Class: Neuro::Network

Inherits:
Object
  • Object
show all
Defined in:
ext/neuro.c

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#new(input_size, hidden_size, output_size) ⇒ Object

Returns a Neuro::Network instance of the given size specification.



548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
 
# File 'ext/neuro.c', line 548

static VALUE rb_network_initialize(int argc, VALUE *argv, VALUE self)
{
    Network *network;
    VALUE input_size, hidden_size, output_size;

    rb_scan_args(argc, argv, "3", &input_size, &hidden_size, &output_size);
  Check_Type(input_size, T_FIXNUM);
  Check_Type(hidden_size, T_FIXNUM);
  Check_Type(output_size, T_FIXNUM);
    Data_Get_Struct(self, Network, network);
    Network_init(network, NUM2INT(input_size), NUM2INT(hidden_size),
        NUM2INT(output_size), 0);
    Network_init_weights(network);
    return self;
}

Class Method Details

.Neuro::Network.load(string) ⇒ Object

Creates a Network object plus state from the Marshal dumped string string, and returns it.



597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
 
# File 'ext/neuro.c', line 597

static VALUE rb_network_load(VALUE klass, VALUE string)
{
    VALUE input_size, hidden_size, output_size, learned,
        hidden_layer, output_layer, pair[2];
    Network *network;
    VALUE hash = rb_marshal_load(string);
    input_size = rb_hash_aref(hash, SYM("input_size"));
    hidden_size = rb_hash_aref(hash, SYM("hidden_size"));
    output_size = rb_hash_aref(hash, SYM("output_size"));
    learned = rb_hash_aref(hash, SYM("learned"));
  Check_Type(input_size, T_FIXNUM);
  Check_Type(hidden_size, T_FIXNUM);
  Check_Type(output_size, T_FIXNUM);
  Check_Type(learned, T_FIXNUM);
    network = Network_allocate();
    Network_init(network, NUM2INT(input_size), NUM2INT(hidden_size),
            NUM2INT(output_size), NUM2INT(learned));
    hidden_layer = rb_hash_aref(hash, SYM("hidden_layer"));
    output_layer = rb_hash_aref(hash, SYM("output_layer"));
    Check_Type(hidden_layer, T_ARRAY);
    Check_Type(output_layer, T_ARRAY);
    pair[0] = (VALUE) network->hidden_layer;
    pair[1] = (VALUE) 0;
    rb_iterate(rb_each, hidden_layer, setup_layer_i, (VALUE) pair);
    pair[0] = (VALUE) network->output_layer;
    pair[1] = (VALUE) 0;
    rb_iterate(rb_each, output_layer, setup_layer_i, (VALUE) pair);
    return Data_Wrap_Struct(klass, NULL, rb_network_free, network);
}

.Neuro::Network.load(string) ⇒ Object

Creates a Network object plus state from the Marshal dumped string string, and returns it.



597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
 
# File 'ext/neuro.c', line 597

static VALUE rb_network_load(VALUE klass, VALUE string)
{
    VALUE input_size, hidden_size, output_size, learned,
        hidden_layer, output_layer, pair[2];
    Network *network;
    VALUE hash = rb_marshal_load(string);
    input_size = rb_hash_aref(hash, SYM("input_size"));
    hidden_size = rb_hash_aref(hash, SYM("hidden_size"));
    output_size = rb_hash_aref(hash, SYM("output_size"));
    learned = rb_hash_aref(hash, SYM("learned"));
  Check_Type(input_size, T_FIXNUM);
  Check_Type(hidden_size, T_FIXNUM);
  Check_Type(output_size, T_FIXNUM);
  Check_Type(learned, T_FIXNUM);
    network = Network_allocate();
    Network_init(network, NUM2INT(input_size), NUM2INT(hidden_size),
            NUM2INT(output_size), NUM2INT(learned));
    hidden_layer = rb_hash_aref(hash, SYM("hidden_layer"));
    output_layer = rb_hash_aref(hash, SYM("output_layer"));
    Check_Type(hidden_layer, T_ARRAY);
    Check_Type(output_layer, T_ARRAY);
    pair[0] = (VALUE) network->hidden_layer;
    pair[1] = (VALUE) 0;
    rb_iterate(rb_each, hidden_layer, setup_layer_i, (VALUE) pair);
    pair[0] = (VALUE) network->output_layer;
    pair[1] = (VALUE) 0;
    rb_iterate(rb_each, output_layer, setup_layer_i, (VALUE) pair);
    return Data_Wrap_Struct(klass, NULL, rb_network_free, network);
}

Instance Method Details

#_dump(*args) ⇒ Object

Returns the serialized data for this Network instance for the Marshal module.



568
569
570
571
572
573
574
575
576
577
 
# File 'ext/neuro.c', line 568

static VALUE rb_network_dump(int argc, VALUE *argv, VALUE self)
{
    VALUE port = Qnil, hash;
    Network *network;

    rb_scan_args(argc, argv, "01", &port);
    Data_Get_Struct(self, Network, network);
    hash = Network_to_hash(network);
    return rb_marshal_dump(hash, port);
}

#debug ⇒ Object

Returns nil, if debugging is switchted off. Returns the IO object, that is used for debugging output, if debugging is switchted on.



393
394
395
396
397
398
399
 
# File 'ext/neuro.c', line 393

static VALUE rb_network_debug(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return network->debug;
}

#debug=(io) ⇒ Object

Switches debugging on, if io is an IO object. If it is nil, debugging is switched off.



407
408
409
410
411
412
413
414
 
# File 'ext/neuro.c', line 407

static VALUE rb_network_debug_set(VALUE self, VALUE io)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    network->debug = io;
    return io;
}

#debug_step ⇒ Object

Returns the Integer number of steps, that are done during learning, before a debugging message is printed to #debug.



420
421
422
423
424
425
426
 
# File 'ext/neuro.c', line 420

static VALUE rb_network_debug_step(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->debug_step);
}

#debug_step=(step) ⇒ Object

Sets the number of steps, that are done during learning, before a debugging message is printed to step. If step is equal to or less than 0 the default value (=1000) is set.



435
436
437
438
439
440
441
442
443
444
 
# File 'ext/neuro.c', line 435

static VALUE rb_network_debug_step_set(VALUE self, VALUE step)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    Check_Type(step, T_FIXNUM);
    network->debug_step = NUM2INT(step);
    if (network->debug_step <= 0) network->debug_step = DEFAULT_DEBUG_STEP;
    return step;
}

#decide(data) ⇒ Object

The network is given the Array data (size has to be == input_size), and it responds with another Array (size == output_size) by returning it.



321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
 
# File 'ext/neuro.c', line 321

static VALUE rb_network_decide(VALUE self, VALUE data)
{
    Network *network;
    VALUE result;
    int i;

    Data_Get_Struct(self, Network, network);

  Check_Type(data, T_ARRAY);
    if (RARRAY(data)->len != network->input_size)
        rb_raise(rb_cNeuroError, "size of data != input_size");
    transform_data(network->tmp_input, data);
    feed;
    result = rb_ary_new2(network->output_size);
    for (i = 0; i < network->output_size; i++) {
        rb_ary_store(result, i,
            rb_float_new(network->output_layer[i]->output));
    }
    return result;
}

#dump(*args) ⇒ Object

Returns the serialized data for this Network instance for the Marshal module.



568
569
570
571
572
573
574
575
576
577
 
# File 'ext/neuro.c', line 568

static VALUE rb_network_dump(int argc, VALUE *argv, VALUE self)
{
    VALUE port = Qnil, hash;
    Network *network;

    rb_scan_args(argc, argv, "01", &port);
    Data_Get_Struct(self, Network, network);
    hash = Network_to_hash(network);
    return rb_marshal_dump(hash, port);
}

#hidden_size ⇒ Object

Returns the hidden_size of this Network as an Integer. This is the number of nodes in the hidden layer.



358
359
360
361
362
363
364
 
# File 'ext/neuro.c', line 358

static VALUE rb_network_hidden_size(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->hidden_size);
}

#input_size ⇒ Object

Returns the input_size of this Network as an Integer. This is the number of weights, that are connected to the input of the hidden layer.



346
347
348
349
350
351
352
 
# File 'ext/neuro.c', line 346

static VALUE rb_network_input_size(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->input_size);
}

#learn(data, desired, max_error, eta) ⇒ Object

The network should respond with the Array desired (size == output_size), if it was given the Array data (size == input_size). The learning process ends, if the resulting error sinks below max_error and convergence is assumed. A lower eta parameter leads to slower learning, because of low weight changes. A too high eta can lead to wildly oscillating weights, and result in slower learning or no learning at all. The last two parameters should be chosen appropriately to the problem at hand. ;)

The return value is an Integer value, that denotes the number of learning steps, which were necessary, to learn the data, or max_iterations, if the data couldn’t be learned.



234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
 
# File 'ext/neuro.c', line 234

static VALUE rb_network_learn(VALUE self, VALUE data, VALUE desired, VALUE
        max_error, VALUE eta)
{
    Network *network;
    double max_error_float, eta_float, error, sum,
        *output_delta, *hidden_delta;
    int i, j, count;

    Data_Get_Struct(self, Network, network);

  Check_Type(data, T_ARRAY);
    if (RARRAY(data)->len != network->input_size)
        rb_raise(rb_cNeuroError, "size of data != input_size");
    transform_data(network->tmp_input, data);

  Check_Type(desired, T_ARRAY);
    if (RARRAY(desired)->len != network->output_size)
        rb_raise(rb_cNeuroError, "size of desired != output_size");
    transform_data(network->tmp_output, desired);
    CAST2FLOAT(max_error);
    max_error_float = RFLOAT(max_error)->value;
    if (max_error_float <= 0) rb_raise(rb_cNeuroError, "max_error <= 0");
    max_error_float *= 2.0;
    CAST2FLOAT(eta);
    eta_float = RFLOAT(eta)->value;
    if (eta_float <= 0) rb_raise(rb_cNeuroError, "eta <= 0");

    output_delta = ALLOCA_N(double, network->output_size);
    hidden_delta = ALLOCA_N(double, network->hidden_size);
    for(count = 0; count < network->max_iterations; count++) {
        feed;

        /* Compute output weight deltas and current error */
        error = 0.0;    
        for (i = 0; i < network->output_size; i++) {
            output_delta[i] = network->tmp_output[i] -
                network->output_layer[i]->output;
            error += output_delta[i] * output_delta[i];
            output_delta[i] *= network->output_layer[i]->output *
                (1.0 - network->output_layer[i]->output);
            /* diff * (sigmoid' = 2 * output  * beta * (1 - output)) */

        }

        if (count % network->debug_step == 0)
            Network_debug_error(network, count, error, max_error_float);

        /* Get out if error is below max_error ^ 2 */
        if (error < max_error_float) goto CONVERGED;

        /* Compute hidden weight deltas */
        
    for (i = 0; i < network->hidden_size; i++) {
            sum = 0.0;
      for (j = 0; j < network->output_size; j++)
        sum += output_delta[j] *
                    network->output_layer[j]->weights[i];
      hidden_delta[i] = sum * network->hidden_layer[i]->output *
                (1.0 - network->hidden_layer[i]->output);
            /* sum * (sigmoid' = 2 * output  * beta * (1 - output)) */
    }
        
        /* Adjust weights */

    for (i = 0; i < network->output_size; i++)
      for (j = 0; j < network->hidden_size; j++)
                network->output_layer[i]->weights[j] +=
                    eta_float * output_delta[i] *
                    network->hidden_layer[j]->output;

    for (i = 0; i < network->hidden_size; i++)
      for (j = 0; j < network->input_size; j++)
        network->hidden_layer[i]->weights[j] += eta_float *
                    hidden_delta[i] * network->tmp_input[j];
    }
    Network_debug_bail_out(network);
CONVERGED:
    network->learned++;
    return INT2NUM(count);
}

#learned ⇒ Object

Returns the number of calls to #learn as an integer.



381
382
383
384
385
386
387
 
# File 'ext/neuro.c', line 381

static VALUE rb_network_learned(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->learned);
}

#max_iterations ⇒ Object

Returns the maximal number of iterations, that are done before #learn gives up and returns without having learned the given data.



450
451
452
453
454
455
456
 
# File 'ext/neuro.c', line 450

static VALUE rb_network_max_iterations(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->max_iterations);
}

#max_iterations=(iterations) ⇒ Object

Sets the maximal number of iterations, that are done before #learn gives up and returns without having learned the given data, to iterations. If iterations is equal to or less than 0, the default value (=10_000) is set.



466
467
468
469
470
471
472
473
474
475
476
 
# File 'ext/neuro.c', line 466

static VALUE rb_network_max_iterations_set(VALUE self, VALUE iterations)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    Check_Type(iterations, T_FIXNUM);
    network->max_iterations = NUM2INT(iterations);
    if (network->max_iterations <= 0)
        network->max_iterations = DEFAULT_MAX_ITERATIONS;
    return iterations;
}

#output_size ⇒ Object

Returns the output_size of this Network as an Integer. This is the number of nodes in the output layer.



370
371
372
373
374
375
376
 
# File 'ext/neuro.c', line 370

static VALUE rb_network_output_size(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return INT2NUM(network->output_size);
}

#to_h ⇒ Object

Returns the state of the network as a Hash.



481
482
483
484
485
486
487
 
# File 'ext/neuro.c', line 481

static VALUE rb_network_to_h(VALUE self)
{
    Network *network;

    Data_Get_Struct(self, Network, network);
    return Network_to_hash(network);
}

#to_s ⇒ Object

Returns a short string for the network.



493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
 
# File 'ext/neuro.c', line 493

static VALUE rb_network_to_s(VALUE self)
{
    Network *network;
    VALUE argv[5];
    int argc = 5;

    Data_Get_Struct(self, Network, network);
    argv[0] = rb_str_new2("#<%s:%u,%u,%u>");
    argv[1] = rb_funcall(self, id_class, 0, 0);
    argv[1] = rb_funcall(argv[1], id_name, 0, 0);
    argv[2] = INT2NUM(network->input_size);
    argv[3] = INT2NUM(network->hidden_size);
    argv[4] = INT2NUM(network->output_size);
    return rb_f_sprintf(argc, argv);
}