autoCategories FFNet_Pattern_to_Categories (FFNet me, Pattern thee, int labeling) {
try {
if (! my outputCategories) {
Melder_throw (U"The FFNet has no output categories.");
}
if (my nInputs != thy nx) {
Melder_throw (U"The number of colums in the Pattern (", thy nx, U") should equal the number of inputs in the FFNet (", my nInputs, U").");
}
if (! _Pattern_checkElements (thee)) {
Melder_throw (U"All Pattern elements must be in the interval [0, 1].\nYou could use \"Formula...\" to scale the Pattern values first.");
}
autoCategories him = Categories_create ();
for (long k = 1; k <= thy ny; k++) {
FFNet_propagate (me, thy z[k], nullptr);
long index = FFNet_getWinningUnit (me, labeling);
autoDaata item = Data_copy ((Daata) my outputCategories -> item[index]);
Collection_addItem_move (him.peek(), item.move());
}
return him;
} catch (MelderError) {
Melder_throw (me, U": no Categories created.");
}
}
Categories FFNet_Pattern_to_Categories (FFNet me, Pattern thee, int labeling) {
try {
if (my outputCategories == 0) {
Melder_throw (U"The FFNet has no output categories.");
}
if (my nInputs != thy nx) Melder_throw (U"The number of colums in the Pattern (", thy nx,
U") should equal the number of inputs in the FFNet (", my nInputs, U").");
if (! _Pattern_checkElements (thee)) Melder_throw
(U"The elements in the Pattern are not all in the interval [0, 1].\n"
U"The input of the neural net can only process values that are between 0 and 1.\n"
U"You could use \"Formula...\" to scale the Pattern values first.");
autoCategories him = Categories_create ();
for (long k = 1; k <= thy ny; k++) {
FFNet_propagate (me, thy z[k], 0);
long index = FFNet_getWinningUnit (me, labeling);
autoDaata item = Data_copy ( (Daata) my outputCategories -> item[index]);
Collection_addItem (him.peek(), item.transfer());
}
return him.transfer();
} catch (MelderError) {
Melder_throw (me, U": no Categories created.");
}
}
Categories FFNet_Activation_to_Categories (FFNet me, Activation activation, int labeling)
{
Categories thee = NULL, categories = my outputCategories;
long i, (*labelingFunction) (I, const double act[]);
if (! my outputCategories)
{
(void) Melder_error1 (L"FFNet & Activation: To Categories\n");
return Melder_errorp1 (L"The neural net has no Categories (has the FFNet been trained yet?).");
}
if (my nOutputs != activation->nx)
{
(void) Melder_error1 (L"FFNet & Activation: To Categories\n");
return Melder_errorp1 (L"The number of columns in the Activation must equal the number of outputs of FFNet.");
}
thee = Categories_create ();
if (thee == NULL) return NULL;
labelingFunction = labeling == 2 ? stochastic : winnerTakesAll;
for (i = 1; i <= activation->ny; i++)
{
long index = labelingFunction (me, activation->z[i]);
Data item = Data_copy (categories->item[index]);
if (item == NULL || ! Collection_addItem (thee, item))
{
forget (thee);
return Melder_errorp3 (L"FFNet & Activation: To Categories\n\nError creating label ",
Melder_integer (i), L".");
}
}
return thee;
}
static void CategoriesEditorCommand_init (CategoriesEditorCommand me, const char32 *name, Thing boss,
Command_Callback execute, Command_Callback undo, int /*nCategories*/, int nSelected) {
my nSelected = nSelected;
Command_init (me, name, boss, execute, undo);
my categories = Categories_create();
my selection = NUMvector<long> (1, nSelected);
}
autoCategories ResultsMFC_to_Categories_responses (ResultsMFC me) {
try {
autoCategories thee = Categories_create ();
for (long trial = 1; trial <= my numberOfTrials; trial ++) {
autoSimpleString category = SimpleString_create (my result [trial]. response);
thy addItem_move (category.move());
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": responses not converted to Categories.");
}
}
Categories ResultsMFC_to_Categories_responses (ResultsMFC me) {
try {
autoCategories thee = Categories_create ();
for (long trial = 1; trial <= my numberOfTrials; trial ++) {
autoSimpleString category = SimpleString_create (my result [trial]. response);
Collection_addItem (thee.peek(), category.transfer());
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": responses not converted to Categories.");
}
}
autoCategories OrderedOfString_to_Categories (OrderedOfString me) {
try {
autoCategories thee = Categories_create();
for (long i = 1; i <= my size; i++) {
autoSimpleString item = Data_copy ( (SimpleString) my item [i]);
Collection_addItem_move (thee.peek(), item.move());
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": not converted to Categories.");
}
}
autoCategories TableOfReal_to_CategoriesColumn (TableOfReal me) {
try {
autoCategories thee = Categories_create ();
for (long i = 1; i <= my numberOfColumns; i++) {
if (my columnLabels[i]) {
autoSimpleString s = SimpleString_create (my columnLabels[i]);
Collection_addItem_move (thee.peek(), s.move());
}
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": columnlabels not converted to Categories.");
}
}
Categories ClassificationTable_to_Categories_maximumProbability (ClassificationTable me) {
try {
autoCategories thee = Categories_create ();
for (long i = 1; i <= my numberOfRows; i++) {
double max = my data[i][1];
long col = 1;
for (long j = 2; j <= my numberOfColumns; j++) {
if (my data[i][j] > max) {
max = my data[i][j]; col = j;
}
}
OrderedOfString_append (thee.peek(), my columnLabels[col]);
}
return thee.transfer();
} catch (MelderError) {
Melder_throw (me, U": no Categories created.");
}
}
autoCategories FFNet_Activation_to_Categories (FFNet me, Activation activation, int labeling) {
try {
long (*labelingFunction) (FFNet me, const double act[]);
if (! my outputCategories) {
Melder_throw (U"No Categories (has the FFNet been trained yet?).");
}
if (my nOutputs != activation -> nx) {
Melder_throw (U"Number of columns and number of outputs must be equal.");
}
autoCategories thee = Categories_create ();
labelingFunction = labeling == 2 ? stochastic : winnerTakesAll;
for (long i = 1; i <= activation->ny; i++) {
long index = labelingFunction (me, activation -> z[i]);
autoSimpleString item = Data_copy (my outputCategories->at [index]);
thy addItem_move (item.move());
}
return thee;
} catch (MelderError) {
Melder_throw (me, U": no Categories created.");
}
}
WORD (L"Name", L"1x1")
NATURAL (L"Dimension of a pattern", L"1")
NATURAL (L"Number of patterns", L"1")
OK
DO
praat_new (Pattern_create (GET_INTEGER (L"Number of patterns"),
GET_INTEGER (L"Dimension of a pattern")), GET_STRING (L"Name"));
END
/**************** New Categories ***************************/
FORM (Categories_create, L"Create Categories", L"")
WORD (L"Name", L"empty")
OK
DO
praat_new (Categories_create (), GET_STRING (L"Name"));
END
DIRECT (FFNet_help) Melder_help (L"Feedforward neural networks"); END
DIRECT (FFNet_getMinimum)
LOOP {
iam (FFNet);
Melder_information (Melder_double (FFNet_getMinimum (me)));
}
END
FORM (FFNet_reset, L"FFNet: Reset", L"FFNet: Reset...")
LABEL (L"", L"Warning: this command destroys all previous learning.")
LABEL (L"", L"New weights will be randomly chosen from the interval [-range, +range].")
POSITIVE (L"Range", L"0.1")
