Programming Assignment - 4 Implementing Backpropagation

CS:5007 Deep Learning
Programming Assignment - 4
Implementing Backpropagation
Follow the instructions given below carefully:
1. You are allowed to use ONLY the following python libraries: NumPy, Pandas, Matplotlib, random, math, sys
2. You are free to take help from the backpropagation code given in ‘Demo 6 - Backpropagation’.
3. You must submit your code in a single python notebook with naming format as follows:
Firstname Lastname assignment4.ipynb
4. Your code must be properly commented explaining each step clearly.
5. Penalty will be there if any of the above instructions are not followed.
6. Zero marks will be provided for plagiarised code.
7. NOTE: This assignment is GRADED and the total points for this assignment is 5. An extra BONUS of 2
points will also be given to the students whose performance is in top 5.
Questions
1. (5 points) Implement your own fully connected feed-forward neural network with backpropagation from scratch
for the task of classification on the given dataset.
• Dataset
This dataset is about the confirmed COVID-19 cases in Ontario. The original data comes from the following file “Confirmed positive cases of COVID19 in Ontario” at: https://data.ontario.ca/dataset/
confirmed-positive-cases-of-covid-19-in-ontario. The original dataset is resampled and divided
into training and test data, namely, data train.csv and data test.csv respectively, for this assignment. The
target label is the column ‘Outcome1’ which is present only in data train.csv. The labels are strings and
three different labels are there as follows: Fatal, Resolved, Not Resolved. You should convert categorical
features to numerical or one-hot encoded feature, as appropriate.
• Task
Classify the confirmed COVID-19 cases among the three different classes: Fatal, Resolved, Not Resolved.
Setup and train your own fully connected feed-forward neural network model to perform this task.
• Evaluation Criteria
(a) (2 points) We will evaluate Mean F1-Score based on the predictions made by your model on
data test.csv. The F1 score, commonly used in information retrieval, measures accuracy using the
statistics precision p and recall r. Precision is the ratio of true positives (tp) to all predicted positives
(tp + f p). Recall is the ratio of true positives to all actual positives (tp + fn). The F1 score is given
by:
F1 = 2
p · r
p + r
where p =
tp
tp + f p, r =
tp
tp + fn
The F1 metric weights recall and precision equally, and a good retrieval algorithm will maximize both
precision and recall simultaneously. Thus, moderately good performance on both will be favoured over
1
Row ID Outcome1
1 Fatal
2 Resolved
3 Not Resolved
extremely good performance on one and poor performance on the other.
You must submit your predictions in a .csv file with the name of the file as ‘Predictions.csv’. The file
should contain two columns: Row ID and Outcome1. The file should contain a header and have the
following format:
(b) (3 points) Write a report including the following:
i. Give a detailed explanation of your approach which must include: Data preprocessing, model
architecture, loss function, optimizer used and relevant hyperparameters.
ii. Mention how you evaluated your model’s generalization performance. Plot the train accuracy and
validation accuracy per iteration curve and write your observations. NOTE: You must submit the
report in PDF format with the name of the file as ‘Report.pdf’.
2