inline void basic_examples()
{
std::ofstream output("output/basic_examples_output.txt");
cl::tape_serializer<double> serializer(output);
add_example(serializer);
mult_example(serializer);
pow_example(serializer);
exponent_example(serializer);
cos_sin_example(serializer);
}
/*
in :
- attribute is the attribute to compute the gain of
- examples is a tab of examples used to compute entropy
- n_ex is explicit
out :
- the gain ration value
*/
static double gain_ratio(const struct attribute_t *attribute, const struct example_t *examples, int n_ex)
{
double ret = 0.0;
struct example_t *ex_set = malloc(sizeof(*ex_set) * n_ex);
int i, j, index;
int index_attr = get_index_attribute(attribute->property, attributes_set, n_attr_set);
for(j = 0; j < attribute->l_tab; ++j)
{
index = 0;
/*ex_set = S_v*/
for(i = 0; i < n_ex; ++i)
{
if(strncmp(attribute->tab_values[j], HIGHER, sizeof(HIGHER)) == 0)
{
double down, up, v;
get_double_from_string(examples[i].tab_values[index_attr], &v);
sscanf(attribute->tab_values[j], "%*c %lf %*c%*c %lf", &down, &up);
if(down < v && (v > up || v - up < DBL_EPSILON))
add_example(ex_set, &index, &examples[i]);
}
else if(strncmp(attribute->tab_values[j], LESS_OR_EQUAL, sizeof(LESS_OR_EQUAL)) == 0)
{
double up, v;
get_double_from_string(examples[i].tab_values[index_attr], &v);
sscanf(attribute->tab_values[j], "%*c %lf", &up);
if(v > up || v - up < DBL_EPSILON)
add_example(ex_set, &index, &examples[i]);
}
else
{
if(strcmp(examples[i].tab_values[index_attr], attribute->tab_values[j]) == 0)
add_example(ex_set, &index, &examples[i]);
}
}
ret += entropy(ex_set, index)/log((double)index/n_ex);
}
free(ex_set);
return ret*(-log(2));
}
/*
in :
- examples is the set of examples where elements of the subset lie
- n_ex is explicit
- len_sebset is a pointer on the length of the subset
- index_attribute is the index of the attribute that is checked to create the subset
- value is the value of this attribute
out :
- a tab of examples
*/
static struct example_t *create_subset_ex_from_attr(const struct example_t *examples, int n_ex, int *len_subset, int index_attribute, string value)
{
int i;
void *tmp_ptr;
struct example_t *ret = NULL;
*len_subset = 0;
for(i = 0; i < n_ex; ++i)
if(strcmp(examples[i].tab_values[index_attribute], value) == 0)
{
tmp_ptr = realloc(ret, (*len_subset+1) * sizeof(*ret));
if(tmp_ptr == NULL)
leave_memory_error("create_subset_ex_from_attr");
ret = tmp_ptr;
add_example(ret, len_subset, &examples[i]);
}
return ret;
}
/*
in :
- attribute is the attribute to compute the gain of
- examples is the set of examples used to compute entropy
- n_ex is explicit
out :
- the (optimized) gain value
*/
static double gain(const struct attribute_t *attribute, const struct example_t *examples, int n_ex)
{
double ret = 0.0;
struct example_t *ex_set = malloc(sizeof(*ex_set) * n_ex);
int i, j, index;
int index_attr = get_index_attribute(attribute->property, attributes_set, n_attr_set);
for(j = 0; j < attribute->l_tab; ++j)
{
index = 0;
/*ex_set = S_v*/
for(i = 0; i < n_ex; ++i)
{
if(strcmp(examples[i].tab_values[index_attr], attribute->tab_values[j]) == 0)
add_example(ex_set, &index, &examples[i]);
}
ret += index*entropy(ex_set, index)/n_ex;
}
free(ex_set);
return ret;
}
