I want to build an ANN from scratch without using frameworks. However, when training the model, the cost function almost does not decrease at all. I’ve reviewed it many times but can’t find anything wrong. Please help me review the code shared on the link below. Thanks so much!
https://drive.google.com/drive/folders/1JADTaaDNUyPipz3f3Xlg8iJqEMEEI--r?usp=sharing
Which DLAI courses have you attended? Many of the courses start with a from-scratch project notebook.
I’d recommend going through Deep learning Specialization especially the first course on Neural Networks.
Go to Course Q&A → Deep Learning Specialization → Neural Networks
Yes, of course I have studied Professor Andrew’s Deep Learning course. However, I want to re-implement a feed forward network from scratch using softmax regression for image classification in the MNIST dataset. In fact, my code is partially modified from assignment in Spec Deep Learning.
I think most of what you need is in Course 1 of DLS.
But you might have to do some outside research to get the backpropagation code for neural network training.