preview

Analysis Of Convolutional Neural Network Using Keras

Decent Essays
Open Document

specify the algorithm performance on the new data. With this model, a test accuracy of 97.65% was achieved after 10 epochs. The baseline error for this model came out to be 1.91%. The below figure 3.1 illustrates the loss and accuracy for every epoch succeeded by the last assessment of the trained model. Figure 3.1: Multi Layer Perceptron Neural Network epochs 3.2 Convolutional Neural Network Using Keras As mentioned in the previous chapter, the Convolutional Neural Network(CNN) is very similar to Neural Network(NN) as they also constitute learnable weights and bias plus. In it, every neuron gets some input, implements a scalar product and follows it with nonlinearity optionally[50]. However, a particular differentiable score function is …show more content…

1 from keras . layers import Conv2D, MaxPooling2D 2 from keras import backend as K_bend 3 seed = 7 4 np . random. seed ( seed ) 5 no_classes = 10 6 epochs = 10 7 img_rows , img_cols = 28 , 28 8 ( x_trn , y_trn ) , ( x_tst , y_t s t ) = mnist . load_data ( ) Preprocessing the input data Like MLP, the training and test set were reshaped and changed to ’float’ data type from ’int’. Also, the pixel values were normalized to the range 0 and 1, and one hot encode the output variables. 1 # reshaping to be of the form [ samples ] [ px ] [wd] [ ht ] 2 x_trn = x_trn . reshape ( x_trn . shape [ 0 ] , 1 , 28 , 28) . astype ( ’ f loat32 ’ ) 3 x_t s t = x_t s t . reshape ( x_t s t . shape [ 0 ] , 1 , 28 , 28) . astype ( ’ f loat32 ’ ) 4 # normalising inputs between 0¡1 5 x_trn = x_trn / 255 6 x_t s t = x_t s t / 255 7 y_trn = np_ut i ls . t o _ cat ego r i cal ( y_trn ) 8 y_t s t = np_ut i ls . t o _ cat ego r i cal ( y_t s t ) 9 no_classes = y_t s t . shape [ 1 ] Define and compile the Neural network model The CNN’s are more complicated as compared to standard multi-layer perceptrons. That is why a simple structure was chosen to start with that utilizes all of the components for state of the art results. Below is the summarized description of network architecture used. 1 # creat ing sequential_model 2 seq_mdl = Sqntl ( ) 3 seq_mdl . add (Conv2D(32 ,

Get Access
Related
Get Access