generative-rnn

Inspiration

We all attended the same secondary school last year, and were very active in our school's competitive programming club. Also being interested in machine learning, we decided to combine the two together to make a machine learning model to solve competitive programming problems.

What it does

We originally envisioned an ML model that could take as input a problem statement and produce as output the solution to that problem. However, we ran into many challenges while building the model, and so decided to settle with a model that just produces code snippets from various examples.

How we built it

We wrote the model in Python, using Keras as the method for implementing TensorFlow to create recurrent neural networks (RNNs) and use long short-term memory (LSTMs) to train the model with examples in Python to produce Python code.

Challenges we ran into

There were many challenges that we ran into; this being our first attempt at an ML project. For example, working with multidimensional (5+) arrays was difficult, and learning what functions did what was also very time-consuming. In particular, we were unable to concatonate our two models; one for the input text and one for the output code, resulting in our decision to abandon our original goal.

Accomplishments that we're proud of

Our most significant accomplishment would be actually finishing (as in, writing code that compiles) an ML model. Since this was our first encounter with machine learning, TensorFlow, and Keras, and having nearly no time in advance to learn such concepts, we were very surprised that our model could produce something (although nowhere coherent).

What we learned

Machine learning is a very advanced field, and its topics cannot be learned over a weekend. However, this forced us to collaborate efficiently as we tried our best to manage our time while still learning about new ideas, participating in workshops, and networking with others.

What's next for generative-rnn

We plan to continue working on this project collaboratively, hoping to eventually achieve our original goal. Since ML is a topic that we are all interested in, and is a field in which some of us wish to conduct our own research in, this project serves as an interesting but difficult gateway into the field.