Not able to reduce loss

#1

Somewhere I am missing some update rule. It would be of great help if you can please take a look into the code at https://ide.codingblocks.com/s/250998

#3

hey @tisandas2011 ,

Yeah you were missing an update rule at each iteration , dues to some error in weightedMat that you used and after that the way you have updated your weights and bias.

I have corrected them and implemeted it again in iterative manner , to understand it much more easily. Have a look at this code https://ide.codingblocks.com/s/251195 , i guess you can understand it easily.

If still there is something ,which you can’t understand then kindly ask me.
Thank You and HappY Learning. :slightly_smiling_face:

#4

Thanks for the help. I have made the changes as pointed out by you. However for the test scenario I am still getting around 30% accuracy. It would be of great help if you can confirm whether still some problem exists or it would be due to the hyperparameters values. The modified code can be checked at https://ide.codingblocks.com/s/251903

#5

hey @tisandas2011 ,
Your implementation is correct and is really good. I really appreciate it.
Actually the thing with SVM is , that you can achieve maximum of 35% with SVM on this dataset. I tried various svm models on this dataset but couldn’t improve it much.
So if even you try to tune it , then you can’t improve it much.

So i would suggest you , as this dataset is mainly about images , then neural networks will work more better on this and can also provide you with good results.

I hope this would have resolved your doubt.
Thank you and Happy Learning. :slightly_smiling_face:.

#6

Thanks for the prompt help. :slight_smile: