Multi-Scale Instrument Scoring

1library(yamaa)23yamaa_spec("spec.yaml")$run()$output

1import yamaa23yamaa.spec("spec.yaml").run().output

1# Reference solution for the yamaa benchmark adam-adqs-multi-scale-scoring (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(purrr, warn.conflicts = FALSE)9library(readr)1011qs <- read_csv(12  "/app/input/qs.csv",13  col_types = cols(14    QSSEQ = col_integer(),15    QSSTRESN = col_double(),16    .default = col_character()17  )18)1920# Each item response, on its own answer scale.21items <- qs |>22  mutate(PARAMCD = QSTESTCD, PARAM = QSTEST, AVAL = QSSTRESN) |>23  select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL)2425# The four scales: member items, anchor item, minimum answered items,26# direction (down = higher score means better, up = higher means worse),27# and the answer range.28scales <- tribble(29  ~pcode, ~pname, ~members, ~anchor, ~min_ans, ~direction, ~rng,30  "F1SCORE", "Physical Functioning Scale Score",31  list(c("F101", "F102", "F103", "F104")), "F101", 2L, "down", 3,32  "F2SCORE", "Role Functioning Scale Score",33  list(c("F201", "F202")), "F201", 1L, "down", 3,34  "SSCORE", "Fatigue and Sleep Symptom Scale Score",35  list(c("S01", "S02", "F104")), "S01", 2L, "up", 3,36  "GSCORE", "Global Health Scale Score",37  list(c("G01", "G02")), "G01", 1L, "up", 638)3940score_one <- function(pcode, pname, members, anchor, min_ans, direction, rng) {41  # `members` arrives as a one-element list from the tribble's list-column;42  # `%in%` against a list never matches, so unwrap it first.43  members <- unlist(members)44  member_rows <- items |> filter(PARAMCD %in% members)45  visits <- items |>46    filter(PARAMCD %in% anchor) |>47    distinct(STUDYID, USUBJID, VISIT)48  stats <- member_rows |>49    group_by(STUDYID, USUBJID, VISIT) |>50    summarise(51      n_answered = sum(!is.na(AVAL)),52      raw_mean = mean(AVAL, na.rm = TRUE),53      .groups = "drop"54    )55  visits |>56    left_join(stats, by = c("STUDYID", "USUBJID", "VISIT")) |>57    mutate(58      AVAL = dplyr::case_when(59        is.na(n_answered) | n_answered < min_ans ~ NA_real_,60        direction == "down" ~ 100 * (1 - (raw_mean - 1) / rng),61        TRUE ~ 100 * (raw_mean - 1) / rng62      ),63      PARAMCD = pcode,64      PARAM = pname65    ) |>66    select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL)67}6869scores <- pmap_dfr(scales, score_one)7071adqs <- bind_rows(items, scores) |>72  arrange(STUDYID, USUBJID, VISIT, PARAMCD)7374dir.create("/app/output", showWarnings = FALSE)75write_csv(adqs, "/app/output/adqs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adqs-multi-scale-scoring (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 1.67import csv8from collections import OrderedDict910INPUT = "/app/input"11OUTPUT = "/app/output"1213# scale: (score PARAMCD, PARAM, items, anchor, min answered, rescale fn)14SCALES = [15    ("F1SCORE", "Physical Functioning Scale Score",16     ["F101", "F102", "F103", "F104"], "F101", 2,17     lambda m: (4.0 - m) / 3.0 * 100.0),18    ("F2SCORE", "Role Functioning Scale Score",19     ["F201", "F202"], "F201", 1,20     lambda m: (4.0 - m) / 3.0 * 100.0),21    ("SSCORE", "Fatigue and Sleep Symptom Scale Score",22     ["S01", "S02", "F104"], "S01", 2,23     lambda m: (m - 1.0) / 3.0 * 100.0),24    ("GSCORE", "Global Health Scale Score",25     ["G01", "G02"], "G01", 1,26     lambda m: (m - 1.0) / 6.0 * 100.0),27]282930def fmt(value):31    if value is None:32        return ""33    if float(value).is_integer():34        return str(int(value))35    return repr(float(value))363738def main():39    with open(f"{INPUT}/qs.csv", newline="") as f:40        rows = list(csv.DictReader(f))4142    groups = OrderedDict()  # (study, subj, visit) -> list of rows43    for r in rows:44        key = (r["STUDYID"], r["USUBJID"], r["VISIT"])45        groups.setdefault(key, []).append(r)4647    out = []48    for (study, subj, visit), grows in groups.items():49        for r in grows:50            aval = r["QSSTRESN"].strip()51            out.append([study, subj, visit, r["QSTESTCD"], r["QSTEST"],52                        aval if aval else ""])53        by_code = {}54        for r in grows:55            by_code.setdefault(r["QSTESTCD"], []).append(r)56        for pcd, pname, items, anchor, min_n, scale_fn in SCALES:57            if anchor not in by_code:58                continue59            answered = []60            for code in items:61                for r in by_code.get(code, []):62                    v = r["QSSTRESN"].strip()63                    if v:64                        answered.append(float(v))65            if len(answered) >= min_n:66                aval = fmt(scale_fn(sum(answered) / len(answered)))67            else:68                aval = ""69            out.append([study, subj, visit, pcd, pname, aval])7071    with open(f"{OUTPUT}/adqs.csv", "w", newline="") as f:72        w = csv.writer(f)73        w.writerow(["STUDYID", "USUBJID", "VISIT", "PARAMCD", "PARAM", "AVAL"])74        w.writerows(out)757677if __name__ == "__main__":78    main()

1/* opensas reference solution for the yamaa benchmark adam-adqs-multi-scale-scoring.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; create table collected as select STUDYID,USUBJID,VISIT,QSTESTCD as PARAMCD,QSTEST as PARAM,QSSTRESN as AVAL from qs; quit;12proc sql;13create table anchor0 as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='F101';14create table members0 as select * from collected where PARAMCD in ('F101','F102','F103','F104');15create table stats0 as select a.STUDYID,a.USUBJID,a.VISIT,count(b.AVAL) as NANSWER,mean(b.AVAL) as RAWMEAN from anchor0 a left join members0 b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT; quit;16data score0; set stats0; length PARAMCD PARAM $1024; PARAMCD='F1SCORE'; PARAM='Physical Functioning Scale Score'; AVAL=.; if NANSWER>=2 then AVAL=100*(1-(RAWMEAN-1)/3); run;17proc sql;18create table anchor1 as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='F201';19create table members1 as select * from collected where PARAMCD in ('F201','F202');20create table stats1 as select a.STUDYID,a.USUBJID,a.VISIT,count(b.AVAL) as NANSWER,mean(b.AVAL) as RAWMEAN from anchor1 a left join members1 b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT; quit;21data score1; set stats1; length PARAMCD PARAM $1024; PARAMCD='F2SCORE'; PARAM='Role Functioning Scale Score'; AVAL=.; if NANSWER>=1 then AVAL=100*(1-(RAWMEAN-1)/3); run;22proc sql;23create table anchor2 as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='S01';24create table members2 as select * from collected where PARAMCD in ('S01','S02','F104');25create table stats2 as select a.STUDYID,a.USUBJID,a.VISIT,count(b.AVAL) as NANSWER,mean(b.AVAL) as RAWMEAN from anchor2 a left join members2 b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT; quit;26data score2; set stats2; length PARAMCD PARAM $1024; PARAMCD='SSCORE'; PARAM='Fatigue and Sleep Symptom Scale Score'; AVAL=.; if NANSWER>=2 then AVAL=100*(RAWMEAN-1)/3; run;27proc sql;28create table anchor3 as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='G01';29create table members3 as select * from collected where PARAMCD in ('G01','G02');30create table stats3 as select a.STUDYID,a.USUBJID,a.VISIT,count(b.AVAL) as NANSWER,mean(b.AVAL) as RAWMEAN from anchor3 a left join members3 b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT; quit;31data score3; set stats3; length PARAMCD PARAM $1024; PARAMCD='GSCORE'; PARAM='Global Health Scale Score'; AVAL=.; if NANSWER>=1 then AVAL=100*(RAWMEAN-1)/6; run;32data result; length PARAMCD PARAM $1024; set collected score0 score1 score2 score3; run;33proc sql;34  create table final as select STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL from result;35quit;36proc export data=final outfile="/app/output/adqs.csv" dbms=csv replace; run;