hogneuralnet's Introduction

HOGNeuralNet

Creation of a neural network to classify HOG data gathered from the Jaffe database.

This nerual network has a single hidden layer which can use a number of activation functions. The output layer consists of 7 neruons using a sigmoid function for an activation. The final output uses the one-hot method, with the maximum value in the set becoming the 1 value, and the rest being reduced to 0.

Emotion Labels

1000000 Angry
0100000 Disgust
0010000 Fear
0001000 Happy
0000100 Neutral
0000010 Sad
0000001 Surprise

Program Output

Program Starting

Read in 213 images from the Jaffe database.

	 30 	 Angry
	 29 	 Disgust
	 32 	 Fear
	 31 	 Happy
	 30 	 Neutral
	 31 	 Sad
	 30 	 Surprise
	---------------------
	 159 	 Training Examples
	 54 	 Testing Examples

Running HOG
Training Neural Network
	---
Cost at 0 4.851773795072829
Accuracy: 13.20754716981132
Cost at 50 2.770205801525372
Accuracy: 30.81761006289308
Cost at 100 2.018388789910542
Accuracy: 66.0377358490566
Cost at 150 1.7183135887095513
Accuracy: 63.52201257861635
Cost at 200 0.5222544032112558
Accuracy: 96.22641509433963
Cost at 250 0.1749239751147752
Accuracy: 99.37106918238993
Cost at 300 0.07886075520084389
Accuracy: 100.0
Cost at 350 0.05115073052317008
Accuracy: 100.0
Cost at 400 0.03672703543742705
Accuracy: 100.0
Cost at 450 0.028140808053845397
Accuracy: 100.0
Cost at 500 0.02254917335028536
Accuracy: 100.0
Cost at 550 0.018662424354947645
Accuracy: 100.0
Cost at 600 0.01582855457931452
Accuracy: 100.0
Cost at 650 0.013683022699696836
Accuracy: 100.0
Cost at 700 0.012009713185312425
Accuracy: 100.0
Cost at 750 0.010673442543944927
Accuracy: 100.0
Cost at 800 0.009583930287963505
Accuracy: 100.0
Cost at 850 0.008680526551903303
Accuracy: 100.0
	---
Making Predictions

Accuracy: 90.74 %


Times:
	 0.10809 	Read In Time
	 19.64181 	HOG Transformation Time
	 182.11333 	Training Time
	 0.05475 	Training Time
	---------------------
	 201.91798 	Total Time

Recommend Projects