diff options
| -rwxr-xr-x | src/ann_mlp.c | 471 |
1 files changed, 320 insertions, 151 deletions
diff --git a/src/ann_mlp.c b/src/ann_mlp.c index 46bdfb0..2cc7b01 100755 --- a/src/ann_mlp.c +++ b/src/ann_mlp.c @@ -5,6 +5,12 @@ FANN is obviously needed for compilation this software is licensed under the GNU General Public License */ + +/* + hacked by Georg Holzmann for some additional methods, bug fixes, ... + 2005, grh@mur.at +*/ + #include <stdio.h> #include <string.h> #include "m_pd.h" @@ -21,9 +27,6 @@ #define TRAIN 0 #define RUN 1 -#define MAXINPUT 1024 -#define MAXOUTPUT 256 - static t_class *ann_mlp_class; typedef struct _ann_mlp { @@ -35,9 +38,47 @@ typedef struct _ann_mlp { float desired_error; unsigned int max_iterations; unsigned int iterations_between_reports; + fann_type *input; // grh: storage for input + t_atom *output; // grh: storage for output (t_atom) + fann_type *out_float; // grh: storage for output (fann_type) + t_canvas *x_canvas; t_outlet *l_out, *f_out; } t_ann_mlp; +// allocation +static void ann_mlp_allocate_storage(t_ann_mlp *x) +{ + int i; + + if(!x->ann) + return; + + x->input = (fann_type *)getbytes(x->ann->num_input*sizeof(fann_type)); + x->output = (t_atom *)getbytes(x->ann->num_output*sizeof(t_atom)); + x->out_float = (fann_type *)getbytes(x->ann->num_output*sizeof(fann_type)); + + // init storage with zeros + for (i=0; i<x->ann->num_input; i++) + x->input[i]=0; + for (i=0; i<x->ann->num_output; i++) + { + SETFLOAT(x->output+i, 0); + x->out_float[i]=0; + } +} + +// deallocation +static void ann_mlp_free(t_ann_mlp *x) +{ + if(!x->ann) + return; + + freebytes(x->input, x->ann->num_input * sizeof(fann_type)); + freebytes(x->output, x->ann->num_output * sizeof(t_atom)); + freebytes(x->out_float, x->ann->num_output * sizeof(fann_type)); + fann_destroy(x->ann); +} + static void ann_mlp_help(t_ann_mlp *x) { post(""); @@ -55,59 +96,94 @@ static void ann_mlp_createFann(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *arg unsigned int num_input = 2; unsigned int num_output = 1; unsigned int num_layers = 3; - unsigned int num_neurons_hidden = 3; + unsigned int *neurons_per_layer = NULL; + int activated=0; + int i, count_args = 0; float connection_rate = 1; float learning_rate = (float)0.7; - - if (argc>0) + + + // okay, start parsing init args ... + + if (argc > count_args++) num_input = atom_getint(argv++); - if (argc>1) + if (argc > count_args++) num_output = atom_getint(argv++); - if (argc>2) - num_layers = atom_getint(argv++); - - if (argc>3) - num_neurons_hidden = atom_getint(argv++); - - if (argc>4) + if (argc > count_args++) + { + int hidden=0; + + num_layers = atom_getint(argv++); + hidden = num_layers-2; + + neurons_per_layer = (unsigned int *)getbytes(num_layers*sizeof(unsigned int)); + + neurons_per_layer[0] = num_input; + + // make standard initialization (if there are too few init args) + for (i=1; i<hidden+1; i++) + neurons_per_layer[i] = 3; + + // now check init args + for (i=1; i<hidden+1; i++) + { + if (argc > count_args++) + neurons_per_layer[i] = atom_getint(argv++); + } + + neurons_per_layer[num_layers-1] = num_output; + + activated=1; + } + + if (argc > count_args++) connection_rate = atom_getfloat(argv++); - if (argc>5) + if (argc > count_args++) learning_rate = atom_getfloat(argv++); - if (num_input>MAXINPUT) - { - error("too many inputs, maximum allowed is MAXINPUT"); - return; - } - - if (num_output>MAXOUTPUT) - { - error("too many outputs, maximum allowed is MAXOUTPUT"); - return; - } - - x->ann = fann_create(connection_rate, learning_rate, num_layers, - num_input, num_neurons_hidden, num_output); - + // make one hidden layer as standard, if there were too few init args + if(!activated) + { + neurons_per_layer = (unsigned int *)getbytes(3*sizeof(unsigned int)); + neurons_per_layer[0] = num_input; + neurons_per_layer[1] = 3; + neurons_per_layer[2] = num_output; + } + + // ... end of parsing init args + + + if(x->ann) + ann_mlp_free(x); + + x->ann = fann_create_array(connection_rate, learning_rate, num_layers, neurons_per_layer); + + // deallocate helper array + freebytes(neurons_per_layer, num_layers * sizeof(unsigned int)); + + if(!x->ann) + { + error("error creating the ann"); + return; + } + + ann_mlp_allocate_storage(x); fann_set_activation_function_hidden(x->ann, FANN_SIGMOID_SYMMETRIC); fann_set_activation_function_output(x->ann, FANN_SIGMOID_SYMMETRIC); - - if (x->ann == 0) - { - error("error creating the ann"); - } else - { - post("created ann with:"); - post("num_input = %i", num_input); - post("num_output = %i", num_output); - post("num_layers = %i", num_layers); - post("num_neurons_hidden = %i", num_neurons_hidden); - post("connection_rate = %f", connection_rate); - post("learning_rate = %f", learning_rate); - } + + // set error log to stdout, so that you see it in the pd console + //fann_set_error_log((struct fann_error*)x->ann, stdout); + // unfortunately this doesn't work ... but it should do in a similar way !! + + post("created ann with:"); + post("num_input = %i", num_input); + post("num_output = %i", num_output); + post("num_layers = %i", num_layers); + post("connection_rate = %f", connection_rate); + post("learning_rate = %f", learning_rate); } static void ann_mlp_print_status(t_ann_mlp *x) @@ -149,26 +225,27 @@ static void ann_mlp_set_mode(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) } } - - -static void ann_mlp_train_on_file(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +static void ann_mlp_train_on_file(t_ann_mlp *x, t_symbol *s) { + // make correct path + char patcher_path[MAXPDSTRING]; + char filename[MAXPDSTRING]; + if (x->ann == 0) { error("ann not initialized"); return; } - if (argc<1) - { - error("you must specify the filename with training data"); - return; - } else - { - x->filenametrain = atom_gensym(argv); - } + // make correct path + canvas_makefilename(x->x_canvas, s->s_name, patcher_path, MAXPDSTRING); + sys_bashfilename(patcher_path, filename); + x->filenametrain = gensym(filename); - //post("nn: starting training on file %s, please be patient and wait for my next message (it could take severeal minutes to complete training)", x->filenametrain->s_name); + if(!x->filenametrain) + return; + + post("nn: starting training on file %s, please be patient and wait ... (it could take severeal minutes to complete training)", x->filenametrain->s_name); fann_train_on_file(x->ann, x->filenametrain->s_name, x->max_iterations, x->iterations_between_reports, x->desired_error); @@ -220,72 +297,47 @@ static void ann_mlp_set_iterations_between_reports(t_ann_mlp *x, t_symbol *sl, i } - // run the ann using floats in list passed to the inlet as input values // and send result to outlet as list of float static void ann_mlp_run_the_net(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) { - int i=0; - fann_type input[MAXINPUT]; + int i=0; fann_type *calc_out; - t_atom lista[MAXOUTPUT]; - int quanti; - float valoreTMP; if (x->ann == 0) { error("ann not initialized"); return; } - - quanti = x->ann->num_output; - - // fill input array with zeros - for (i=0; i<MAXINPUT; i++) - { - input[i]=0; - } - - // fill output array with zeros - for (i=0; i<MAXOUTPUT; i++) - { - SETFLOAT(lista + i,0); - } + + if(argc < x->ann->num_input) + { + error("ann_mlp: too few input values!!"); + return; + } // fill input array with actual data sent to inlet - for (i=0;i<argc;i++) + for (i=0;i<x->ann->num_input;i++) { - input[i] = atom_getfloat(argv++); + x->input[i] = atom_getfloat(argv++); } // run the ann - calc_out = fann_run(x->ann, input); + calc_out = fann_run(x->ann, x->input); // fill the output array with result from ann - for (i=0;i<quanti;i++) - { - valoreTMP = calc_out[i]; - //post("calc_out[%i]=%f", i, calc_out[i]); - SETFLOAT(lista+i, valoreTMP); - } + for (i=0;i<x->ann->num_output;i++) + SETFLOAT(x->output+i, calc_out[i]); // send output array to outlet - outlet_anything(x->l_out, - gensym("list") , - quanti, - lista); - + outlet_anything(x->l_out, gensym("list"), + x->ann->num_output, x->output); } static void ann_mlp_train_on_the_fly(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) { int i=0; - fann_type input[MAXINPUT]; - fann_type output[MAXOUTPUT]; - //fann_type *calcMSE; - //t_atom lista[MAXOUTPUT]; - int quantiINs; - int quantiOUTs; + int quantiINs, quantiOUTs; float mse; if (x->ann == 0) @@ -303,37 +355,20 @@ static void ann_mlp_train_on_the_fly(t_ann_mlp *x, t_symbol *sl, int argc, t_ato return; } - // fill input array with zeros - for (i=0; i<MAXINPUT; i++) - { - input[i]=0; - } - // fill input array with zeros - for (i=0; i<MAXOUTPUT; i++) - { - output[i]=0; - } - // fill input array with actual data sent to inlet for (i=0;i<quantiINs;i++) - { - input[i] = atom_getfloat(argv++); - } + x->input[i] = atom_getfloat(argv++); for (i=0;i<quantiOUTs;i++) - { - output[i] = atom_getfloat(argv++); - } + x->out_float[i] = atom_getfloat(argv++); //fann_reset_MSE(x->ann); - fann_train(x->ann, input, output); + fann_train(x->ann, x->input, x->out_float); mse = fann_get_MSE(x->ann); outlet_float(x->f_out, mse); - - } static void ann_mlp_manage_list(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) @@ -346,35 +381,57 @@ static void ann_mlp_manage_list(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *ar } } -static void ann_mlp_set_filename(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +static void ann_mlp_set_filename(t_ann_mlp *x, t_symbol *s) { - if (argc>0) { - x->filename = atom_gensym(argv); - } else - { - error("you must specify the filename"); - } - post("nn:filename set to %s", x->filename->s_name); + // make correct path + char patcher_path[MAXPDSTRING]; + char filename[MAXPDSTRING]; + + if(!s) + return; + + // make correct path + canvas_makefilename(x->x_canvas, s->s_name, patcher_path, MAXPDSTRING); + sys_bashfilename(patcher_path, filename); + x->filename = gensym(filename); } -static void ann_mlp_load_ann_from_file(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +static void ann_mlp_load_ann_from_file(t_ann_mlp *x, t_symbol *s) { - if (argc>0) { - x->filename = atom_gensym(argv); - } - x->ann = fann_create_from_file(x->filename->s_name); - if (x->ann == 0) - error("error opening %s", x->filename->s_name); - else - post("nn:ann loaded fom file %s", x->filename->s_name); + ann_mlp_set_filename(x,s); + + if(!x->filename) + { + error("ann: no filename !!!"); + return; + } + + // deallocate storage + if(x->ann) + ann_mlp_free(x); + + x->ann = fann_create_from_file(x->filename->s_name); + + if (x->ann == 0) + error("error opening %s", x->filename->s_name); + else + post("nn:ann loaded fom file %s", x->filename->s_name); + + // allocate storage + ann_mlp_allocate_storage(x); } -static void ann_mlp_save_ann_to_file(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +static void ann_mlp_save_ann_to_file(t_ann_mlp *x, t_symbol *s) { - if (argc>0) { - x->filename = atom_gensym(argv); - } - if (x->ann == 0) + ann_mlp_set_filename(x,s); + + if(!x->filename) + { + error("ann: no filename !!!"); + return; + } + + if (x->ann == 0) { error("ann is not initialized"); } else @@ -466,6 +523,100 @@ static void ann_mlp_set_activation_function_output(t_ann_mlp *x, t_symbol *sl, i } +static void ann_mlp_set_activation_function_hidden(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +{ + t_symbol *parametro = 0; + int funzione = 0; + + if (x->ann == 0) + { + error("ann not initialized"); + return; + } + + if (argc>0) { + parametro = atom_gensym(argv); + if (strcmp(parametro->s_name, "FANN_THRESHOLD")==0) + funzione = FANN_THRESHOLD; + if (strcmp(parametro->s_name, "FANN_THRESHOLD_SYMMETRIC")==0) + funzione = FANN_THRESHOLD_SYMMETRIC; + if (strcmp(parametro->s_name, "FANN_LINEAR")==0) + funzione = FANN_LINEAR; + if (strcmp(parametro->s_name, "FANN_SIGMOID")==0) + funzione = FANN_SIGMOID; + if (strcmp(parametro->s_name, "FANN_SIGMOID_STEPWISE")==0) + funzione = FANN_SIGMOID_STEPWISE; + if (strcmp(parametro->s_name, "FANN_SIGMOID_SYMMETRIC")==0) + funzione = FANN_SIGMOID_SYMMETRIC; + if (strcmp(parametro->s_name, "FANN_SIGMOID_SYMMETRIC_STEPWISE")==0) + funzione = FANN_SIGMOID_SYMMETRIC_STEPWISE; + fann_set_activation_function_hidden(x->ann, funzione); + } else + { + error("you must specify the activation function"); + } + post("nn:activation function set to %s (%i)", parametro->s_name, funzione); + +} + +static void ann_mlp_randomize_weights(t_ann_mlp *x, t_symbol *sl, int argc, t_atom *argv) +{ + t_float min = -1; + t_float max = 1; + + if(!x->ann) + { + post("ann_mlp: ann is not initialized"); + return; + } + + if (argc>0) + min = atom_getfloat(argv++); + + if (argc>1) + max = atom_getfloat(argv++); + + fann_randomize_weights(x->ann, min, max); +} + +static void ann_mlp_learnrate(t_ann_mlp *x, t_float f) +{ + int learnrate = 0; + + if(!x->ann) + { + post("ann_mlp: ann is not initialized"); + return; + } + + learnrate = (f<0) ? 0 : f; + fann_set_learning_rate(x->ann, learnrate); +} + +static void ann_mlp_set_activation_steepness_hidden(t_ann_mlp *x, t_float f) +{ + if(!x->ann) + { + post("ann_mlp: ann is not initialized"); + return; + } + + fann_set_activation_steepness_hidden(x->ann, f); +} + +static void ann_mlp_set_activation_steepness_output(t_ann_mlp *x, t_float f) +{ + if(!x->ann) + { + post("ann_mlp: ann is not initialized"); + return; + } + + fann_set_activation_steepness_output(x->ann, f); +} + +void fann_set_activation_steepness_hidden(struct fann * ann, fann_type steepness); + static void ann_mlp_print_ann_details(t_ann_mlp *x) { if (x->ann == 0) @@ -491,6 +642,17 @@ static void ann_mlp_print_ann_details(t_ann_mlp *x) } } +static void ann_mlp_print_ann_print(t_ann_mlp *x) +{ + if(!x->ann) + { + post("ann_mlp: ann is not initialized"); + return; + } + + fann_print_connections(x->ann); + fann_print_parameters(x->ann); +} static void *ann_mlp_new(t_symbol *s, int argc, t_atom *argv) { @@ -502,23 +664,22 @@ static void *ann_mlp_new(t_symbol *s, int argc, t_atom *argv) x->max_iterations = 500000; x->iterations_between_reports = 1000; x->mode=RUN; + x->x_canvas = canvas_getcurrent(); + x->filename = NULL; + x->filenametrain = NULL; + x->ann = NULL; + x->input = NULL; + x->output = NULL; + x->out_float = NULL; if (argc>0) { x->filename = atom_gensym(argv); - ann_mlp_load_ann_from_file(x, NULL , 0, NULL); + ann_mlp_load_ann_from_file(x, NULL); } return (void *)x; } -// free resources -static void ann_mlp_free(t_ann_mlp *x) -{ - struct fann *ann = x->ann; - fann_destroy(ann); - // TODO: free other resources! -} - void ann_mlp_setup(void) { post(""); post("ann_mlp: multilayer perceptron for PD"); @@ -538,17 +699,19 @@ void ann_mlp_setup(void) { class_addmethod(ann_mlp_class, (t_method)ann_mlp_train, gensym("train"), 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_run, gensym("run"), 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_mode, gensym("setmode"), A_GIMME, 0); - class_addmethod(ann_mlp_class, (t_method)ann_mlp_train_on_file, gensym("train-on-file"), A_GIMME, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_train_on_file, gensym("train-on-file"), A_DEFSYMBOL, 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_manage_list, gensym("data"), A_GIMME, 0); - class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_filename, gensym("filename"), A_GIMME, 0); - class_addmethod(ann_mlp_class, (t_method)ann_mlp_load_ann_from_file, gensym("load"),A_GIMME, 0); - class_addmethod(ann_mlp_class, (t_method)ann_mlp_save_ann_to_file, gensym("save"),A_GIMME, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_filename, gensym("filename"), A_DEFSYMBOL, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_load_ann_from_file, gensym("load"),A_DEFSYMBOL, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_save_ann_to_file, gensym("save"),A_DEFSYMBOL, 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_print_ann_details, gensym("details"), 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_print_ann_print, gensym("print"), 0); // change training parameters class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_desired_error, gensym("desired_error"),A_GIMME, 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_max_iterations, gensym("max_iterations"),A_GIMME, 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_iterations_between_reports, gensym("iterations_between_reports"),A_GIMME, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_learnrate, gensym("learnrate"), A_FLOAT, 0); // change training and activation algorithms class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_FANN_TRAIN_INCREMENTAL, gensym("FANN_TRAIN_INCREMENTAL"), 0); @@ -556,7 +719,13 @@ void ann_mlp_setup(void) { class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_FANN_TRAIN_RPROP, gensym("FANN_TRAIN_RPROP"), 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_FANN_TRAIN_QUICKPROP, gensym("FANN_TRAIN_QUICKPROP"), 0); class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_activation_function_output, gensym("set_activation_function_output"),A_GIMME, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_activation_function_hidden, gensym("set_activation_function_hidden"),A_GIMME, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_activation_steepness_hidden, gensym("set_activation_steepness_hidden"), A_FLOAT, 0); + class_addmethod(ann_mlp_class, (t_method)ann_mlp_set_activation_steepness_output, gensym("set_activation_steepness_output"), A_FLOAT, 0); + // initialization: + class_addmethod(ann_mlp_class, (t_method)ann_mlp_randomize_weights, gensym("randomize_weights"),A_GIMME, 0); + // the most important one: running the ann class_addlist(ann_mlp_class, (t_method)ann_mlp_manage_list); |