Progress Highlights
Project Research / Discussion
Received FYP monitoring feedbacks from panels (overall satisfactory!)
Created a Google Doc to compile the FYP monitoring assessment results, comments and suggestions for improvement from panels and provide our own analysis, review and proposed actions
Created and disseminated a Google Form survey to collect smart watch user requirements following feedback from FYP monitoring
Project Development
Successfully completed the labelling phase for all data sets (10,149 tweets in total)
Analysed the labelled data set to obtain weightage of emotions labelled
Attended a meeting with supervisor to discuss regarding FYP monitoring feedback and current progress update post mid semester break
Successfully developed the Word2Seq feature extraction algorithm using Keras and Tensorflow (achieved 78% accuracy but with terrible overfitting issues due to imbalance data as shown in the learning curve figures)
Officially created a new Django project for the Smart Watch User and Administrator dashboard web-app with Faidz
What I'm stuck at?
Currently stuck in developing the Word2Vec feature extraction algorithm (pending help from Ms. Chempaka Seri)
Trying to find ways to improve accuracy of the model by adding more preprocessing method such as translating Malay tweets to English beforehand (but Google Translate API package for Python is broken at the moment)
Finish the
How will next week be?
Complete the features extraction using Word2Vec
Assist Faidz in developing the Smart Watch User and Administrator dashboard web-app in 1 week's time (wish us luck!)
Proceed with 2nd phase of FYP research (developing the project activity diagram)
Start curating the FYP1 report in IEEE single-column paper format
Amend the viva presentation slide contents based on the FYP monitoring feedback
Lessons Learnt
TensorFlow and Keras is a pain in the ass to use at first. As it turns out, there's a lot of discrepancies with v1.x.x and v2.x.x versions of TensorFlow. Hence, I cannot simply directly refer to online examples as most of them used the outdated version of these dependencies. Thankfully, with a lot of Googling and referring to official documentations of TensorFlow and Keras, I was able to resolve the resolving dependencies import errors.
The notion that machine learning is easy AF because it merely "uses existing packages developed by someone else" couldn't be further from the truth. After doing a bit of machine learning coding, I can actually appreciate its sheer complexity and have a lot of respect towards my AI peers.
A lot of my peers including myself opted to using agile approach specifically Kanban flow for our FYP. This ended up becoming an issue among the panels as they are suspecting that us students are merely using Kanban flow to escape from the tedious process from other Agile flows such as Scrum (e.g. need to have daily standup meetings, product backlogs etc.). The lesson to be learnt here is that for every single decision made in FYP (but also pretty much in life, really), be prepared to come up with strong justifications. Don't be a sheep that follows things blindly.