Inspiration
We were inspired by a research paper we found created by Harvard and MIT professors on the use of TimeX and it is helpful for time-series models.
What it does
It is supposed to show the most influential attributes using TimeX of the ECG dataset to determine whether there are anomalies within the ECG.
How we built it
We first used Pytorch to create a base model. We trained 60% of the data given and then tested the 40% of the data with the model. We then replicated this but instead with the TimeX model to see the differences and spot the most influential attributes.
Challenges we ran into
We had many issues with underfitting and the autoencoder collapsing, which led to the TimeX model "working" but not efficiently.
Accomplishments that we're proud of
We are proud of being able to use the TimeX model and figuring out how to use the TimeX model with another dataset to see if we can get positive results. We were able to get the TimeX model to work and get data visualizations to prove it.
What we learned
We learned about TimeX and how there are different models online that can be used to create stronger models and interpret data in different ways.
What's next for TimeX Comparison and Review
This project has a lot of flaws that can be improved on. There are still underfitting issues and the autoencoder is still weak. With improvements, this project can more accurately give data and tell the most influential attributes to the dataset.
Log in or sign up for Devpost to join the conversation.