A neural network architecture built from scratch doesn't seem to work!

Using all the knowledge gained from the first course “Neural Networks and Deep Learning”, I tried to construct a neural network architecture from scratch in Google Colab. But it didn’t work at all. I tried to use the same cats classifier dataset to train the model. But the cost remains unchanged after every iteration. What’s more surprising is that, even after copying the entire code in Google Colab and Kaggle from the lab, and using the same dataset, the code still performs horribly. But as it turns out, in the lab, it performs fine. Why so? Can the same code perform differently in different platforms?

Normally speaking, not unless there are libraries and functions used specifically in one environment and not used in the other.

Instead of copying all your code to Colab, why not upload all the files to Colab and replicate the assignment? Check this guide on how to do this. You will get the same results as in the assignment.