Basics of ML and Deep Learning

Description

Objective : The assignment aims at familiarizing you with numpy and the basics of ML and
feed-forward neural networks in python. We will be working mostly on the ipython notebooks
for the assignment. You can refer http://cs231n.github.io/ipython-tutorial/ for installing the
required software for working with ipython notebooks and getting started with it.
Assignment 1a : Multiple linear regression model
This part of the assignment consist of a python notebook that depicts the outline of using
multiple linear regression to learn a linear function to map input data to the output. The
function should be learned with the objective to minimize the Mean Square Error (MSE)
loss between the predicted output and the ground truth data.
The dataset for training and testing is provided in the assignment itself (‘Concrete_Data.csv’)
Assignment 1b : Feedforward Neural Network model
This part of the assignment consists of learning a feedforward neural network to classify
three types of flowers on the basis of certain input features. As in the previous assignment,
the objective is to decrease the categorical cross-entropy loss between the predicted
output and the ground truth data.
The dataset for training and testing is provided in the assignment itself (‘Iris_Data.csv‘)
For both the two assignments, you have to fill all the missing parts which is indicated using
question marks (?). Please don’t declare any function other than given in the notebook. The
functions have their usual meanings and is described below in brief. Some of the functions
are partially implemented and some of them are left for you to complete it.
Since the objectives are similar, we describe the common functions you need to write for
both these cases (i.e Assignment 1a and 1b).
● shuffle_dataset() : Simple function to shuffle the X and Y dataset. The dataset has to
be shuffled in place. (0.5 + 0.5)
● __init__ () : Initialize the linear regression and the neural network models by
specifying the biases and the weight matrices (2 + 5)
● forward() : Perform a forward pass by taking the entire dataset as the input for both
the models and calculates the output. Compute the predicted scores (for the LR
model) and labels ( for the NN models). (3 + 8 )
● backward() : Computes the gradients for every parameter of the network and update
the corresponding parameter using gradient descent step. This is also called a
backward pass as a prediction has been made and now the network is updated
accordingly, so as to minimize the loss function. (5 + 10)
● MSE_loss() : It computes the Mean Squared Loss between the predicted output and
ground truth data. (3)
● crossEntropy_loss(): It computes the categorical cross entropy loss between the
predicted output and ground truth data. (5)
● Accuracy: Computes the accuracy between the predicted and actual values for the
NN model (3).
● Compute the values of the following:
○ Final training loss (1+1)
○ Final test loss (1+1)
○ Final accuracy of the NN model (1)
Submission :
You have to submit two ipython notebooks with the .ipynb extension for the two
assignments . Please make sure that all the cells are running. The ipynb notebooks with the
skeleton code are given for your convenience. Please make the requisite changes there and
submit. Outline of the results mentioned above has been implemented in the notebook, you
just have to complete the blanks.
Please submit the completed file named as Assignment1a_.ipynb and
Assignment1b_.ipynb
No report is required.