1e-1
# Defines optimizer
optimizer = torch.optim.SGD([a, b], lr=lr)
# Defines MSE loss
loss_fn = torch.nn.MSELoss()
# Defines number of epochs
n_epochs = 1000
for epoch in range(n_epochs):
# Computes our model's predicted output
yhat = a + b * x_train
# How wrong is our model? That's the error!
# error = (y_train - yhat)
# It is a regression, so it computes mean squared error (MSE)
# loss = (error ** 2).mean()
loss = loss_fn(yhat, y_train)
# Computes gradients for both "a" and "b" parameters
# a_grad = -2 * error.mean()
# b_grad = -2 * (x_train * error).mean()
loss.backward() # simple!
# Updates parameters using gradients and the learning rate
# a = a - lr * a_grad
# b = b - lr * b_grad
optimizer.step() # simple!
# Resets gradients!
optimizer.zero_grad()
yhat = a + b * x_train
final_error = loss_fn(yhat, y_train)
print('Final value of a: ', a)
print('Final value of b: ', b)
print('Final loss : ',final_error)
Final value of a: tensor([1.0235], requires_grad=True)
Final value of b: tensor([1.9690], requires_grad=True)
Final loss : tensor(0.0080, dtype=torch.float64, grad_fn=<MseLossBackward0>)
pytorch FC
class NeuralNetwork(torch.nn.Module):
def __init__(self, n_input, n_unit1, n_output):
super().__init__()
# Inputs to 1st hidden layer linear transformation
self.hidden = torch.nn.Linear(n_input, n_unit1)
self.sigmoid = torch.nn.Sigmoid()
# Output layer
self.output = torch.nn.Linear(n_unit1, n_output)
self.softmax = torch.nn.Softmax(dim=1)
def forward(self, x):
x = self.hidden(x)
x = self.sigmoid(x)
x = self.output(x)
x = self.softmax(x)
return x
nn_model = NeuralNetwork(10, 30, 2)
print(nn_model)
NeuralNetwork(
(hidden): Linear(in_features=10, out_features=30, bias=True)
(sigmoid): Sigmoid()
(output): Linear(in_features=30, out_features=2, bias=True)
(softmax): Softmax(dim=1)
)
model = torch.nn.Sequential(
torch.nn.Linear(10, 30),
torch.nn.Sigmoid(),
torch.nn.Linear(30, 2),
torch.nn.Softmax()
)
print(model)
Sequential(
(0): Linear(in_features=10, out_features=30, bias=True)
(1): Sigmoid()
(2): Linear(in_features=30, out_features=2, bias=True)
(3): Softmax(dim=None)
)
class LinRegModel(torch.nn.Module):
def __init__(self):
super().__init__()
# Inputs to 1st hidden layer linear transformation
self.neuron = torch.nn.Linear(1, 1)
def forward(self, x):
x = x.float()
y = self.neuron(x)
return y.float()
lreg_model = LinRegModel()
print(lreg_model)
LinRegModel(
(neuron): Linear(in_features=1, out_features=1, bias=True)
)
np.random.seed(42)
x = np.random.rand(100, 1) #make random vector with size 100
y = 1 + 2 * x + .1 * np.random.randn(100, 1) #make y depends on x with noise by equation y = 1 + 2 * x + (noise)
# Shuffles the indices
idx = np.arange(100)
np.random.shuffle(idx)
# Uses first 80 random indices for train
train_idx = idx[:80]
# Uses the remaining indices for validation
val_idx = idx[80:]
# Generates train and validation sets
x_train, y_train = x[train_idx], y[train_idx]
x_val, y_val = x[val_idx], y[val_idx]
# Converts numpy arrays to tensor
x_train = torch.from_numpy(x_train)
x_train = x_train.float()
y_train = torch.from_numpy(y_train)
y_train = y_train.float()
# Sets learning rate
lr = 1e-1
# training step
lreg_model.train()
# Defines optimizer
optimizer = torch.optim.SGD(lreg_model.parameters(), lr=lr)
# Defines MSE loss
loss_fn = torch.nn.MSELoss()
# Defines number of epochs
n_epochs = 1000
# evaluate model
# model.eval()
# with torch.no_grad():
# ...
# out_data = model(data)
# ...
for epoch in range(n_epochs):
# Computes our model's predicted output
# yhat = a + b * x_train
yhat = lreg_model(x_train)
# How wrong is our model? That's the error!
# error = (y_train - yhat)
# It is a regression, so it computes mean squared error (MSE)
# loss = (error ** 2).mean()
loss = loss_fn(yhat, y_train)
# Computes gradients for both "a" and "b" parameters
# a_grad = -2 * error.mean()
# b_grad = -2 * (x_train * error).mean()
loss.backward() # simple!
# Updates parameters using gradients and the learning rate
# a = a - lr * a_grad
# b = b - lr * b_grad
optimizer.step() # simple!
# Resets gradients!
optimizer.zero_grad()
# print('Final value of a: ', a)
# print('Final value of b: ', a)
print(lreg_model.state_dict())
OrderedDict([('neuron.weight', tensor([[1.9690]])), ('neuron.bias', tensor([1.0235]))])
Classification
pre-process data
# lets work with MNIST dataset!
from sklearn.datasets import load_digits
digits = load_digits()
digits.keys()
dict_keys(['data', 'target', 'frame', 'feature_names', 'target_names', 'images', 'DESCR'])
digits['data'].shape
(1797, 64)
digits['data'][0]
array([ 0., 0., 5., 13., 9., 1., 0., 0., 0., 0., 13., 15., 10.,
15., 5., 0., 0., 3., 15., 2., 0., 11., 8., 0., 0., 4.,
12., 0., 0., 8., 8., 0., 0., 5., 8., 0., 0., 9., 8.,
0., 0., 4., 11., 0., 1., 12., 7., 0., 0., 2., 14., 5.,
10., 12., 0., 0., 0., 0., 6., 13., 10., 0., 0., 0.])
digits['target']
array([0, 1, 2, ..., 8, 9, 8])
digits.images.shape
(1797, 8, 8)
import matplotlib.pyplot as plt
plt.figure(figsize=(10,4))
plt.title(f'label: {digits.target[0]}')
plt.imshow(digits.images[0], cmap='gray')
<matplotlib.image.AxesImage at 0x7f6e7d3207f0>
# build csv file of meta data
import pandas as pd
file_names = [f'/content/images/{i}.jpg' for i in range(100)]
df = pd.DataFrame({'file_name':file_names , 'target':digits.target[:100]})
df.head()
| file_name | target | |
|---|---|---|
| 0 | /content/images/0.jpg | 0 |
| 1 | /content/images/1.jpg | 1 |
| 2 | /content/images/2.jpg | 2 |
| 3 | /content/images/3.jpg | 3 |
| 4 | /content/images/4.jpg | 4 |
df['mode'] = 'train'
test_size = int(len(df) * 0.2)
ix = df.sample(test_size).index
df.loc[ix, 'mode'] = 'test'
df.sample(5)
| file_name | target | mode | |
|---|---|---|---|
| 92 | /content/images/92.jpg | 9 | train |
| 83 | /content/images/83.jpg | 3 | train |
| 5 | /content/images/5.jpg | 5 | train |
| 93 | /content/images/93.jpg | 1 | train |
| 16 | /content/images/16.jpg | 6 | train |
df['mode'].value_counts()
train 80
test 20
Name: mode, dtype: int64
df.to_csv('data.csv', index=False)
# save mnist images as files!
# build folder
! mkdir 'images'
import imageio
for i, image in enumerate(digits.images[:100]):
file_name = f'/content/images/{i}.jpg'
imageio.imwrite(file_name, image)
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 15.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 15.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 15.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
WARNING:imageio:Lossy conversion from float64 to uint8. Range [0.0, 16.0]. Convert image to uint8 prior to saving to suppress this warning.
# build dataset
from torch.utils.data import Dataset
class CustomImageDataset(Dataset):
def __init__(self, data_path, mode):
self.data = pd.read_csv(data_path)
self.mode = mode
# filter train/test data
self.data = self.data[self.data['mode'] == self.mode]
self.data.reset_index(drop=True, inplace=True)
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
record = self.data.iloc[idx]
image_path = record['file_name']
label = record['target']
if self.mode == 'train':
label_one_hot = [0.0]*10
label_one_hot[label] = 1
else:
label_one_hot = label
# read image file
# we can use different package like opencv,imageio,PIL,matplotlib,...
image_array = plt.imread(image_path)
image_array = image_array.reshape(64)
label_one_hot = torch.tensor(label_one_hot)
image_array = torch.tensor(image_array)
return image_array, label_one_hot
data_path = '/content/data.csv'
train_dataset = CustomImageDataset(data_path=data_path, mode='train')
test_dataset = CustomImageDataset(data_path=data_path, mode='test')
next(iter(train_dataset))
(tensor([ 0, 2, 88, 217, 158, 21, 0, 0, 16, 0, 228, 236, 166, 253,
95, 0, 0, 41, 255, 44, 11, 182, 132, 3, 11, 80, 220, 0,
2, 143, 133, 5, 6, 71, 122, 0, 5, 155, 140, 0, 0, 76,
194, 0, 26, 207, 105, 15, 0, 18, 255, 83, 146, 211, 4, 0,
4, 3, 87, 228, 182, 0, 9, 2], dtype=torch.uint8),
tensor([1., 0., 0., 0., 0., 0., 0., 0., 0., 0.]))
next(iter(test_dataset))
(tensor([ 0, 7, 145, 223, 128, 13, 0, 2, 0, 0, 204, 215, 228, 199,
5, 0, 3, 5, 128, 164, 0, 231, 59, 10, 0, 0, 74, 241,
186, 238, 38, 0, 7, 0, 69, 255, 255, 13, 15, 14, 1, 53,
238, 126, 159, 207, 21, 3, 0, 19, 248, 29, 43, 255, 133, 0,
0, 3, 169, 249, 245, 163, 26, 1], dtype=torch.uint8),
tensor(8))
from torch.utils.data import DataLoader
train_dataloader = DataLoader(train_dataset, batch_size=5, shuffle=True)
test_dataloader = DataLoader(test_dataset, batch_size=20, shuffle=True)
# build network !
import torch.nn as nn
class NeuralNetwork(torch.nn.Module):
def __init__(self):
super().__init__()
self.hidden = torch.nn.Linear(64, 32)
self.sigmoid = torch.nn.Sigmoid()
self.hidden1 = torch.nn.Linear(32, 16)
# Output layer
self.output = torch.nn.Linear(16, 10)
def forward(self, x):
x = x.to(torch.float32)
x = self.hidden(x)
x = self.sigmoid(x)
x = self.hidden1(x)
x = self.output(x)
return x
net = NeuralNetwork()
net
NeuralNetwork(
(hidden): Linear(in_features=64, out_features=32, bias=True)
(sigmoid): Sigmoid()
(hidden1): Linear(in_features=32, out_features=16, bias=True)
(output): Linear(in_features=16, out_features=10, bias=True)
)
import torch.optim as optim
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# Assuming that we are on a CUDA machine, this should print a CUDA device:
print(device)
cuda:0
# if we can gpu, let use it!
net.to(device)
NeuralNetwork(
(hidden): Linear(in_features=64, out_features=32, bias=True)
(sigmoid): Sigmoid()
(hidden1): Linear(in_features=32, out_features=16, bias=True)
(output): Linear(in_features=16, out_features=10, bias=True)
)
no_epochs = 10
for epoch in range(no_epochs):
for i, data in enumerate(train_dataloader, 0):
# get the inputs; data is a list of [inputs, labels]
inputs, labels = data
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward + backward + optimize
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
print(f'epoch:{epoch + 1}, batch: {i+1}, loss: {loss.item()}')
epoch:1, batch: 1, loss: 2.1908631324768066
epoch:1, batch: 2, loss: 2.224562168121338
epoch:1, batch: 3, loss: 2.406986951828003
epoch:1, batch: 4, loss: 2.4482572078704834
epoch:1, batch: 5, loss: 2.319631338119507
epoch:1, batch: 6, loss: 2.207183599472046
epoch:1, batch: 7, loss: 2.2555484771728516
epoch:1, batch: 8, loss: 2.279235601425171
epoch:1, batch: 9, loss: 2.2593040466308594
epoch:1, batch: 10, loss: 2.286088705062866
epoch:1, batch: 11, loss: 2.545666456222534
epoch:1, batch: 12, loss: 2.4073219299316406
epoch:1, batch: 13, loss: 2.331043004989624
epoch:1, batch: 14, loss: 2.3777008056640625
epoch:1, batch: 15, loss: 2.261925458908081
epoch:1, batch: 16, loss: 2.1023812294006348
epoch:2, batch: 1, loss: 2.1612069606781006
epoch:2, batch: 2, loss: 2.2677056789398193
epoch:2, batch: 3, loss: 2.446007490158081
epoch:2, batch: 4, loss: 2.2930259704589844
epoch:2, batch: 5, loss: 2.2307610511779785
epoch:2, batch: 6, loss: 2.1262547969818115
epoch:2, batch: 7, loss: 2.174713373184204
epoch:2, batch: 8, loss: 2.4077744483947754
epoch:2, batch: 9, loss: 2.1118643283843994
epoch:2, batch: 10, loss: 2.1846072673797607
epoch:2, batch: 11, loss: 2.208353042602539
epoch:2, batch: 12, loss: 2.111048936843872
epoch:2, batch: 13, loss: 2.3346307277679443
epoch:2, batch: 14, loss: 2.375053644180298
epoch:2, batch: 15, loss: 2.3144726753234863
epoch:2, batch: 16, loss: 2.29840350151062
epoch:3, batch: 1, loss: 2.3953609466552734
epoch:3, batch: 2, loss: 2.074617624282837
epoch:3, batch: 3, loss: 2.0533435344696045
epoch:3, batch: 4, loss: 2.272300958633423
epoch:3, batch: 5, loss: 2.291416883468628
epoch:3, batch: 6, loss: 2.365598678588867
epoch:3, batch: 7, loss: 2.2665815353393555
epoch:3, batch: 8, loss: 2.2684059143066406
epoch:3, batch: 9, loss: 2.0873420238494873
epoch:3, batch: 10, loss: 2.2886240482330322
epoch:3, batch: 11, loss: 2.3690507411956787
epoch:3, batch: 12, loss: 2.2202694416046143
epoch:3, batch: 13, loss: 2.313382863998413
epoch:3, batch: 14, loss: 2.104445457458496
epoch:3, batch: 15, loss: 2.260211706161499
epoch:3, batch: 16, loss: 2.050424575805664
epoch:4, batch: 1, loss: 2.464210271835327
epoch:4, batch: 2, loss: 2.3058598041534424
epoch:4, batch: 3, loss: 2.3367178440093994
epoch:4, batch: 4, loss: 2.1102194786071777
epoch:4, batch: 5, loss: 2.1745903491973877
epoch:4, batch: 6, loss: 2.1484971046447754
epoch:4, batch: 7, loss: 2.1926991939544678
epoch:4, batch: 8, loss: 2.150862216949463
epoch:4, batch: 9, loss: 1.9954556226730347
epoch:4, batch: 10, loss: 2.0690479278564453
epoch:4, batch: 11, loss: 2.132089853286743
epoch:4, batch: 12, loss: 2.0676677227020264
epoch:4, batch: 13, loss: 2.3002305030822754
epoch:4, batch: 14, loss: 2.38397479057312
epoch:4, batch: 15, loss: 2.123269557952881
epoch:4, batch: 16, loss: 2.004640579223633
epoch:5, batch: 1, loss: 2.1065430641174316
epoch:5, batch: 2, loss: 2.1438424587249756
epoch:5, batch: 3, loss: 2.191101312637329
epoch:5, batch: 4, loss: 2.313643217086792
epoch:5, batch: 5, loss: 2.162065267562866
epoch:5, batch: 6, loss: 2.1472058296203613
epoch:5, batch: 7, loss: 2.22294020652771
epoch:5, batch: 8, loss: 2.1290218830108643
epoch:5, batch: 9, loss: 2.1258203983306885
epoch:5, batch: 10, loss: 2.240772008895874
epoch:5, batch: 11, loss: 2.1791350841522217
epoch:5, batch: 12, loss: 2.0856432914733887
epoch:5, batch: 13, loss: 2.0481395721435547
epoch:5, batch: 14, loss: 1.9638092517852783
epoch:5, batch: 15, loss: 2.2188031673431396
epoch:5, batch: 16, loss: 2.0778210163116455
epoch:6, batch: 1, loss: 2.144806146621704
epoch:6, batch: 2, loss: 1.9162601232528687
epoch:6, batch: 3, loss: 2.0913197994232178
epoch:6, batch: 4, loss: 2.2266457080841064
epoch:6, batch: 5, loss: 1.9286491870880127
epoch:6, batch: 6, loss: 2.006355047225952
epoch:6, batch: 7, loss: 2.143359899520874
epoch:6, batch: 8, loss: 2.02276873588562
epoch:6, batch: 9, loss: 2.2432315349578857
epoch:6, batch: 10, loss: 2.1693453788757324
epoch:6, batch: 11, loss: 2.2603983879089355
epoch:6, batch: 12, loss: 2.103860855102539
epoch:6, batch: 13, loss: 2.0979373455047607
epoch:6, batch: 14, loss: 2.1934289932250977
epoch:6, batch: 15, loss: 2.1618306636810303
epoch:6, batch: 16, loss: 2.1570911407470703
epoch:7, batch: 1, loss: 1.9818881750106812
epoch:7, batch: 2, loss: 2.334798574447632
epoch:7, batch: 3, loss: 2.0481016635894775
epoch:7, batch: 4, loss: 2.0699918270111084
epoch:7, batch: 5, loss: 2.1774942874908447
epoch:7, batch: 6, loss: 2.181626558303833
epoch:7, batch: 7, loss: 2.166632652282715
epoch:7, batch: 8, loss: 1.9819879531860352
epoch:7, batch: 9, loss: 1.9898792505264282
epoch:7, batch: 10, loss: 2.0292041301727295
epoch:7, batch: 11, loss: 2.022294282913208
epoch:7, batch: 12, loss: 2.221928358078003
epoch:7, batch: 13, loss: 1.9757635593414307
epoch:7, batch: 14, loss: 2.09120774269104
epoch:7, batch: 15, loss: 2.0597217082977295
epoch:7, batch: 16, loss: 2.1298887729644775
epoch:8, batch: 1, loss: 2.0941884517669678
epoch:8, batch: 2, loss: 2.136359214782715
epoch:8, batch: 3, loss: 1.9520286321640015
epoch:8, batch: 4, loss: 2.267650604248047
epoch:8, batch: 5, loss: 2.1401755809783936
epoch:8, batch: 6, loss: 1.916703462600708
epoch:8, batch: 7, loss: 2.0482423305511475
epoch:8, batch: 8, loss: 1.917816162109375
epoch:8, batch: 9, loss: 2.0706393718719482
epoch:8, batch: 10, loss: 2.071678638458252
epoch:8, batch: 11, loss: 2.1015563011169434
epoch:8, batch: 12, loss: 1.9428249597549438
epoch:8, batch: 13, loss: 2.277313709259033
epoch:8, batch: 14, loss: 1.9660637378692627
epoch:8, batch: 15, loss: 2.267744541168213
epoch:8, batch: 16, loss: 2.0209202766418457
epoch:9, batch: 1, loss: 1.977026343345642
epoch:9, batch: 2, loss: 1.980362892150879
epoch:9, batch: 3, loss: 2.0005500316619873
epoch:9, batch: 4, loss: 2.1845550537109375
epoch:9, batch: 5, loss: 1.9319108724594116
epoch:9, batch: 6, loss: 2.2868587970733643
epoch:9, batch: 7, loss: 2.0340218544006348
epoch:9, batch: 8, loss: 2.094358444213867
epoch:9, batch: 9, loss: 2.0003163814544678
epoch:9, batch: 10, loss: 2.023909330368042
epoch:9, batch: 11, loss: 2.0238759517669678
epoch:9, batch: 12, loss: 2.014430284500122
epoch:9, batch: 13, loss: 2.0417122840881348
epoch:9, batch: 14, loss: 2.031055450439453
epoch:9, batch: 15, loss: 1.9574825763702393
epoch:9, batch: 16, loss: 2.12681245803833
epoch:10, batch: 1, loss: 2.175635576248169
epoch:10, batch: 2, loss: 2.1380021572113037
epoch:10, batch: 3, loss: 1.9599369764328003
epoch:10, batch: 4, loss: 1.9095171689987183
epoch:10, batch: 5, loss: 2.0523219108581543
epoch:10, batch: 6, loss: 1.9941107034683228
epoch:10, batch: 7, loss: 1.8930692672729492
epoch:10, batch: 8, loss: 1.8279964923858643
epoch:10, batch: 9, loss: 1.901199221611023
epoch:10, batch: 10, loss: 2.0580239295959473
epoch:10, batch: 11, loss: 2.076740026473999
epoch:10, batch: 12, loss: 2.0451810359954834
epoch:10, batch: 13, loss: 1.9357643127441406
epoch:10, batch: 14, loss: 1.924949049949646
epoch:10, batch: 15, loss: 2.1606929302215576
epoch:10, batch: 16, loss: 2.0498740673065186
# lets visualize network graph
!brew install graphviz
!pip install torchviz
/bin/bash: brew: command not found
Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/
Collecting torchviz
Downloading torchviz-0.0.2.tar.gz (4.9 kB)
Preparing metadata (setup.py) ... [?25l[?25hdone
Requirement already satisfied: torch in /usr/local/lib/python3.8/dist-packages (from torchviz) (1.13.1+cu116)
Requirement already satisfied: graphviz in /usr/local/lib/python3.8/dist-packages (from torchviz) (0.10.1)
Requirement already satisfied: typing-extensions in /usr/local/lib/python3.8/dist-packages (from torch->torchviz) (4.5.0)
Building wheels for collected packages: torchviz
Building wheel for torchviz (setup.py) ... [?25l[?25hdone
Created wheel for torchviz: filename=torchviz-0.0.2-py3-none-any.whl size=4151 sha256=030220c2b38c9333a7a7d7090dea5d04906438c9890c83b4647d7e3b3b323480
Stored in directory: /root/.cache/pip/wheels/05/7d/1b/8306781244e42ede119edbb053bdcda1c1f424ca226165a417
Successfully built torchviz
Installing collected packages: torchviz
Successfully installed torchviz-0.0.2
from torchviz import make_dot
image = next(iter(train_dataset))[0]
image = image.to(device)
output = net(image)
make_dot(output, params=dict(net.named_parameters()))
correct = 0
total = 0
# since we're not training, we don't need to calculate the gradients for our outputs
with torch.no_grad():
for data in test_dataloader:
inputs, labels = data
inputs = inputs.to(device)
labels = labels.to(device)
# calculate outputs by running images through the network
outputs = net(inputs)
# the class with the highest energy is what we choose as prediction
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print(f'Accuracy of the network on the 20 test images: {100 * correct // total} %')
Accuracy of the network on the 20 test images: 70 %
test_dataloader1 = DataLoader(test_dataset, batch_size=1, shuffle=True)
with torch.no_grad():
for data in test_dataloader1:
inputs, labels = data
image = inputs.numpy().reshape(8,8)
inputs = inputs.to(device)
outputs = net(inputs)
_, predicted = torch.max(outputs.data, 1)
plt.figure(figsize=(4,4))
plt.title(f'true label :{labels}, predict label {predicted}')
plt.imshow(image, cmap='gray')
