martinferianc / wineclassification-eie3

Classifying wines based on physicochemical properties done as a coursework for EE3-23

👋 Hi there! I'm Martin

Website • Google Scholar • LinkedIn

---

Martin obtained an MEng in Electronic and Information Engineering from Imperial College London, London, UK in 2015. He is currently a PhD candidate in the Department of Electronic and Electrical Engineering at University College London. His research interests include Bayesian neural networks, deep learning , hardware acceleration and confidence calibration. He has hands-on experience from industrial/academic placements in different countries.

wineclassification-eie3's Issues

Linear Regression

Fix testing at the moment it is potato

To-Do:

• Record training accuracy per epoch, per fold

• Record validation accuracy per epoch, per fold

• Record loss per epoch, per fold

• Combine validation accuracy, training accuracy per epoch per fold

• Combine it all in main.py also with postprocessing and confusion matrix generation for the final models

• Add comments

• Read the paper which was initially written about this

• Ask about what values does he actually want in the report

To-Do:

• Cross-check with others
• Correct the readme after the final chcek - conclusion
• Add ensemble learning to the conclusion
• Write that the performance measures largely depend on the number of examples in each class - we have an imbalanced dataset!

To-Do:

George

• Correct the regularization

• Reference the activation function

• Go through the manual to check if we have everything

Martin

• Make visualizations of the final model

• Write README to run our thing

• Go through the manual to check if we have everything

Report:

1. Introduction
   a. talk about the problem in hand and how am I planning on solving it
   b. talk about the data, how many features there are and how many classes and what are their number
   c. talk about splitting into n-folds to do cross validation 80%, 10%, 10%
   d. talk about normalisation and why is it significant
   e. talk about cross correlation - plot correlation matrix and mention anything significant

2. Linear method
   a. propose linear regression classifier with MSE loss and explain the overall OvA approach
   b. perform the different measurements and summarise the outcomes of :
   c. Learning rate, mini batch size, Number of epochs, different regulariser, different penalties
   i. Learning rate - vary from 0.1 to 0.0001 plot the average loss and display the average validation accuracy Batch size 256, 5 epochs
   ii. With the best learning rate vary the number of epochs 1 - 5 keep batch size 256 plot the loss function and again the validation accuracy
   iii. Regularizer try L1 0.1 to 0.00001 try L2 from 0.1 to 0.00001
   iv. Plot the training vs. validation accuracy
   v. include confusion matrix in the appendix

3. Neural Network

4. Conclusion
   a. add the MAD error as show in the original paper

To Do

change it into linearr regression