siddeshpillai/deep-learning

Image Captioning java based library

deep-learning

An efficient architecture to perform deep learning of convolution neural networks on GPU using P2J library.

Work is still in progress. Keep an eye on the project for updates.

Contributors

1. Suhas Pillai
2. Siddesh Pillai

Sequential code: Download files and CIFAR data set. Set path and train the system

deep-learning's Issues

Update the papers

dude add some papers.

Logic for numerical gradientcheck

Hi Siddesh,

Refer the gradient check code. Also, check my code on data_load ...try to use object of numjava class for matrix computation rather than directly using an array....because I have written most of the code and matrix functions using numjava object----> which has a matrix.

so the call to the matrix is
numjava n= new ....
n.finalmatrix------> which will give you the matrix for computation.

So, why we need this is because we want to verify the whether our back propagation is correct or not.

In Short, what we calculate using partial derivatives is analytical gradient and numerical gradient (the code which you will write ) helps to verify whether our analytical gradient calculation is correct or not.

eval_numerical_gradient()
eval_numerical_gradient_array()
write only for these two functions.

Regards,
Suhas.Pillai

Test

• Convol - currently in progress
• Layers
• NumJava
• DataLoad
• GradiantCheck

Mathods to implement

Methods to implement
def svm_loss(x, y):
def softmax_loss(x, y):
def eval_numerical_gradient(f, x, verbose=True, h=0.00001):
def eval_numerical_gradient_array(f, x, df, h=1e-5):
class ClassifierTrainer(object): ---> big one for entire training.

One of the too few implementation in java

Thank you for that.
I wish to improve a trading strategy, can i use your code easyly ?
(tryed the karpathy js before)
And made an android app wich need more improvment, than my particule swarn or other i've tryed