Skip to content

Conversation

castelao
Copy link
Member

We already reduced the size and optimized the memory usage related to input fields and cost calculation, but we still consume too much memory to run in a regular computer. This PR is an MVP to reduce the memory footprint on the searching process itself to accommodate in a regular computer memory.

The core idea here is on using an active searching cache and an archive to dump processed nodes that can be written to disk to be recovered only in the end of the process.

@castelao castelao self-assigned this Aug 12, 2025
@castelao castelao added the enhancement Update to logic or general code improvements label Aug 12, 2025
@castelao
Copy link
Member Author

For the records, this solution already allows cost of type f32. Until now, we only accepted integers, so the easiest solution was to scale the estimated cost and truncate, which reduced the resolution and limited the maximum total cost.

@castelao castelao requested a review from ppinchuk August 12, 2025 06:03
@castelao castelao added the p-high Priority: high label Aug 26, 2025
A custom Network for our needs. Assumes a regular grid and edges are
stored with the node, so that when pop a node, the associated nodes are
removed with it.
It was initially fixed to ArrayIndex, but we don't need to restrict to
that.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement Update to logic or general code improvements p-high Priority: high

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant