Feature/mct optim fix #173
Conversation
|
I checked the dis.tss results of the branch master and the current branch using cold.xml. Results after the commit "split and optimization of the new solve1pixel function " are bit identical. Results after "minor optimisations" are slightly different but I think they are still OK: 29.2018 -> 29.2017 |
|
Is 'minor optimisations' strictly necessary? I would very much prefer having bit identical results. |
|
I think we can maybe do something in between that allows bit reproducibility. I think moving around some parts of the code (counts, upstream inflow, etc.) helped but maybe not the removal of closureError and the change in the while loop, which is what impacts may the bit reproducibility. |
| num_upstream_pixels, a_dx_div_dt[pix], b_a_dx_div_dt[pix], beta, inv_beta, b_minus_1): | ||
| solve1PixelAvg(pix, discharge_avg, discharge_before, discharge[pix], lateral_inflow[pix], upstream_lookup,\ | ||
| num_upstream_pixels, inv_time_delta, beta, a_dx[pix]) | ||
|
|
There was a problem hiding this comment.
Please @ecCinziaMazzetti and @StefaniaGrimaldi just check the changes here, as I think we were using the whole arrays previously in the computation, while only the "pix" element of the a_dx, constant, lateral_inflow, a_dx_div_dt and b_a_dx_div_dt arrays were needed in the solve1pixel functions, but I can be wrong.
| discharge[pix] = 0 | ||
| return | ||
| return False |
There was a problem hiding this comment.
@ecCinziaMazzetti @StefaniaGrimaldi Here we skip the computation of the discharge_avg value, but discharge[pix] is changed, as we set it to zero. Is that correct? Should we still need to compute channel_volume_start and channel_volume_end for dischage_avg as the quantity of discharge[pix] has changed?
No description provided.