/* --- Copyright University of Birmingham 2000. All rights reserved. ------ > File: $poplocal/local/rclib/demo/rc_neural.p > Purpose: Demonstrate simple neural nets > Author: Riccardo Poli, October 1997 (see revisions) > Documentation: > Related Files: */ /* rc_neural.p Artificial Neuron Simulator in Pop-11 This is based on workshop5.p (the answers to the exercises in SEM1A9 workshop 5) plus a graphical interface. To use this program, just compile this file and execute one of the train_net(...) commands at the end of this file. A graphical panel will appear. This represents the weights, bias, output and inputs of an artificial neuron with step activation function. You will see that the weights and the bias change during the learning phase. After the training is completed you can use the network by changing weights, bias or inputs and then activating the network pressing the "propagate" button. */ uses rclib; uses rc_control_panel; vars rc_neural_panel = false, rc_neural_ninputs = 0; /* Solution to Exercise A */ define activation(net) -> result; if net > 0 then 1 -> result; else 0 -> result; endif; enddefine; /* activation(2.5) => activation(0.0) => activation(-1.0) => */ /* Solution to Exercise B */ define net_input(inputs,weights,bias) -> result; lvars i, w; bias -> result; for i, w in inputs, weights do i * w + result -> result; endfor; enddefine; /* net_input([1 0],[1 1],-1) => net_input([1 0 1],[1 1 -1],1) => */ /* Solution to Exercise C */ define neuron(inputs,weights,bias) -> result; activation(net_input(inputs,weights,bias)) -> result; enddefine; /* neuron([1 0],[1 1],-1) => neuron([1 1],[1 1],1) => neuron([1 0 1],[1 1 -1],1) => neuron([1 0 1],[1 1 -1],-1) => */ /* Solution to Exercise D */ define error(inputs,weights,bias,target) -> result; neuron(inputs,weights,bias) - target -> result; enddefine; /* error([1 0],[1 1],-1,1) => error([1 1],[1 1],1,0) => error([1 1],[1 1],1,1) => error([1 0 1],[1 1 -1],1,0) => error([1 0 1],[1 1 -1],-1,0) => */ /* Solution to Exercise E */ define train(inputs,weights,bias,target) -> ( new_weights, new_bias ); lvars e, w, i; error(inputs,weights,bias,target) -> e; if e == 0 then weights -> new_weights; bias -> new_bias; elseif e > 0 then bias - 1 -> new_bias; [% for w, i in weights, inputs do if i == 1 then w - 1; else w; endif; endfor; %] -> new_weights; else bias + 1 -> new_bias; [% for w, i in weights, inputs do if i == 1 then w + 1; else w; endif; endfor; %] -> new_weights; endif; enddefine; /* train([1 0],[1 1],-1,1) => train([1 1],[1 1],1,0) => train([1 1],[1 1],1,1) => train([1 0 1],[1 1 -1],1,0) => train([1 0 1],[1 1 -1],-1,0) => */ /* Solution to Exercise F */ define train_all(training_set,weights,bias) -> (new_weights,new_bias); lvars example; weights -> new_weights; bias -> new_bias; for example in training_set do train(example(1),new_weights,new_bias,example(2)) -> (new_weights, new_bias); endfor; enddefine; /* train_all([[[1 0] 1][[0 1] 1]],[1 1],-1) => train_all([[[1 0] 1][[0 1] 0]],[1 1],-1) => train_all([[[1 0] 1][[0 1] 0][[1 1] 0]],[1 1],-1) => train_all([[[0 0] 0][[1 0] 1][[0 1] 0][[1 1] 0]],[1 1],-1) => */ define rc_neural_setup(ninputs ); lvars x, slider_specs = [ {width panel} {radius 6} {fieldfg 'blue'} {offset 15} {gap 0} {barcol 'blue'} {blobcol 'red'} {framewidth 4}], panel_specs = [ {width 300} {bg 'black'} [TEXT {margin 5} {align centre} : 'Artificial Neuron Demo' 'Inputs'] [SLIDERS ^slider_specs {framecol 'green'} {label inputs}: % for x to ninputs do [{0 1 0} indentfn [{-5 12 %'i'> rc_neural_panel]] ] ]; /* dlocal ;;; Make the blobs fit in the slider bars rc_slider_blob_bar_ratio = 1; */ rc_control_panel(10, 10, panel_specs, 'Artificial Neuron Demo') -> rc_neural_panel; ninputs -> rc_neural_ninputs; enddefine; /* rc_neural_setup( 5 ); */ define rc_neural_propagate(); lvars x, bias, inputs, weights; slider_value_of_name(rc_neural_panel,"bias", 1) -> bias; [% for x to rc_neural_ninputs do slider_value_of_name(rc_neural_panel,"weights", x); endfor; %] -> weights; [% for x to rc_neural_ninputs do slider_value_of_name(rc_neural_panel,"inputs", x); endfor; %] -> inputs; neuron(inputs,weights,bias) -> slider_value_of_name(rc_neural_panel,"output", 1); enddefine; define rc_neural_set_weights(weights, bias); lvars x; if not(rc_neural_panel) or length(weights) /== rc_neural_ninputs then rc_neural_setup( length(weights) ) endif; bias -> slider_value_of_name(rc_neural_panel,"bias", 1); for x to rc_neural_ninputs do weights(x) -> slider_value_of_name(rc_neural_panel,"weights", x); endfor; enddefine; /* rc_neural_set_weights([1 1 2 3],1); */ /* Solution to Exercise G */ define train_net(training_set) -> (new_weights,new_bias); lvars weights, bias, ninputs = length(hd(hd(training_set))); [% repeat ninputs times random(11)-5; endrepeat; %] -> weights; random(11) - 5 -> bias; repeat 100 times pr('w='); pr(weights); pr(' b='); npr(bias); rc_neural_set_weights(weights, bias); syssleep(slider_value_of_name(rc_neural_panel,"delay", 1)); train_all(training_set,weights,bias) -> (new_weights,new_bias); if weights = new_weights and bias == new_bias then return endif; (new_weights,new_bias) -> (weights,bias); endrepeat; ( false, false ) -> (new_weights,new_bias); enddefine; /* train_net([[[1 0] 1][[0 1] 1]]) => train_net([[[1 0] 1][[0 1] 0]]) => train_net([[[1 0] 1][[0 1] 0][[1 1] 0]]) => train_net([[[0 0] 0][[1 0] 1][[0 1] 0][[1 1] 0]]) => ;;; OR function train_net([[[0 0] 0][[0 1] 1][[1 0] 1][[1 1] 1]]) => ;;; AND function train_net([[[0 0] 0][[0 1] 0][[1 0] 0][[1 1] 1]]) => ;;; XOR function train_net([[[0 0] 0][[0 1] 1][[1 0] 1][[1 1] 0]]) => ;;; MAJORITY function (3 inputs) train_net([[[0 0 0] 0] [[0 0 1] 0] [[0 1 0] 0] [[0 1 1] 1] [[1 0 0] 0] [[1 0 1] 1] [[1 1 0] 1] [[1 1 1] 1]]); ;;; EVEN-3 Parity function train_net([[[0 0 0] 1] [[0 0 1] 0] [[0 1 0] 0] [[0 1 1] 1] [[1 0 0] 0] [[1 0 1] 1] [[1 1 0] 1] [[1 1 1] 0]]); */ /* --- Revision History --------------------------------------------------- --- Aaron Sloman, Jul 4 2000 Changed colours, because of new slider facilities: slider backgrounds no longer need to be white. --- Aaron Sloman, Apr 19 1999 Slightly modified slider specs again --- Aaron Sloman, Feb 6 1998 Altered to use new slider specs */