Ndr/cost model performance #111
Conversation
| CostModel { | ||
| ) -> Result<CostModel, CostError> { | ||
| // map the state variable coefficiencies and rates to the state variable indices | ||
| let mut state_variable_coefficients = vec![0.0; state_variable_indices.len()]; |
There was a problem hiding this comment.
oh, gotcha! this makes sense.
| value | default | meaning |
|---|---|---|
| vehicle cost rate | Raw | use the observed values directly |
| coefficients | 0.0 | if a coef is not specified for a state variable, ignore the variable in the cost model |
but we may want to add some error test here to protect against the case where someone forgets to populate a cost model in the TOML and the defaults result in zero-valued costs, like if all coefficients are zeroes at the end of this method -> return Err(_), and possibly the same for the rate values (if the user zeroes out all rate values with Factors).
There was a problem hiding this comment.
we may want to add some error test here to protect against the case where someone forgets to populate a cost model in the TOML and the defaults result in zero-valued costs
yeah that's a great point, I'll add that in
it did reduce the search runtime a bit but nothing to write home about. I'm hoping that there are some other areas we haven't touched that might result in even bigger gains.. |
Actually, when I tested this out on a national run it had more of an effect. Before these changes I got these numbers for running batches of long queries over the national graph: average route runtime: 137.328 seconds
average seconds per distance: 0.10788 seconds
average milliseconds per tree edge: 0.02720 millisecondsAnd after average route runtime: 33.852 seconds
average seconds per distance: 0.02219 seconds
average milliseconds per tree edge: 0.00545 milliseconds |
|
awesome! great stuff nick 👨🎤 🎸 |
Applies some minor modifications to the cost model to use iterators instead of allocating a vector for the cost model aggregation.
Also flattens the state variable coefficients, vehicle rates and network rates into a fixed size vector for index based lookup.