Adline stands for adaptive linear neuron. It makes use of linear activation function, and it uses the delta rule for training to minimize the mean squared errors between the actual output and the desired target output. The weights and bias are adjustable. Here, we perform 10 epochs of training and calculate total mean error in each case, the total mean error decreases after certain epochs and later becomes nearly constant.
Truth table for OR gate
| x | y | x or y |
|---|---|---|
| -1 | -1 | -1 |
| -1 | 1 | 1 |
| 1 | -1 | 1 |
| 1 | 1 | 1 |
Below is the implementation.
Python3
# import the module numpyimport numpy as np# the features for the or model , here we have# taken the possible values for combination of# two inputsfeatures = np.array( [ [-1, -1], [-1, 1], [1, -1], [1, 1] ])# labels for the or model, here the output for # the features is taken as an arraylabels = np.array([-1, 1, 1, 1])# to print the features and the labels for # which the model has to be trainedprint(features, labels)# initialise weights, bias , learning rate, epochweight = [0.5, 0.5]bias = 0.1learning_rate = 0.2epoch = 10for i in range(epoch): # epoch is the number of the model is trained # with the same data print("epoch :", i+1) # variable to check if there is no change in previous # weight and present calculated weight # initial error is kept as 0 sum_squared_error = 0.0 # for each of the possible input given in the features for j in range(features.shape[0]): # actual output to be obtained actual = labels[j] # the value of two features as given in the features # array x1 = features[j][0] x2 = features[j][1] # net unit value computation performed to obtain the # sum of features multiplied with their weights unit = (x1 * weight[0]) + (x2 * weight[1]) + bias # error is computed so as to update the weights error = actual - unit # print statement to print the actual value , predicted # value and the error print("error =", error) # summation of squared error is calculated sum_squared_error += error * error # updation of weights, summing up of product of learning rate , # sum of squared error and feature value weight[0] += learning_rate * error * x1 weight[1] += learning_rate * error * x2 # updation of bias, summing up of product of learning rate and # sum of squared error bias += learning_rate * error print("sum of squared error = ", sum_squared_error/4, "\n\n") |
Output:
[[-1 -1] [-1 1] [ 1 -1] [ 1 1]] [-1 1 1 1] epoch : 1 error = -0.09999999999999998 error = 0.9199999999999999 error = 1.1039999999999999 error = -0.5247999999999999 sum of squared error = 0.5876577599999998 epoch : 2 error = -0.54976 error = 0.803712 error = 0.8172543999999999 error = -0.64406528 sum of squared error = 0.5077284689412096 epoch : 3 error = -0.6729103360000002 error = 0.7483308032 error = 0.7399630438400001 error = -0.6898669486079996 sum of squared error = 0.5090672560860652 epoch : 4 error = -0.7047962935296 error = 0.72625757847552 error = 0.7201693816586239 error = -0.7061914301759491 sum of squared error = 0.5103845399996764 epoch : 5 error = -0.7124421954738586 error = 0.7182636328518943 error = 0.7154472043637898 error = -0.7117071786082882 sum of squared error = 0.5104670846209363 epoch : 6 error = -0.714060481354338 error = 0.715548426006041 error = 0.7144420989392495 error = -0.7134930727032405 sum of squared error = 0.5103479496309858 epoch : 7 error = -0.7143209120714415 error = 0.7146705871452027 error = 0.7142737539596766 error = -0.7140502797165604 sum of squared error = 0.5102658027779979 epoch : 8 error = -0.7143272889928647 error = 0.7143984993919014 error = 0.7142647152041359 error = -0.7142182126044045 sum of squared error = 0.510227607583693 epoch : 9 error = -0.7143072010372341 error = 0.7143174255259156 error = 0.7142744539151652 error = -0.7142671011374249 sum of squared error = 0.5102124122866718 epoch : 10 error = -0.7142946765305948 error = 0.7142942165270032 error = 0.7142809804050706 error = -0.7142808151475037 sum of squared error = 0.5102068786350209