I am wondering if it would be worth it to cache the intermediate results in decision tree estimators to avoid that the sum_total and weighted_n_node_samples attributes are re-computed.
In particular, these for loops will not be needed:
|
for p in range(start, end): |
|
i = samples[p] |
|
|
|
# w is originally set to be 1.0, meaning that if no sample weights |
|
# are given, the default weight of each sample is 1.0 |
|
if sample_weight != NULL: |
|
w = sample_weight[i] |
|
|
|
# Count weighted class frequency for each target |
|
for k in range(self.n_outputs): |
|
c = <SIZE_t> self.y[i, k] |
|
sum_total[k * self.sum_stride + c] += w |
|
|
|
self.weighted_n_node_samples += w |
and:
|
for p in range(start, end): |
|
i = samples[p] |
|
|
|
if sample_weight != NULL: |
|
w = sample_weight[i] |
|
|
|
for k in range(self.n_outputs): |
|
y_ik = self.y[i, k] |
|
w_y_ik = w * y_ik |
|
self.sum_total[k] += w_y_ik |
|
self.sq_sum_total += w_y_ik * y_ik |
|
|
|
self.weighted_n_node_samples += w |
and:
|
for p in range(start, end): |
|
i = samples[p] |
|
|
|
if sample_weight != NULL: |
|
w = sample_weight[i] |
|
|
|
for k in range(self.n_outputs): |
|
# push method ends up calling safe_realloc, hence `except -1` |
|
# push all values to the right side, |
|
# since pos = start initially anyway |
|
(<WeightedMedianCalculator> right_child[k]).push(self.y[i, k], w) |
|
|
|
self.weighted_n_node_samples += w |
It should be easy to cache these intermediate results in Stack and PriorityHeap, but I think we should discuss before addressing the issue.
