This week I continued tuning hyperparameters on the support vector machine to get better F1 score. F1 score is the main performance metric that I decided to use because the nature of our data is such that there are a lot more negatives than positives so even a bad model can give a high accuracy. Hence, precision is very important. Therefore, F1 score is a really good metric to use.
It takes a long time to train and test so I spent the whole week on just experimenting with different values for the hyperparamenters of the svm.
It was slightly frustating to see how long things were taking and I was running my code on google collab so sometimes if I left it to compile and/or train overnight, I would come back the next morning to “runtime disconnect” and I wouldn’t be able to test and get results.