Repository navigation
Benchmark the authors' Irriblend-DSW optimiser, now MIT-licensed (#40) - #89
Merged
Merged
Conversation
Irriblend-DSW (Gallego-Elvira et al. 2021, doi:10.1016/j.agwat.2021.107012) had no licence, so the formulation track used only an LP after its published method. The authors added the MIT licence on 2026-10-20 (commit a046d5c). irriblend_adapter imports their unmodified irriblend_dsw_funs.py (SHA-256 checked) and calls its optimiser ferti_eco_min directly; the Colab notebook mounts Google Drive and cannot run elsewhere. Through the same calls it reproduces all 97 water blends of their two published case-study outputs exactly. The FlahaX catalogue is written in their 'ferts' convention (mmol/g, phosphate as H2PO4-, acid H+ as minus the anion); products the model cannot represent (micronutrients, chelates, urea, citrate) are left out with the reason. Neutral settings: equal prices, a non-binding EC cap, residual HCO3- 0. The engine runs as irriblend_dsw beside the method LP, which stays, labelled as FlahaX's own code. Every engine is also scored on macronutrients only, since Irriblend-DSW models none of the micronutrients.
…lend-DSW (#40) Measured at 1ade5b6 with the pinned competitor environment (Python 3.13.15, Windows 11), published with tools/publish_competitor_snapshot.py; track outputs kept byte for byte. The comparison page now covers the authors' Irriblend-DSW optimiser beside the LP after its method, with a macronutrient-only column for every engine and a section explaining its results on these inputs: 30 of 52 recipes returned (18 of the 22 rejections are formulas without an ammonium target, where its exact NH4 constraint excludes calcium nitrate), N, P, K and S met in every returned recipe, Mg or Ca overshoots as its lower bounds allow, and up to 708 mg/L chloride at equal prices. Every accuracy figure also in the 2026-10-21 snapshot is unchanged; timings were three to four times slower for every tool in this run, which the page states.
20 tasks done
rafatahmed
added a commit
that referenced
this pull request
Nov 7, 2026
Benchmark the authors' Irriblend-DSW optimiser, now MIT-licensed (#40)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-up to #40. The Irriblend-DSW authors added the MIT licence on 2026-10-20 (commit
a046d5c), so the comparison can now run their code, not only the LP after their method.What changed
tools/competitors/irriblend_adapter.py.code/irriblend_dsw_funs.py, after checking its SHA-256.ferti_eco_mindirectly. The notebook mounts Google Drive and writes to a fixed Colab path, so it cannot run elsewhere.irriblend_dsw. The LP after their method stays, labelled as FlahaX's own code. Every engine also gets a macronutrient-only score, because Irriblend-DSW models no micronutrients.2026-10-22-competitors-1ade5b6. Measured at the harness commit and published with the repo tool, with track outputs kept byte for byte. The competitor-comparison page is rewritten from it.Results (fixed benchmark inputs)
recommend, Cl ceilingThe page explains that Irriblend-DSW's low scores follow from its design, not a defect:
Every accuracy figure also in the 2026-10-21 snapshot is unchanged. Timings are three to four times slower for every tool in this run, PHREEQC included (machine load), which the page states.
Validation
FLAHAX_REQUIRE_PHREEQC=1unittest: 433 tests OKverify_distribution: PASSsphinx -W(fresh build): clean--check: okgit diff --checkok; no CRLF outside the byte-exact snapshotEmail to the authors
A draft results email to Dr Gallego-Elvira is in the owner's working copy (
outputs/irriblend-email-draft.md), for review. It is not sent.