Regarding Implementation of Neural Network

amankharb · June 22, 2020, 8:58am

well i was implementing neural network algorithm by the help of videos and i notice that my accuracy is not up to the mark .So can u take a look at my code as sometimes when i change the number of neurons in hidden layer it perform worst (worst than Logistic regression (sometimes))
.As i have written vectorized implementation of delta2 as delta2=np.dot(delta3,w3.T)((a2)(1-a2)) instead of delta2 =np.dot(delta3,w3.T)(1-np.square(a2)) that is used in videos by prateek bhaiya.How a2(1-a2) is equivalent to (1-np.square(a2)).

github.com

Ornate123/Implemenataion_Neural_Network/blob/master/Artificial_Neural_Network.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Implementation of a 3 Layer Neural Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np "
   ]
  },
  {
   "cell_type": "code",

This file has been truncated. show original

prashant_ml · June 22, 2020, 3:46pm

hey @amankharb ,
there is no such problem in your model code. But you need to understand overfitting and underfitting in neural networks. i guess your model is getting over fitted at around epoch 685.
So just first search about them and understand them how to resolve them. If still there is something you can’t understand in them , i will surely help you out.

amankharb · June 23, 2020, 8:02am

i also want to ask what difference will it create when i use delta2=np.dot(delta3,w3.T)((a2)(1-a2)) instead of delta2 =np.dot(delta3,w3.T)*(1-np.square(a2)). ??

prashant_ml · June 23, 2020, 12:39pm

Its just the way you update your weight values and nothing else. You can also create any other formulae to perform updation of those values.
There is no such particular difference between them , other then as stated above.

I hope this would have helped you.
Thank You.

prashant_ml · January 6, 2021, 6:55pm

I hope I’ve cleared your doubt. I ask you to please rate your experience here
Your feedback is very important. It helps us improve our platform and hence provide you
the learning experience you deserve.

On the off chance, you still have some questions or not find the answers satisfactory, you may reopen
the doubt.