networking - Python Backpropagation - How to Initialize the starting activation? -


i having troubles implementing backprop network. i'm not understanding how start off because in network first layer has 8 nodes. prompt gives me 10 in training set.

in first group example, have 

[0.0,0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0,0.0]

but 8 nodes can assign starting activation each node or have 2 left over.

####################################################################################### #  #                               preparations  #  #######################################################################################  import random import math import pygame import sys network1 = [] network2 = [] layer1 = [] layer2 = [] layer3 = [] set1 = [] set2 = [] set3 = [] mu = .5 eta = .1  ####################################################################################### #  #                                   node class #  #######################################################################################  class node(object):       def __init__(self,name=none):          self.name = name         self.error = none         self.activation_threshold = 0.0         self.net_input = 0.0         self.net_output = 0.0         self.outgoing_connections = []          self.incoming_connections = []         self.activation = none      def __str__(self):         return self.name      def addin(self,sender,weight=random.random):         self.incoming_connections.append(connection(sender,self,weight=random.random))     def addout(self,sender,weight=random.random):         self.outgoing_connections.append(connection(sender,self,weight=random.random))       def update_input(self):          self.net_input=0.0         conn in self.incoming_connections:             self.net_input += conn.weight * conn.sender.activation      def update_output(self):          self.net_output=0.0         conn in self.outgoing_connections:              self.net_output += conn.weight * self.activation      def update_activation(self):         self.activation = 1 / (1 + math.exp(-self.net_input))      def update_weight(self):         in self.incoming_connections:             i.weight = (2*i.reciever.activation - 1)*(2*i.sender.activation-1)         in self.outgoing_connections:             i.weight = (2*i.reciever.activation - 1)*(2*i.sender.activation-1)      def update_error(self):         pass  ####################################################################################### #  #                                   connection class #  #######################################################################################  class connection(object):      def __init__(self, sender, reciever, weight=random.random):          self.weight=weight          self.sender=sender          self.reciever=reciever      def __str__(self):         string = "connection " + str(self.sender) + " " + str(self.reciever) + ", weight = " + str(self.weight)         return string  ####################################################################################### #  #                                     creating nodes & connections  #  #######################################################################################  in xrange(8):      layer1.append(node(str(i))) network1.append(layer1) in xrange(3):     layer2.append(node(str(i))) network1.append(layer2) in xrange(8):      layer3.append(node(str(i))) network1.append(layer3)  in xrange(8):     j in xrange(3):         layer1[i].addin(layer2[j]) in xrange(3):     j in xrange(8):         layer2[i].addin(layer3[j])  ####################################################################################### # #                                         training patterns # #######################################################################################  """non-overlapping categories""" cata11=[0.0,0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0,0.0] set1.append(cata11) cata12=[0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,1.0] set1.append(cata12) catb11=[0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0,1.0,0.0] set1.append(catb11) catb12=[0.0,0.0,0.0,1.0,0.0,0.0,0.0,0.0,0.0,0.0] set1.append(catb12) catc11=[0.0,0.0,0.0,0.0,0.0,0.0,0.0,1.0,0.0,0.0] set1.append(catc11) catc12=[0.0,0.0,1.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0] set1.append(catc12) catd11=[0.0,1.0,0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0] set1.append(catd11) catd12=[1.0,0.0,0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0] set1.append(catd12)  """linearly independent instances , linearly separable categories""" cata21=[1.0,1.0,0.0,1.0,0.0,0.0,0.0,0.0,0.0,1.0] set2.append(cata21) cata22=[1.0,1.0,0.0,0.0,1.0,0.0,0.0,0.0,0.0,1.0] set2.append(cata22) catb21=[1.0,1.0,0.0,0.0,0.0,1.0,0.0,0.0,1.0,0.0] set2.append(catb21) catb22=[1.0,1.0,0.0,0.0,0.0,0.0,0.0,0.0,1.0,0.0] set2.append(catb22) catc21=[1.0,0.0,1.0,1.0,0.0,0.0,0.0,1.0,0.0,0.0] set2.append(catc21) catc22=[1.0,0.0,1.0,0.0,1.0,0.0,0.0,1.0,0.0,0.0] set2.append(catc22) catd21=[1.0,0.0,1.0,0.0,0.0,1.0,1.0,0.0,0.0,0.0] set2.append(catd21) catd22=[1.0,0.0,1.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0] set2.append(catd22)  """not linearly separable categories""" cata31=[1.0,1.0,1.0,1.0,1.0,0.0,0.0,0.0,0.0,0.0] set3.append(cata31) cata32=[0.0,0.0,0.0,0.0,0.0,1.0,1.0,1.0,1.0,1.0] set3.append(cata32) catb31=[1.0,1.0,0.0,1.0,1.0,0.0,0.0,1.0,0.0,0.0] set3.append(catb31) catb32=[0.0,0.0,1.0,0.0,0.0,1.0,1.0,0.0,1.0,1.0] set3.append(catb32)  ##set_activations(cata11) in network1:     i.update_weight() ##for thing in node1: ##    thing.update_weight() ##set_activations(cata12) ##for thing in node1: ##    thing.update_weight() ##set_activations(catb11) ##for thing in node1: ##    thing.update_weight() ##set_activations(catb12) ##for thing in node1: ##    thing.update_weight() ##set_activations(catc11) ##for thing in node1: ##    thing.update_weight() ##set_activations(catc12) ##for thing in node1: ##    thing.update_weight() ##set_activations(catd11) ##for thing in node1: ##    thing.update_weight() ##set_activations(catd12) ##for thing in node1: ##    thing.update_weight()  ####################################################################################### # #                                        updating network # #######################################################################################  in layer1:     i.update_input()     print 'node', str(i), 'input : ', i.net_input in layer1:     i.update_activation()     print 'act:', i.activation in layer1:     i.update_output()     print 'output', i.net_output in layer2:     i.update_input()     print 'node', str(i), 'input : ', i.net_input in layer2:     i.update_activation()     print 'act:', i.activation

neural network structure

i believe you've misunderstood how neural network structure build up. of input values should sent neurons in subsequent layer. of 10 input signals should sent 8 of neurons. please see appended graphics. i've included connections few of neurons make drawing easier understand.

graphical representation

enter image description here


Comments

Post a Comment

Popular posts from this blog

Why can rails not find a route created by a helper? -

i created member route in rails 4: resources :line_items post 'decrement', on: :member end and gave matching method in line_items controller: def decrement @cart = current_cart @line_item = @cart.line_items.find_by_id(params[:id]) @line_item.decrement_quantity respond_to |format| if @line_item.save format.html { redirect_to shop_path, notice: 'line item updated.' } format.js {@current_item = @line_item} format.json { head :ok } else format.html { render action: "edit" } format.js {@current_item = @line_item} format.json { render json: @line_item.errors, status: :unprocessable_entity } end end end but when try make button: <%= button_to 'x', decrement_line_item_path(item) %> i error: no route matches [post] "/carts/25" what gives? your error message is: no route matches [post] "/carts/25" but expectin...
Read more

javascript - jquery or ashx not working -

i using ajax file upload control. it creating html: <form action="/pages/handlers/asyncfileuploadhandler.ashx" method="post" enctype="multipart/form-data" target="ajaxuploader-iframe-1397144545"> <input type="hidden" name="type" value="mindvoice"> <input type="file" id="imagefile" style="display: block; width: 80%; margin-right: auto; margin-left: auto;"> </form> and target is: <iframe width="0" height="0" style="display:none;" name="ajaxuploader-iframe-1397144545" id="ajaxuploader-iframe-1397144545"></iframe> the form submitted using .submit() . my ashx handler: <%@ webhandler language="c#" class="asyncfileuploadhandler" %> using system; using system.web; using nettpals.core; public class asyncfileuploadhandler : ihttphandler { public void processre...
Read more

opencv - DataType<cv::detail::deriv_type>::depth what is it used for -

i wondering why line used in lucas kanade tracker in opencv: datatype<cv::detail::deriv_type>::depth can explain me? in opencv, depth of mat refers type of data contained in mat 's data buffer. represented integer values correspond given data type. these integers commonly abstracted appropriate macro definition (e.g. uchar data represented macro cv_8u ). cv::datatype type-traits class provides method obtain corresponding integer value without having memorize macro means data type. there few cases user code needs use datatype::depth . more common datatype::type . a simple example shows 1 possible use of datatype::depth : cv::mat uchar_data = cv::mat::ones(3, 3, cv_8uc1); cv::mat float_data; uchar_data.convertto(float_data, cv::datatype<float>::depth); // ^^ equivalently replaced // cv_32f macro float_data.at<float>(0,1) += 0.5f; std::cout << float_data << std::endl;...
Read more