// destructive simplification procedure, prunes subtrees with too low counts
long long int simplify_subtree(int pos, int divisor, int min_size, int indent) {
PropertyDecisionNode &n = inner_node[pos];
if (n.property == -1) {
for (int i=0;i<indent;i++) v_printf(10," ");
v_printf(10,"* leaf: count=%lli, size=%llu bits, bits per int: %f\n", (long long int)leaf_node[n.leafID].count, (unsigned long long int)leaf_node[n.leafID].realSize/5461, (leaf_node[n.leafID].count > 0 ? leaf_node[n.leafID].realSize/leaf_node[n.leafID].count*1.0/5461 : -1));
if (leaf_node[n.leafID].count == 0) return -100; // avoid empty leafs by giving them an extra penalty
return leaf_node[n.leafID].count;
} else {
for (int i=0;i<indent;i++) v_printf(10," ");
v_printf(10,"* test: property %i, value > %i ? (after %lli steps)\n", n.property, n.splitval, (long long int)n.count);
long long int subtree_size = 0;
subtree_size += simplify_subtree(n.childID, divisor, min_size, indent+1);
subtree_size += simplify_subtree(n.childID+1, divisor, min_size, indent+1);
n.count /= divisor;
if (n.count > CONTEXT_TREE_MAX_COUNT) {
n.count = CONTEXT_TREE_MAX_COUNT;
}
if (n.count < CONTEXT_TREE_MIN_COUNT) n.count=CONTEXT_TREE_MIN_COUNT;
if (n.count > 0xf) n.count &= 0xfff8; // remove some lsb entropy
// printf("%li COUNT\n",n.count);
if (subtree_size < min_size) {
// printf(" PRUNING THE ABOVE SUBTREE\n");
n.property = -1; // procedure is destructive because the leafID is not set
}
return subtree_size;
}
}
// destructive simplification procedure, prunes subtrees with too low counts
long long int simplify_subtree(int pos, int divisor, int min_size) {
PropertyDecisionNode &n = inner_node[pos];
if (n.property == -1) {
// printf("* leaf %i : count=%lli, size=%llu bits, bits per int: %f\n", n.leafID, (long long int)leaf_node[n.leafID].count, (unsigned long long int)leaf_node[n.leafID].realSize/5461, (leaf_node[n.leafID].count > 0 ? leaf_node[n.leafID].realSize/leaf_node[n.leafID].count*1.0/5461 : -1));
if (leaf_node[n.leafID].count == 0) return -100; // avoid empty leafs by giving them an extra penalty
return leaf_node[n.leafID].count;
} else {
// printf("* split on prop %i at val %i after %lli steps\n", n.property, n.splitval, (long long int)n.count);
// printf("* split on prop %i\n", n.property);
long long int subtree_size = 0;
subtree_size += simplify_subtree(n.childID, divisor, min_size);
subtree_size += simplify_subtree(n.childID+1, divisor, min_size);
n.count /= divisor;
if (n.count > CONTEXT_TREE_MAX_COUNT) {
n.count = CONTEXT_TREE_MAX_COUNT;
}
if (n.count < CONTEXT_TREE_MIN_COUNT) n.count=CONTEXT_TREE_MIN_COUNT;
// printf("%li COUNT\n",n.count);
if (subtree_size < min_size) {
// printf(" PRUNING THE ABOVE SUBTREE\n");
n.property = -1; // procedure is destructive because the leafID is not set
}
return subtree_size;
}
}
void simplify(int divisor=CONTEXT_TREE_COUNT_DIV, int min_size=CONTEXT_TREE_MIN_SUBTREE_SIZE) {
v_printf(10,"TREE BEFORE SIMPLIFICATION:\n");
simplify_subtree(0, divisor, min_size, 0);
}
void simplify(int divisor=CONTEXT_TREE_COUNT_DIV, int min_size=CONTEXT_TREE_MIN_SUBTREE_SIZE) {
simplify_subtree(0, divisor, min_size);
}
void simplify(int divisor=CONTEXT_TREE_COUNT_DIV, int min_size=CONTEXT_TREE_MIN_SUBTREE_SIZE, int plane=0) {
v_printf(10,"PLANE %i: TREE BEFORE SIMPLIFICATION:\n",plane);
simplify_subtree(0, divisor, min_size, 0, plane);
}