JaeDukSeo

random_direction1 = []
random_direction2 = []

for w in copy_of_the_weights:
    
    if w.dim() == 1: 
        random_direction1.append(torch.zeros_like(w))
        random_direction2.append(torch.zeros_like(w))
        
    else:

JaeDukSeo

random_direction1 = []
random_direction2 = []

for w in copy_of_the_weights:
    
    if w.dim() == 1: 
        random_direction1.append(torch.zeros_like(w))
        random_direction2.append(torch.zeros_like(w))
        
    else:

JaeDukSeo

random_direction1 = get_random_weights(copy_of_the_weights)
random_direction2 = get_random_weights(copy_of_the_weights)

for d1,d2,w in zip(random_direction1,random_direction2,copy_of_the_weights):
    
    if w.dim() == 1:
        d1.data = torch.zeros_like(w)
        d2.data = torch.zeros_like(w)
        
    elif w.shape[0] == 10:

JaeDukSeo

random_direction1 = get_random_weights(copy_of_the_weights)
random_direction2 = get_random_weights(copy_of_the_weights)

for d1,d2,w in zip(random_direction1,random_direction2,copy_of_the_weights):
    
    w_norm  = w.view((w.shape[0],-1))  .norm(dim=(1),keepdim=True)[:,:,None,None]
    d_norm1 = d1.view((d1.shape[0],-1)).norm(dim=(1),keepdim=True)[:,:,None,None]
    d_norm2 = d2.view((d2.shape[0],-1)).norm(dim=(1),keepdim=True)[:,:,None,None]
    
    d1.data = d1.cuda() * (w_norm/(d_norm1.cuda()+1e-10))

JaeDukSeo

#include <stdio.h>
#include <string.h>
/* the planets structure */
typedef struct 
{
	char name [10];
	double diameter;
	int moons;
	double orbit, rotation;
} planets_t;

JaeDukSeo

#include <stdio.h>
#include <string.h>

// this is the lab 7 homework problem tenfold (this is not the solution rather reference code)
// see here http://cps125.scs.ryerson.ca/labs/manual.html
void tenfold(int* array_pointer1,int* array_pointer2,int array_size){
  for (int i=0;i<array_size;i++){
    if(array_pointer1[i]<0){
      array_pointer2[i] = array_pointer1[i] + 4;
    }else{

JaeDukSeo

#include <stdio.h>

int check_prime(int a)
{
   int c;
   
   for ( c = 2 ; c <= a - 1 ; c++ )
   {
    if ( a%c == 0 ){
      return 0;

JaeDukSeo

class CNN():

    def __init__(self,k,inc,out, stddev=0.05,which_reg='A',act=tf_iden,d_act=d_tf_iden):
        self.w              = tf.Variable(tf.random_normal([k,k,inc,out],stddev=stddev,seed=2,dtype=tf.float32))
        self.m,self.v       = tf.Variable(tf.zeros_like(self.w)),tf.Variable(tf.zeros_like(self.w))
        self.act,self.d_act = act,d_act
        self.current_case   = which_reg
        
    def getw(self): return self.w
    def feedforward(self,input,stride=1,padding='SAME',training_phase=True,std_value=0.0005):

JaeDukSeo

class RELU_as_Reg():
    
    def __init__(self,batch,width,channel,regularizer):
        self.w = tf.Variable(tf.ones([batch,width,width,channel],tf.float32) )
        self.m,self.v       = tf.Variable(tf.zeros_like(self.w)),tf.Variable(tf.zeros_like(self.w))
        self.regularizer = regularizer
        self.lamda = 0.0001
    
    def feedforward(self,input):
        self.input  = input

JaeDukSeo

# m gradient base
m_pull_count   = np.zeros((num_ep,num_bandit))
m_estimation   = np.zeros((num_ep,num_bandit))
m_reward       = np.zeros((num_ep,num_iter))
m_optimal_pull = np.zeros((num_ep,num_iter))
m_regret_total = np.zeros((num_ep,num_iter))
                    
for eps in range(num_ep):
    temp_pull_count   = np.zeros(num_bandit)
    temp_estimation   = np.zeros(num_bandit) + 1/num_bandit