Google kaggle competition on TPU performing metrics.
Link to the competition.
For this project, we participated in a Kaggle competition that was organized by Google. The goal of the competition was to predict the time that a model would run for based on the configurations and parameters used.
The repo is organized in the following manner:
- Results: contains the csv files output by the model, which are required for the submission.
- Code: contains the notebooks and Python files required to train and test the model.
- Plots: contains plots that were used for the presentation.
- Summary: contains the draft and final version of the write-up.
A big part of the project was understanding the provided notebooks. We took Google's intro files, along with files provided by other participants, and created a spliced version them to form our training/testing notebook.