Subscale Scoring

1library(yamaa)23adqs <- yamaa_domain("spec.yaml")$output4adqs

1import yamaa23adqs = yamaa.yamaa_domain("spec.yaml").output4adqs

1# Reference solution for the yamaa benchmark adam-adqs-subscale-score (R track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67library(dplyr, warn.conflicts = FALSE)8library(readr)910qs <- read_csv(11  "/app/input/qs.csv",12  col_types = cols(13    QSSEQ = col_integer(),14    QSSTRESN = col_double(),15    .default = col_character()16  )17)1819# Each physical functioning item response, on its zero to four scale.20items <- qs |>21  filter(QSTESTCD %in% c("PF01", "PF02", "PF03", "PF04")) |>22  mutate(PARAMCD = QSTESTCD, PARAM = QSTEST, AVAL = QSSTRESN) |>23  select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL)2425# One score record per visit with a PF01 item record: the mean of the26# answered items on a zero to one hundred scale, with no value when fewer27# than three of the four items were answered.28visits <- items |>29  filter(PARAMCD %in% "PF01") |>30  distinct(STUDYID, USUBJID, VISIT)3132stats <- items |>33  group_by(STUDYID, USUBJID, VISIT) |>34  summarise(35    n_answered = sum(!is.na(AVAL)),36    mean_answered = mean(AVAL, na.rm = TRUE),37    .groups = "drop"38  ) |>39  mutate(40    mean_answered = if_else(41      n_answered < 3 | is.nan(mean_answered),42      NA_real_,43      mean_answered * 2544    )45  )4647scores <- visits |>48  left_join(stats, by = c("STUDYID", "USUBJID", "VISIT")) |>49  mutate(50    PARAMCD = "PFSCORE",51    PARAM = "Physical Functioning Subscale Score",52    AVAL = mean_answered53  ) |>54  select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL)5556adqs <- bind_rows(items, scores) |>57  arrange(STUDYID, USUBJID, VISIT, PARAMCD)5859dir.create("/app/output", showWarnings = FALSE)60write_csv(adqs, "/app/output/adqs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adqs-subscale-score (Python track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67from pathlib import Path89import polars as pl1011qs = pl.read_csv("/app/input/qs.csv", infer_schema=False).with_columns(12    pl.col("QSSEQ").cast(pl.Int64),13    pl.col("QSSTRESN").cast(pl.Float64),14)1516# Each physical functioning item response, on its zero to four scale.17items = (18    qs.filter(pl.col("QSTESTCD").is_in(["PF01", "PF02", "PF03", "PF04"]))19    .with_columns(20        PARAMCD=pl.col("QSTESTCD"), PARAM=pl.col("QSTEST"), AVAL=pl.col("QSSTRESN")21    )22    .select("STUDYID", "USUBJID", "VISIT", "PARAMCD", "PARAM", "AVAL")23)2425# One score record per visit with a PF01 item record: the mean of the26# answered items on a zero to one hundred scale, with no value when fewer27# than three of the four items were answered.28visits = items.filter(pl.col("PARAMCD") == "PF01").select(29    "STUDYID", "USUBJID", "VISIT"30).unique(maintain_order=True)3132stats = items.group_by(["STUDYID", "USUBJID", "VISIT"]).agg(33    n_answered=pl.col("AVAL").drop_nulls().len(),34    mean_answered=pl.col("AVAL").mean(),35)3637scores = (38    visits.join(stats, on=["STUDYID", "USUBJID", "VISIT"], how="left")39    .with_columns(40        AVAL=pl.when(pl.col("n_answered") >= 3)41        .then(pl.col("mean_answered") * 25)42        .otherwise(None)43    )44    .with_columns(45        PARAMCD=pl.lit("PFSCORE"),46        PARAM=pl.lit("Physical Functioning Subscale Score"),47    )48    .select("STUDYID", "USUBJID", "VISIT", "PARAMCD", "PARAM", "AVAL")49)5051adqs = pl.concat([items, scores], how="diagonal").sort(52    ["STUDYID", "USUBJID", "VISIT", "PARAMCD"]53)5455Path("/app/output").mkdir(exist_ok=True)56adqs.write_csv("/app/output/adqs.csv")

1/* opensas reference solution for the yamaa benchmark adam-adqs-subscale-score.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data qs;6  length STUDYID USUBJID VISIT QSCAT QSTESTCD QSTEST $1024;7  infile "/app/input/qs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. VISIT : $char1024. QSSEQ QSCAT : $char1024. QSTESTCD : $char1024. QSTEST : $char1024. QSSTRESN;9run;1011proc sql;12  create table collected as select STUDYID,USUBJID,VISIT,QSTESTCD as PARAMCD,QSTEST as PARAM,QSSTRESN as AVAL from qs where QSTESTCD in ('PF01','PF02','PF03','PF04');13  create table anchors as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='PF01';14  create table scores as select a.STUDYID,a.USUBJID,a.VISIT,'PFSCORE' as PARAMCD,'Physical Functioning Subscale Score' as PARAM,15    case when count(b.AVAL)>=3 then 25*mean(b.AVAL) else . end as AVAL from anchors as a left join collected as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT;16quit;17data result; set collected scores; run;1819proc sql;20  create table final as select STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL from result;21quit;22proc export data=final outfile="/app/output/adqs.csv" dbms=csv replace; run;