For the past week, we have been focusing on data augmentation. I remember we started off researching for frequently adopted data augmentation methods. Sadly, most were only applicable to images. For instance, cropping of images was one data augmentation method suggested. Yet, given that our data were a bunch of numbers, this was irrelevant. We spent a whole day researching on it. However, we were as clueless as ever. It was truly disappointing. Just then, enlightenment came. We realize we could apply the idea of cropping – taking segments of the data – to our numerical data. Yes, there was finally progress! Just like that, ideas started flowing and what’s left was only coding.
After a few days of coding, we were almost done. I was overjoyed. Finally our job was complete and I was hoping we could see an increase in our model’s accuracy. But i was wrong. Another issue… Python could only return a list with X number of elements. So parts of our processed data was lost. This sounds simple. All we needed to do was to write a few data points in at a time. Yet, having to skim through the long lines of code and finally realise that it was due to python’s inbuilt issues was time consuming and draining. Finally the issue was solved. We were ready to present our work to our TA for feedback and end our work for the day.
I had never been so anxious and frustrated. Our TA told us to try out another completely different algorithm as our model. Having to complete it by our theme meeting the next day was not the worst since I had written the code for the algorithm previously. This means all that was left was to alter our data such that it could be fitted into the model. The worst was our time tight schedule needed to be pushed back again. We were planning to combine our EEG based classification model and image-based classification model to produce our final classification model this week. Yet having to change our model meant that we had no fixed model now and combing model was totally out of reach. Long story short, the new model was lacking as compared to the initial one and we reverted to the initial model.
Theme Meeting! As usual, we had our theme meeting at 蒙民伟楼. Except this time, we were no longer reporting on our progress and brainstorming on our next step. Instead, our Professor requested that we shift our focus from “prototyping” to preparing for our upcoming presentation. Thankfully, we had two teammates who took the initiative to start preparing for our presentation and were thus able to clearly distribute work among us. In my case, I was in charge of writing report. So, that was how I spent the next few days at yueya.
Well, that’s about it for the week. Bye!