Description
1. Perform the following implementation and analysis of binary SVM classification on the
provided data 40 marks
a. Randomly pick 80% of the data as a training set and the rest as a test set.
Implement the binary SVM classifier using the following kernels: 20 marks
i. Linear
ii. Quadratic
iii. Radial basis function
b. Report both the training and test set classification accuracies for the most suitable
value of generalization constant C for each of the three kernels. Consider C
empirically. Report the accuracy in the form of a comparison table with
corresponding C values (C value which provides the best test set accuracy is the
most suitable one out of all those different trial values that you try.) 20 marks
2. Build a MLP classifier for the given dataset. Use stochastic gradient descent optimiser for
the models. Find the number of nodes in the input and output layer according to the
dataset and justify it in the report. Randomly pick 80% of the data as a training set and
the rest as a test set. Use packages of pytorch. 55 marks
a. Vary the number of hidden layers and number of nodes in each hidden layer as
follows. 20 marks
i. 0 hidden layer
ii. 1 hidden layer with 2 nodes
iii. 1 hidden layer with 6 nodes
iv. 2 hidden layers with 2 and 3 nodes respectively
v. 2 hidden layers with 3 and 2 nodes respectively
b. For each of the architectures, vary the learning rates as 0.1, 0.01, 0.001, 0.0001,
0.00001. Plot graph for the results with respect to accuracy. (Learning rate vs
accuracy for each model and model vs accuracy for each learning rate.) 20 marks
c. Mention the architecture and hyper parameters (including optimizer and learning
rate) of the best found model in the report. Try to justify it. 15 marks
d. (Optional) Compare execution time in CPU and GPU for each model. Conclude
your findings.
3. Compare the performances of both the classifiers. 5 marks
Submission instructions:
1. Submit your codes by implementing them only in PYTHON. No other programming
language is allowed. You may use inbuilt library functions to perform the tasks.
2. Submit a README file which will contain the instructions on how to execute your code.
3. Submit a report briefly explaining the procedure and the results.
If required, any data cleaning or pre-processing can be done. Mention them in the report. The
source code, README file, and the report must be uploaded as a single compressed file (.tar.gz
or .zip). The compressed file should be named as: {Group_Number}_ML_A3.zip or
{Group_NUMBER}_ML_A3.tar.gz.
Datasets:
The links for the available datasets are provided below. The descriptions related to these datasets
are provided in the corresponding pages.
1. https://archive.ics.uci.edu/ml/datasets/QSAR+biodegradation