This is the eleventh week of my GSoC journey.

GSoC project

• GSoC project URL
• Work Repository

Tasks

1. After a simple bug fix, where the argument shuffle=True should have been set to shuffle=False for the test dataloader so that the prediction rows align with the indices specified in random_pos.csv, Figure 2C generated from 500 CellBox Pytorch models now resembles very closely the figure in the original paper, even capturing some outliers in the original scatterplot.
2. Meet Dr. Chris Sander and Augustin in person during trip to Boston.

Relevant issues:

• Blogs about Pytorch having suboptimal convergence when compared with TensorFlow model (this and this)

Notes

Check out my personal GSoC notebook, where I keep my observations and comments in one place.