L4 Linear Regression
Simple Linear Regression Basics
• Linear regression is a linear model, a model that assumes a linear
relationship between the input variables (x) and the single output
variable (y).
• More specifically, that y can be calculated from a linear combination
of the input variables (x).
• When there is a single input variable (x), the method is referred to as
simple linear regression.
• When there are multiple input variables, multiple linear regression.
Simple Linear Regression
• Allows us to understand relationship between two continuous
variables X Y
• Example
5 40
• x: independent variable
7 120
• weight
• y: dependent variable 12 180
• height 16 210
• y=wx+b 20 240
Simple Linear Regression
• Aim of Linear Regression
• Minimize the distance between the points and the line (y=αx+β)
• Adjusting 300
• Coefficient: w 250
• Bias/intercept: b 200
• Learnable parameters: w, b 150
y
Set of training points
Predicted Fit
100
50
0
0 5 10 15 20 25
x
Simple Linear Regression
• MSE Loss: Mean Squared Error
Simple Linear Regression
• Review of Gradient Descent
Building a Linear Regression Model with
PyTorch
# Create the tensors x and y. They are the training
# examples in the dataset for the linear regression (Labelled Dataset)
x = torch.tensor([5.0, 7.0, 12.0, 16.0, 20.0])
y = torch.tensor([40.0, 120.0, 180.0, 210.0, 240.0])
# The learning rate is set to alpha = 0.001
learning_rate = torch.tensor(0.001)
Building a Linear Regression Model with
PyTorch
• # The parameters to be learnt w, and b in the
• # prediction y_p = wx +b
b = torch.rand([1], requires_grad=True)
w = torch.rand([1], requires_grad=True)
print("The parameters are {}, and {}".format(w, b))
Simple Linear Regression Basics
• Use the simple linear equation y=wx+b as an example
• The parameters of the equation are
• slope w
• intercept b
• 2 ways
1. Building simple regression
2. Building a Custom Linear Class
Simple Linear Regression
• one-dimensional linear regression having only two parameters.
• Create the linear expression Y=3x+1
• Define the parameters w and b as tensors in PyTorch.
• set the requires_grad parameter to True, indicating that our model has to
learn these parameters:
import torch
# defining the parameters 'w' and 'b'
w = torch.tensor(3.0, requires_grad = True)
b = torch.tensor(1.0, requires_grad = True)
Simple Linear Regression
• Write a function to make predictions for y at any given value of x
• In PyTorch prediction step is called forward step.
# function of the linear equation for making predictions
def forward(x):
y_pred = w * x + b
return y_pred
# predict y_pred at x = 2
x = torch.tensor([[2.0]])
y_pred = forward(x)
print("prediction of y at 'x = 2' is: ", y_pred)
Simple Linear Regression– complete code
import torch
# defining the parameters 'w' and 'b'
w = torch.tensor(3.0, requires_grad = True)
b = torch.tensor(1.0, requires_grad = True)
# function of the linear equation for making predictions
def forward(x):
y_pred = w * x + b
return y_pred
# predict y_pred at x = 2
x = torch.tensor([[2.0]])
y_pred = forward(x)
print("prediction of y at 'x = 2' is: ", y_pred)
# making predictions at multiple values of x
x = torch.tensor([[3.0], [4.0]])
y_pred = forward(x)
print("prediction of y at 'x = 3 & 4' is: ", y_pred)
Simple Linear Regression
prediction of y at 'x = 2' is: tensor([[7.]], grad_fn=<AddBackward0>)
prediction of y at 'x = 3 & 4' is: tensor([[10.],
[13.]], grad_fn=<AddBackward0>)
Train a simple linear regression model
• In order to train a linear regression model, we need to define a cost function and
an optimizer.
• The cost function is used to measure how well our model fits the data,
• while the optimizer decides which direction to move in order to improve this fit.
• Training the model and updating the parameters after going through a single
iteration of training data is known as one epoch
• We should train the model for several epochs so that weights and biases can
learn the linear relationship between the input features and output labels.
• So far, used simple predictions with only a linear regression forward pass
• Now, train a linear regression model and update its learning parameters
Building a Linear Regression Model with
PyTorch
Building a Linear Regression Model with PyTorch
# end of for loop
Note: Resetting gradients to zero because PyTorch accumulates gradients
Problem 1
• Find the value of w.grad and b.grad using analytical solution for the
given linear regression problem.
• Initial value of w = b =1
• MSE Loss
x y
2 20
4 40
Problem 1– Verify loss, w.grad, b.grad, new w, new b
x = torch.tensor([2, 4]) loss.backward()
y= torch.tensor([20, 40]) print("w.grad=", w.grad)
b = torch.tensor([1.0], requires_grad=True) with torch.no_grad():
w = torch.tensor([1.0], requires_grad=True) w -= learning_rate * w.grad
learning_rate = torch.tensor(0.001) b -= learning_rate * b.grad
loss = 0.0 w.grad.zero_()
for j in range(len(x)):
b.grad.zero_()
a = w * x[j]
y_p = a + b
loss += ( y_p-y[j]) ** 2
loss = loss / len(x)
print("avg loss=", loss)
Sample data – Verify loss, w.grad, b.grad, new w, new b
x = torch.tensor([2, 4]) loss.backward()
y= torch.tensor([20, 40]) print("w.grad=", w.grad)
b = torch.tensor([1.0], requires_grad=True) with torch.no_grad():
w = torch.tensor([1.0], requires_grad=True) w -= learning_rate * w.grad
learning_rate = torch.tensor(0.001) b -= learning_rate * b.grad
loss = 0.0 w.grad.zero_()
for j in range(len(x)):
b.grad.zero_()
a = w * x[j]
y_p = a + b w.grad= tensor([-174.])
b.grad= tensor([-52.])
loss += ( y_p-y[j]) ** 2
The parameters are
loss = loss / len(x) w=1.1740000247955322,
print("avg loss=", loss) b=1.0520000457763672, and
loss=757.0
Complete code with epoch
import torch for epochs in range(10):
from matplotlib import pyplot as plt #Compute the average loss for the training samples
# Create the tensors x and y. They are the training loss = 0.0
# examples in the dataset for the linear regression #Accumulate the loss for all the samples
for j in range(len(x)):
x = torch.tensor([2, 4])
a = w * x[j]
y= torch.tensor([20, 40]) y_p = a + b
b = torch.tensor([1.0], requires_grad=True) loss += ( y_p-y[j]) ** 2
w = torch.tensor([1.0], requires_grad=True) #loss += (y[j]-y_p)**2
print("The parameters are {}, and {}".format(w, b)) #Find the average loss
# The learning rate is set to alpha = 0.001 loss = loss / len(x)
print("avg loss=", loss)
learning_rate = torch.tensor(0.001)
#Add the loss to a list for the plotting purpose
#The list of loss values for the plotting purpose loss_list.append(loss.item())
loss_list = []
Additional question: Try w.grad if loss=(y[j]-y_p)**2
Complete code with epoch
#Compute the gradients using backward w.grad.zero_()
# dl/dw and dl/db b.grad.zero_()
loss.backward()
#w.grad = None
print("w.grad=", w.grad) #b.grad = None
# Without modifying the gradient in this block
# perform the operation print("The parameters are w={}, b={}, and
with torch.no_grad(): loss={}".format(w.item(), b.item(), loss.item()))
# Update the weight based on gradient descent #Display the plot – End of outer epoch for loop
plt.plot(loss_list)
w -= learning_rate * w.grad plt.show()
b -= learning_rate * b.grad print(loss.item(), loss.data)
Revised Version of the Implementation
Revised Version of the Implementation
Revised Version of the Implementation
Revised Version of the Implementation
Simple Linear Regression Basics -Building a Custom Linear Class
• PyTorch offers the possibility to build custom linear class.
• This method will be used for building complex models.
• Start by importing the nn module from PyTorch in order to build a custom linear class.
• from torch import nn
• Custom modules in PyTorch are classes derived from nn.Module.
• Build a class for simple linear regression and name it as Linear_Regression. make it a
child class of the nn.Module.
• So all the methods and attributes will be inherited into this class.
• In the object constructor, declare the input and output parameters.
• Also, create a super constructor to call linear class from the nn.Module.
• In order to generate prediction from the input samples, define a forward function in the
class.
• Once we have created the dataset, define the model architecture. The code is shown
next
Simple Linear Regression Basics -Building a Custom Linear Class
class Linear_Regression(nn.Module):
def __init__(self, input_sample, output_sample):
# Inheriting properties from the parent calss
super(Linear_Regression, self).__init__()
self.linear = nn.Linear(input_sample, output_sample)
# define function to make predictions
def forward(self, x):
output = self.linear(x)
return output
Linear Regression using PyTorch built-ins
• PyTorch provides the modules and classes torch.nn Dataset , and DataLoader to create and train
neural networks.
• Dataset : PyTorch’s TensorDataset is a Dataset wrapping tensors. By defining a length and way of
indexing, this also gives us a way to iterate, index, and slice along the first dimension of a tensor.
This will make it easier to access both the independent and dependent variables in the same line
as we train.
• DataLoader: Pytorch’s DataLoader is responsible for managing batches. We can create a
DataLoader from any Dataset. DataLoader makes it easier to iterate over batches.
• Model: Instead of initializing the weights & biases manually, we can define the model using
torch.nn.Linear
• Optimizer: We can make use of the built Stochastic Gradient Descent optimizer (
torch.optim.SGD)
Optim Module
• The Optim module in PyTorch has pre-written codes for most of the
optimizers that are used while building a neural network.
• We just have to import them and then they can be used to build
models.
• Ex: to get SDG Optimizer supported in PyTorch
# importing the optim module
from torch import optim
SGD = optim.SGD(model.parameters(), lr=learning_rate)
PyTorch Deep Learning Model Life-Cycle
• Life-cycle for a deep learning model and the PyTorch API that we can use to
define models
• The five steps in the life-cycle are as follows:
1. Prepare the Data
2. Define the Model
3. Train the Model
4. Evaluate the Model
5. Make Predictions
Step 1: Prepare the Data
• The first step is to load and prepare our data.
• Neural network models require numerical input data and numerical output data.
• PyTorch provides the Dataset class that we can extend and customize to load our dataset.
• the constructor of our dataset object can load our data file (e.g. a CSV file).
• We can then override
• __len__() function that can be used to get the length of the dataset (number of rows or samples), and
• __getitem__() function that is used to get a specific sample by index.
• Note: When loading our dataset, we can also perform any required transforms, such as scaling or
encoding.
• A skeleton of a custom Dataset class is provided below.
Linear Regression Implementation: Fully PyTorch way
Step 1: Prepare the Data
#dataset definition
class MyDataset(Dataset):
# load the dataset
def __init__(self, path):
# store the inputs and outputs
self.X = ...
self.y = ...
# number of rows in the dataset
def __len__(self):
return len(self.X)
# get a row at an index
def __getitem__(self, idx):
return [self.X[idx], self.y[idx]]
Data read - DataLoader
from torch.utils.data import Dataset, DataLoader
import torch
x = torch.tensor(
[12.4, 14.3, 14.5, 14.9, 16.1, 16.9, 16.5, 15.4, 17.0, 17.9, 18.8, 20.3, 22.4, 19.4, 15.5, 16.7,
17.3, 18.4, 19.2,
17.4, 19.5, 19.7, 21.2])
y = torch.tensor(
[11.2, 12.5, 12.7, 13.1, 14.1, 14.8, 14.4, 13.4, 14.9, 15.6, 16.4, 17.7, 19.6, 16.9, 14.0, 14.6,
15.1, 16.1, 16.8,
15.2, 17.0, 17.2, 18.6])
#23 values
#Find if CUDA is available to load the model and device on to the available device CPU/GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
Data read - DataLoader
class MyDataset(Dataset):
def __init__(self, X, Y):
self.X = X
self.Y = Y
def __len__(self):
return len(self.X)
def __getitem__(self, idx):
return self.X[idx].to(device), self.Y[idx].to(device)
dataset = MyDataset(x,y)
dataloader = DataLoader(dataset, batch_size=4, shuffle=True)
for data in iter(dataloader):
print(data)
Data read - DataLoader
x = torch.tensor( [12.4, 14.3, 14.5, 14.9, 16.1, 16.9, 16.5, 15.4, 17.0, 17.9, 18.8, 20.3, 22.4, 19.4, 15.5, 16.7, 17.3, 18.4, 19.2,
17.4, 19.5, 19.7, 21.2])
y = torch.tensor( [11.2, 12.5, 12.7, 13.1, 14.1, 14.8, 14.4, 13.4, 14.9, 15.6, 16.4, 17.7, 19.6, 16.9, 14.0, 14.6, 15.1, 16.1, 16.8,
15.2, 17.0, 17.2, 18.6])
#Total 23 = 4*5+3
dataloader = DataLoader(dataset, batch_size=4, shuffle=True)
[tensor([19.5000, 14.9000, 20.3000, 14.5000]), tensor([17.0000, 13.1000, 17.7000, 12.7000])]
[tensor([17.4000, 15.4000, 19.7000, 14.3000]), tensor([15.2000, 13.4000, 17.2000, 12.5000])]
[tensor([17.0000, 22.4000, 19.2000, 16.9000]), tensor([14.9000, 19.6000, 16.8000, 14.8000])]
[tensor([16.7000, 17.9000, 18.8000, 18.4000]), tensor([14.6000, 15.6000, 16.4000, 16.1000])]
[tensor([12.4000, 16.5000, 21.2000, 19.4000]), tensor([11.2000, 14.4000, 18.6000, 16.9000])]
[tensor([17.3000, 16.1000, 15.5000]), tensor([15.1000, 14.1000, 14.0000])] #3 values
dataloader = DataLoader(dataset, batch_size=4, shuffle=False)
[tensor([12.4000, 14.3000, 14.5000, 14.9000]), tensor([11.2000, 12.5000, 12.7000, 13.1000])]
[tensor([16.1000, 16.9000, 16.5000, 15.4000]), tensor([14.1000, 14.8000, 14.4000, 13.4000])]
[tensor([17.0000, 17.9000, 18.8000, 20.3000]), tensor([14.9000, 15.6000, 16.4000, 17.7000])]
[tensor([22.4000, 19.4000, 15.5000, 16.7000]), tensor([19.6000, 16.9000, 14.0000, 14.6000])]
[tensor([17.3000, 18.4000, 19.2000, 17.4000]), tensor([15.1000, 16.1000, 16.8000, 15.2000])]
[tensor([19.5000, 19.7000, 21.2000]), tensor([17.0000, 17.2000, 18.6000])] #3 values
Step 1: Prepare the Data
• Once loaded, PyTorch provides the DataLoader class to navigate a Dataset
instance during the training and evaluation of our model.
• A DataLoader instance can be created for the training dataset, test dataset, and
even a validation dataset.
• The random_split() function can be used to split a dataset into train and test sets.
• Once split, a selection of rows from the Dataset can be provided to a DataLoader,
along with the batch size and whether the data should be shuffled every epoch.
• Ex: we can define a DataLoader by passing in a selected sample of rows in the
dataset.
Step 1: Prepare the Data
...
# create the dataset
dataset = CSVDataset(...)
# select rows from the dataset
train, test = random_split(dataset, [[...], [...]])
# create a data loader for train and test sets
train_dl = DataLoader(train, batch_size=32, shuffle=True)
test_dl = DataLoader(test, batch_size=1024, shuffle=False)
Once defined, a DataLoader can be enumerated, yielding one batch worth of
samples each iteration.
...
# train the model
for i, (inputs, targets) in enumerate(train_dl):
...
Step 2: Define the Model
• The next step is to define a model.
• Defining a model in PyTorch involves defining a class that extends the Module
class.
• The constructor of our class defines the layers of the model and
• the forward() function is the override that defines how to forward propagate
input through the defined layers of the model.
• Many layers are available, such as Linear for fully connected layers, Conv2d for
convolutional layers, and MaxPool2d for pooling layers.
• Activation functions can also be defined as layers, such as ReLU, Softmax, and
Sigmoid.
• Ex: a simple MLP model with one layer
Step 2: Define the Model
The super call delegates the function call to its parent
# model definition class, which is nn.Module here.
class MLP(nn.Module): This is needed to initialize the nn.Module in a proper
manner
# define model elements A Multilayer Perceptron model, or MLP is a standard
def __init__(self, n_inputs): fully connected neural network model.
super(MLP, self).__init__()
self.layer = Linear(n_inputs, 1) It is comprised of layers of nodes where each node is
connected to all outputs from the previous layer and
self.activation = Sigmoid() the output of each node is connected to all inputs for
nodes in the next layer.
# forward propagate input
def forward(self, X): An MLP is a model with one or more fully connected
layers. This model is appropriate for tabular data, with
X = self.layer(X) one column for each variable and one row for each
X = self.activation(X) variable.
return X
Step 3: Train the Model
• The training process requires that we define a loss function and an
optimization algorithm.
• Common loss functions include the following:
• BCELoss: Binary cross-entropy loss for binary classification.
• CrossEntropyLoss: Categorical cross-entropy loss for multi-class classification.
• MSELoss: Mean squared loss for regression.
• Stochastic gradient descent is used for optimization, and the standard
algorithm is provided by the SGD class
• Ex: other versions of the algorithm are available, such as Adam.
• # define the optimization
criterion = MSELoss()
optimizer = SGD(model.parameters(), lr=0.001)
Step 3: Train the Model
• Training the model involves enumerating the DataLoader for the training
dataset.
• First, a loop is required for the number of training epochs. Then an inner
loop is required for the mini-batches for stochastic gradient descent.
...
# enumerate epochs
for epoch in range(100):
# enumerate mini batches
for i, (inputs, targets) in enumerate(train_dl):
...
Step 3: Train the Model
Each update to the model involves the same general pattern comprised
of:
• Clearing the last error gradient.
• A forward pass of the input through the model.
• Calculating the loss for the model output.
• Backpropagating the error through the model.
• Update the model in an effort to reduce loss.
Step 3: Train the Model
...
# clear the gradients
optimizer.zero_grad()
# compute the model output
yhat = model(inputs)
# calculate loss
loss = criterion(yhat, targets)
# credit assignment
loss.backward()
# update model weights
optimizer.step()
Step 4: Evaluate the model
• Once the model is fit, it can be evaluated on the test dataset.
• This can be achieved by using the DataLoader for the test dataset and
• collecting the predictions for the test set, then
• comparing the predictions to the expected values of the test set and
calculating a performance metric.
...
for i, (inputs, targets) in enumerate(test_dl):
# evaluate the model on the test set
yhat = model(inputs)
...
Step 5: Make predictions
• A fit model can be used to make a prediction on new data.
• Ex: we have a single image or a single row of data and want to make a prediction.
• This requires that we wrap the data in a PyTorch Tensor data structure.
• A Tensor is the PyTorch array for holding data. It also allows to perform the automatic
differentiation tasks in the model graph, like calling backward() when training the model.
• The prediction too will be a Tensor, although we can retrieve the NumPy array by
detaching the Tensor from the automatic differentiation graph and calling the NumPy
function.
Step 5: Make predictions
...
# convert row to data
row = Variable(Tensor([row]).float())
# make prediction
yhat = model(row)
# retrieve numpy array
yhat = yhat.detach().numpy()
Building a Linear Class
import torch model = Linear_Regression(input_sample=1,
from torch import nn output_sample=1)
torch.manual_seed(42) print("printing the model parameters: ",
list(model.parameters()))
class Linear_Regression(nn.Module): x = torch.tensor([[2.0]])
def __init__(self, input_sample, output_sample): y_pred = model(x)
# Inheriting properties from the parent calss print("getting the prediction for x: ", y_pred)
x = torch.tensor([[3.0], [4.0]])
super(Linear_Regression, self).__init__()
y_pred = model(x)
self.linear = nn.Linear(input_sample, output_sample) print("prediction of y at 'x = 3 & 4' is: ", y_pred)
# define function to make predictions
def forward(self, x):
output = self.linear(x)
return output
Building a Linear Class - Output
printing the model parameters: [Parameter containing:
tensor([[0.7645]], requires_grad=True), Parameter containing:
tensor([0.8300], requires_grad=True)]
getting the prediction for x: tensor([[2.3591]],
grad_fn=<AddmmBackward0>)
prediction of y at 'x = 3 & 4' is: tensor([[3.1236],
[3.8882]], grad_fn=<AddmmBackward0>)
Multilinear regression model
• The multilinear regression model is a supervised learning algorithm
that can be used to predict the target variable y given multiple input
variables x.
• It is a linear regression problem where more than one input variables
x or features are used to predict the target variable y.
• A typical use case of this algorithm is predicting the price of a house
given its size, number of rooms, and age.
