https://github.com/sim98b/neural_network

This repo has various handmade neural networks
https://github.com/sim98b/neural_network

Last synced: 9 days ago
JSON representation

This repo has various handmade neural networks

• Host: GitHub
• URL: https://github.com/sim98b/neural_network
• Owner: Sim98B
• Created: 2023-07-26T09:26:42.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-08-18T16:21:02.000Z (over 1 year ago)
• Last Synced: 2023-08-18T17:45:34.556Z (over 1 year ago)
• Language: Jupyter Notebook
• Size: 2.46 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Neural_Network

This repo has various handmade neural networks

1. 1 Layer Classifier: a simple 1 layer neural network to classify random generatyed data

   Random data are generated with make_blobs() which parameters can be adjusted to get easier/harder problems; the function generating    data also return a difficulty index of the task, based standard deviatiation of blobs and number of classes, between 0 and 1           repsectively the easiest and the hardest problem.

   To get a significative index the value passed as the maximum of the standard deviation of clusters' centers should be half of the higher value passed to center box.

   

   The neural network has to minimize the cross entropy loss function and it returns the weights and the bias value to evaluate           model's    performance on data never seen, I.E. test set.

   

   The function provides numeric and graphic outcome about loss decreasing and a confusion matrix.

   

   Parameters: epochs = [**int**] number of learning epochs (defult = 200); lr = [**float**] learning_rate (default = 0.001); verbose = [**int**] verbosity, in epochs, for displaying loss decreasing (default = 10); bs = [**int**] batch size of the train set used to adjust weights and bias (default = 32); es = [**int**] number of epochs without a loss decrease beyond which learning is stopped (default = 30); dropout = [**float**] percentage of feature to drop and don't use during learning (default = 0.2); plot = [**0 or 1**] whether display train and validation loss decreasing and confusion matrix's plots (default = 1)