Data Scientist
Data science allows one to create innovate complex systems with mere zeros and ones.
Creating such systems from the ground up to advance and improve the economic, academic state and digital security of society
Framework Techniques
Frameworks that I have practiced ranging from machine learning, data science and web develepemont. Differing between several programming languages and disciplines.
Over a span of 2 years and the SSI data science/statistics and research bootcamp I have practiced many methods and techniques in processing, calculating and interpretung data. Processing text; using NLP methods (i.e, calculating document similarity with cosine similarity and tfidf vectorization and sentiment analysis); Images classification (using convolutional neural networks); and TimeSeries analysis for stock regression (using Long-short Term neural networks).
Web development was an important step for me to be able to fully make use of my software. I realized that the applications I would create would ultimately be obsolete if I can't create an interface that allows it to best used. I can't exactly make a stock regression application and return the user a mean squared error score.
There are many machine learning algorithms and method variatians I have explored. Ranging from linear regression to deep learning. I have learned how to process different types of data and use them in machine learning problems. Including tabular data, image data, text and datetime data. Using the Tensorflow library varying between the Sequential, Function and Subclass API.
During my time doing the Summer STEM Institute course, the statistics lecturer would at the end of every lecture give us a moral for us to take away. One of them was that data always tells a story and they say that a picture is worth a thousand words. Visualizations allow one to convey information and send a message that is understandable and attactive to the reciever.
My Projects
RevisionBank
Made Using:
RevisionBank is a web application that provides/collects topic specific questions for A level subjects. Emailing the selected question papers to refer back to. Including past papers, solution banks and markschemes. The RevisionBank Tools cover: Maths, Further Maths and Physics with plans to expand into further subjects, exam boards and key stages. The initial goal was to tackle the tsunami of Further Maths A level homework, in an ambtious attempt at automating my A levels. The plan for future features is to implement a search engine that will collect the papers and decide if the papers collected from that site fall legally under copyright laws. Then allow a more interactive experience with the user, with revision scheduling software, note taking and more.
STEMRoadmap Startup Project
Made Using:
STEMRoadMap is my first startup project attempt. With the goal to collect information from the internet and create a learning roadmap for any STEM Subject. Niching to students, undergraduates or even someone interested in a career change. The software implements a host of many machine learning and natural language processing methods. I managed to make the software collect the the exact data required, reducing the data collection and filter process from 7 minutes to 40 seconds each query. This long runtime in the wild would prove impractical so I implemented a method of precaching the data on a database. However at the time I didn't know how to deploy python scripts to be run as a long running script. After practicing with a new project RevisionBank I was able to learn how to run long running scripts on heroku. Next I plan to run my scripts using AWS. Previously I was deterred from AWS due to the very small budget that I had so heroku was a good free alternative. However heroku still has some disadvantages. The project took 9 months to implement the main bulk of the functionality, this was due to the slow process of self teaching myself all the tools I required.
AS Level Revision Card Scheduler
Made Using:
AS Level Revision Card Scheduler is an application that takes a random revision card you make and sends one to you every hour. Allowing me to revise with whilst doing other things. This project started with an excrutiating 4 hour cram revision of AS Level Physics a few weeks before my exams. No matter how much I stared at my textbook I still cou;dn't remember the elementary charge of an electron. A defeated google search of 'how to revise AS level physics' led me with the two answers; making revision cards and exam questions. The tedious and boring nature of making paper revision cards repelled me and got me falling asleep thinking about it. This led to the verdict that I would create revision cards but online, this would allow me to store them in a safe place and schedule the provision of these revision cards to myself every hour and have easy access.
NLBash Attention Transformer Model
Made Using:
NLBash started with my dream of making my own smart house with all the software in there I created. Initially this started with the urge to create an A.I assistant. I wanted to provide funxtionality to the A.I assistant that would make it different to things like Alexa and Siri. Then I was inspired by J.A.R.V.I.S from Iron man. J.A.R.V.I.S controls every aspect of Tony starks house and thinking on how that would be implemented in a real-life standpoint. J.A.R.V.I.S would be translating what Tony Stark says and be formulated running scripts. I could create predefined scripts that can be used by the A.I assistant but where's the fun in that. At the moment NLBash is will simply take natural language and translate it into built-in bash commands. However, further development would lead to usecase specific arguments and non built-in commands.
Stock Forecasting/Regression using Machine Learning/Deep Learning LSTM's and Sentiment Analysis
Made Using:
This project was made for my initial A Level Extended Project Qualification idea. The application was a stock predictor with the goal of finding correlation using stock historical data and finding causation by using sentiment analysis on Twitter posts. Then using both models to make a final prediction. I managed to make forecast predictions with LSTM's with considerable accuracy. However the sentiment model was having problems. The sentiment model used an embedding layer,several LSTM Layers and an Attention Layer. Due to little knowledge in sentiment analysis at the time. I had a problem that each time I try to predict after training, I would come across new voabulary that the model had never seen before so I would have to randomly initalize it which wasn't ideal in accuracy. After reading an article on Natural language to sql, it made me realize I should use a pretrained glove model to handle the general language then train my own embeddings to allow for analysis specfic to the twitter posts and headlines.
Digit Recognizer using Tensorflow CNN's
Made Using:
The Hello World of Image classification. This project was my first attempt at learning deep learning. With the goal of learning how to analyse all possible datatypes for a more all-rounded skillset. Ranging from tabular data; to images; then text and finally audio. Realizing that a model is completely useless if it can't be used in a convienent and accessible, I to learned how to deploy a model using docker. I learned the basics of creating a docker image and running docker containers in order allow for deployment, but in this case only locally. Seeing that I had not yet come across a platform to deploy my projects to the cloud. Later I come across Heroku and Azure which allowed me to deploy my RESTAPIS.
My Skills
