Derive each parameter's reference limits from the study's results

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

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

1# Reference solution for the yamaa benchmark adam-adlb-reference-limits (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 1.67lb <- read.csv(8  "/app/input/lb.csv",9  colClasses = c(LBSTRESN = "numeric"),10  na.strings = "",11  stringsAsFactors = FALSE12)1314# Each parameter's 95% reference range from the study's results: the mean15# less and plus the standard normal 0.975 quantile times the sample standard16# deviation, a test not done left out.17limit <- function(values, sign) {18  mean(values, na.rm = TRUE) + sign * qnorm(0.975) * sd(values, na.rm = TRUE)19}2021adlb <- data.frame(22  STUDYID = lb$STUDYID,23  USUBJID = lb$USUBJID,24  PARAMCD = lb$LBTESTCD,25  AVAL = lb$LBSTRESN,26  ANRLO = ave(lb$LBSTRESN, lb$LBTESTCD, FUN = function(x) limit(x, -1)),27  ANRHI = ave(lb$LBSTRESN, lb$LBTESTCD, FUN = function(x) limit(x, 1))28)2930dir.create("/app/output", showWarnings = FALSE)31write.csv(adlb, "/app/output/adlb.csv", row.names = FALSE, na = "")

1# Reference solution for the yamaa benchmark adam-adlb-reference-limits (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 csv8import statistics910LB_CSV = "/app/input/lb.csv"11OUTPUT_CSV = "/app/output/adlb.csv"121314def number(text):15    text = text.strip()16    return float(text) if text else None171819def cell(value):20    return "" if value is None else repr(value)212223def main():24    with open(LB_CSV, newline="") as f:25        lb = list(csv.DictReader(f))2627    # Each parameter's results, a test not done left out.28    results = {}29    for row in lb:30        value = number(row["LBSTRESN"])31        if value is not None:32            results.setdefault(row["LBTESTCD"], []).append(value)3334    z = statistics.NormalDist().inv_cdf(0.975)35    limits = {}36    for test, values in results.items():37        mean = statistics.fmean(values)38        sd = statistics.stdev(values)39        limits[test] = (mean - z * sd, mean + z * sd)4041    fieldnames = ["STUDYID", "USUBJID", "PARAMCD", "AVAL", "ANRLO", "ANRHI"]42    with open(OUTPUT_CSV, "w", newline="") as f:43        writer = csv.DictWriter(f, fieldnames=fieldnames)44        writer.writeheader()45        for row in lb:46            low, high = limits.get(row["LBTESTCD"], (None, None))47            writer.writerow(48                {49                    "STUDYID": row["STUDYID"],50                    "USUBJID": row["USUBJID"],51                    "PARAMCD": row["LBTESTCD"],52                    "AVAL": cell(number(row["LBSTRESN"])),53                    "ANRLO": cell(low),54                    "ANRHI": cell(high),55                }56            )575859if __name__ == "__main__":60    main()

1/* opensas reference solution for the yamaa benchmark adam-adlb-reference-limits.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data lb;6  length STUDYID DOMAIN USUBJID LBTESTCD LBTEST LBSTRESU LBSTAT $40;7  infile "/app/input/lb.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char40. DOMAIN : $char40. USUBJID : $char40. LBSEQ9        LBTESTCD : $char40. LBTEST : $char40. LBSTRESN LBSTRESU : $char40.10        LBSTAT : $char40.;11run;1213/* The results of each parameter; a test not done has none and takes no14   part. */15data results;16  set lb;17  if not missing(LBSTRESN);18run;1920/* Each parameter's mean and sample standard deviation (divisor n - 1), and21   its limits at the mean less and plus the standard normal 0.975 quantile,22   probit(0.975) = 1.959963984540054, times that deviation. */23proc sql;24  create table means as25    select LBTESTCD, mean(LBSTRESN) as M, count(*) as N26    from results group by LBTESTCD;27  create table squares as28    select r.LBTESTCD, m.M, m.N, (r.LBSTRESN - m.M) * (r.LBSTRESN - m.M) as SQ29    from results as r inner join means as m on m.LBTESTCD = r.LBTESTCD;30  create table limits as31    select LBTESTCD,32           M - 1.959963984540054 * sqrt(sum(SQ) / (N - 1)) as ANRLO,33           M + 1.959963984540054 * sqrt(sum(SQ) / (N - 1)) as ANRHI34    from squares group by LBTESTCD, M, N;35  create table final as36    select l.STUDYID, l.USUBJID, l.LBTESTCD as PARAMCD length=40,37           l.LBSTRESN as AVAL, s.ANRLO, s.ANRHI38    from lb as l left join limits as s on s.LBTESTCD = l.LBTESTCD39    order by 1, 2, 3;40quit;4142proc export data=final outfile="/app/output/adlb.csv" dbms=csv replace; run;