Others

IMBD Movie Sentiment Classifier

This is a binary sentiment classifier for the IMDB movie dataset. IMDB dataset has 50K samples of reviews with labels of either positive or negative sentiments. The dataset is pre-processed and trained with traditional machine learning and neural net deep learning algorithms, and results are analysed and compared with each other. All three traditional machine learning algorithms of Naïve Bayes, Logistics Regression and SVM perform well with Accuracy, Precision and Recall at around 0.9. For deep learning, a CNN architecture is used with about 500,000 trainable parameters. Best results reached at 2 epochs with Accuracy, Precision and Recall at 0.86. Neural networks should work better for much larger dataset. The code is written in Python and executed on Jupyter Notebook.

Python code & slides are available at:

CIFAR-1o Image Classifier

This is an image classifier for the well-known CIFAR-10 dataset. The dataset contains 50,000 train data and 10,000 test data of 32x32 RGB colour images of 10 categories of objects. This project compares the performance and efficiency of several deep learning architectures of Multi-Layer Perceptron, LeNet CNN and 3 block VGG. The best model is the Optimised VGG-3 block with 85% accuracy & quick convergence while LeNet offers the best efficiency (in terms of accuracy per unit of training time) and achieved 69% accuracy in just about one and half minutes. The code is written in Python and executed either on a Jupyter Notebook or Google Colab.

Python code & slides are available at:

Jackie Hair Salon (celebrities edition)

Jackie Hair Salon Special Celebrities edition is an education project based on a fictitious hair salon to illustrate how Dialogflow is used to automate the functions of a hair salon receptionist. It incorporates Google Calendar, Gmail and Firestore cloud database. Jackie Hair Salon chatbot is deployed to Google Assistant for voice interface, and Telegram. It is also available on the website (text interface). Special Celebrities edition uses celebrities as the customer names. It is a fun way to learn chatbot development on Dialogflow. Try it out on the Chat Now page, for the web, telegram or Google Assistant versions. Have fun!

All Dialogflow and Javascript code and configuration instructions are available at:

tk restaurant guide website

open university DATA analysis: tableau

Open University Learning Analytics Dataset contains data about courses, students and their demographics, grades etc for seven modules of about 30,000 enrolled students. Using Tableau’s data visualisation, we spot trends such as student demographics, gender preferences for certain Science-Technology-Engineering-Mathematics (STEM) and social science courses, student behaviour approaching assignment submission deadlines, grades and others. Once developed, Tableau visualisation dashboard allows interactive engagement with the data to explore and discover the trends. Some trends are obvious like urban dwellers prefer Open University online courses as they can study while still working. Other trends are counter-intuitive like STEM courses which are gender neutral. Yet other trends are more striking when visualised like the great number of students submitting assignments at the deadline itself. Based on these findings, three broad recommendations are made on targeted recruitment based on student demographics, module transformation to avoid a mad rush of students handing in assignments at the deadline, and empowerment of the weaker students.

Tableau code and slides are available at: