We were inspired as a group by our common interest in machine learning and our desire of finding more user friendly, practical, and educational applications that would benefit the largest amount of users. English and translations is a popular topic in AI and we came up with an application that finds grammatical errors in conversation given by a user who is practicing how to write and speak english. Our neural network returns their conversational input with highlighted grammatical mistakes.
What did we do? We created a secure, single page website that uses Django to create a front end interface that can handle machine learning python scripts. With a submit button, we give our neural network input given by a user for interpretation and display feedback to them.
A few of the issues we faced along the way was connecting a database from our server to our environment. Initially we were most comfortable and aiming to use a MySQL database but had to shift to the preferred Django sqlite to be able to move forward with our project. This proved to be a more practical solution due to the amount of documentation provided by Django in the use of a sqlite database. We also found a challenge when training our neural network with english data sets. Tensorflow documentation on the subject was limited for two recursive neural networks and our available GPU resources were lacking the power for faster training and computation of the algorithm.
I am proud of our continued progress and compatability as a group. Our hourly goals were continously reached throughout the night. We always found another way to accomplish a task anytime that an obstacle was presented. The front end was succesfully implemented and worked as planned, ready for a beta launch. The tensorflow backend made tremendous progress throughout the night even after many setbacks and redesigning. Our Encoding/Decoding was a unique solution to our problem and I want to think Evan for his continued dedication on optimizing the process. In the end we encounted that training our nueral net was limited to the hardware we had available.
Today, we learned how to implement a Django frontend and the ability to run python scripts. We also came to realize that the Keras framework is easier to work with than the AI tech we chose this time because it uses Tensorflow as a backend but is easier to implement. As a group, we all gained the experience of putting together a successful front end with a backend that uses Python as the main language.
And training our NN on a more powerful computer to increase accuracy and power. We would also like to implement more languages for more users to be able to practice with!