r - Does the number of hidden nodes in the fully connected layer has to be equal to the number of output categories? -


i tried tutorial image recognition in r using mxnet package https://www.r-bloggers.com/image-recognition-tutorial-in-r-using-deep-convolutional-neural-networks-mxnet-package . aim of tutorial recognize faces of 40 persons. dataframe consists of 400 pictures (10 pictures per person). cnn looks this:

data <- mx.symbol.variable('data') # 1st convolutional layer conv_1 <- mx.symbol.convolution(data = data, kernel = c(5, 5), num_filter = 20) tanh_1 <- mx.symbol.activation(data = conv_1, act_type = "tanh") pool_1 <- mx.symbol.pooling(data = tanh_1, pool_type = "max", kernel =    c(2, 2), stride = c(2, 2))  # 2nd convolutional layer conv_2 <- mx.symbol.convolution(data = pool_1, kernel = c(5, 5), num_filter = 50) tanh_2 <- mx.symbol.activation(data = conv_2, act_type = "tanh") pool_2 <- mx.symbol.pooling(data=tanh_2, pool_type = "max", kernel = c(2, 2), stride = c(2, 2))  # 1st connected layer flatten <- mx.symbol.flatten(data = pool_2) fc_1 <- mx.symbol.fullyconnected(data = flatten, num_hidden = 500) tanh_3 <- mx.symbol.activation(data = fc_1, act_type = "tanh")  # 2nd connected layer fc_2 <- mx.symbol.fullyconnected(data = tanh_3, num_hidden = 40) # output. softmax output since we'd probabilities. nn_model <- mx.symbol.softmaxoutput(data = fc_2)

i used same neuronal network own dataset, consists of 1600 pictures of 5 persons. adjusted number of nodes in connected layer 5.

fc_2 <- mx.symbol.fullyconnected(data = tanh_3, num_hidden = 5)

the results of model bad, set ceteris paribus number of nodes in connected layer 80 , got great results (accurancy: 100%).

fc_2 <- mx.symbol.fullyconnected(data = tanh_3, num_hidden = 80)

the model generates probabilities 80 categories although got 5, accurancy excellent. don’t understand event. tried add third connected layer right number of catgories:

# 2nd connected layer fc_2 <- mx.symbol.fullyconnected(data = tanh_3, num_hidden = 80) tanh_4 <- mx.symbol.activation(data = fc_2, act_type = "tanh") # 3rd connected layer fc_3 <- mx.symbol.fullyconnected(data = tanh_4, num_hidden = 5) # output. softmax output since we'd probabilities. nn_model <- mx.symbol.softmaxoutput(data = fc_3)

but results bad. thought number of nodes in connected layer represents number of output categories model should try distinguish.

  1. is possible explain event?
  2. does number of hidden nodes in connected layer has equal number of output categories?

thanks help.

you have more parameters in model samples. typically bad, , can cause on fitting.

another approach can take taking pre-trained model, , re-train last layer data (aka transfer learning). here's mxnet tutorial that: https://mxnet.incubator.apache.org/how_to/finetune.html


Comments

Post a Comment

Popular posts from this blog

Sort a complex associative array in PHP -

this question has answer here: php sort array subarray value 5 answers i have associative array in php like: $newarray1 = array ( [producta] => array ( [link] => http://sanjosespartan.com/blog/products/agile-java-development/ [visitcount] => 4 ) [productd] => array ( [link] => http://sanjosespartan.com/blog/products/intro-to-go/ [visitcount] => 1 ) [productg] => array ( [link] => http://sanjosespartan.com/blog/products/node-js-for-developers/ [visitcount] => 1 ) [productb] => array ( [link] => http://sanjosespartan.com/blog/products/beginning-mysql/ [visitcount] => 1 ) [productc] => array ( [link] => http://sanjosespartan.com/blog/products/computer-networks/ [visitcount] => 1 ) ...
Read more

vb.net - How to ignore if a cell is empty nothing -

how use roscolor() ignore value (like empty cell) ? color row if value upper 5 when there nothing in cell, want ignore roscolor() apply, how? public sub roscolor() integer = 0 quotedatagridview1.rows.count() - 1 step +1 dim val integer val = quotedatagridview1.rows(i).cells(3).value if val = vbempty quotedatagridview1.rows(i).defaultcellstyle.backcolor = color.white elseif val < 5 quotedatagridview1.rows(i).defaultcellstyle.backcolor = color.red elseif val > 5 , val < 10 quotedatagridview1.rows(i).defaultcellstyle.backcolor = color.lightyellow end if next end sub you can check empty value following public sub roscolor() integer = 0 quotedatagridview1.rows.count() - 1 step +1 dim val = quotedatagridview1.rows(i).cells(3).value if isdbnull(val) or val nothing quotedatagridview1.rows(i).defaultcellstyle.backcolor = color.white ...
Read more

recursion - Can every recursive algorithm be improved with dynamic programming? -

i first year undergraduate csc student looking competitive programming. recursion involves defining , solving sub problems. understand, top down dynamic programming (dp) involves memoizing solutions sub problems reduce time complexity of algorithm. can top down dp used improve efficiency of every recursive algorithm overlapping sub problems? dp fail work , how can identify this? the short answer is: yes. however, there constraints. obvious 1 recursive calls must overlap. i.e. during execution of algorithm, recursive function must called multiple times same parameters. lets truncate recursion tree memoization. can use memoization reduce number of calls. however, reduction of calls comes price. need store results somewhere. next obvious constraint need have enough memory. comes not-so obvious constraint. memory access requires time. first need find result stored , maybe copy location. in cases, might faster let recursion calculate result instead of loading somewher...
Read more