Visit Study Day

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

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

1# Reference solution for the yamaa benchmark sdtm-vs-study-day (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(lubridate)9library(purrr)10library(readr)1112study_day <- function(ref, coll) {13  if (is.na(ref) || is.na(coll)) {14    return(NA_integer_)15  }16  delta <- as.integer(coll - ref)17  if (delta >= 0L) delta + 1L else delta18}1920dm <- read_csv(21  "/app/input/dm.csv",22  col_types = cols(.default = col_character())23)24se <- read_csv(25  "/app/input/se.csv",26  col_types = cols(.default = col_character())27)28tv <- read_csv(29  "/app/input/tv.csv",30  col_types = cols(.default = col_character())31)32vs_raw <- read_csv(33  "/app/input/vs_raw.csv",34  col_types = cols(VSSEQ = col_integer(), .default = col_character())35)3637tv_map <- setNames(suppressWarnings(as.numeric(tv$VISITNUM)), tv$VISIT)38ref_map <- setNames(suppressWarnings(ymd(dm$RFSTDTC)), dm$USUBJID)3940vs <- vs_raw |>41  mutate(42    coll = suppressWarnings(ymd(VSDTC)),43    ref = ref_map[USUBJID],44    VSDY = map2_int(ref, coll, study_day, .progress = FALSE),45    DOMAIN = "VS"46  )47vs$VISITNUM <- pmap_dbl(48  list(vs$USUBJID, vs$VISIT, vs$VSDTC),49  function(usubjid, visit, vsdtc) {50    if (visit %in% names(tv_map)) {51      return(tv_map[[visit]])52    }53    day <- suppressWarnings(ymd(vsdtc))54    if (is.na(day)) {55      return(NA_real_)56    }57    prior <- vs_raw |>58      filter(59        USUBJID == usubjid,60        VISIT %in% names(tv_map),61        !is.na(suppressWarnings(ymd(VSDTC))),62        suppressWarnings(ymd(VSDTC)) < day63      )64    if (nrow(prior) == 0) {65      return(NA_real_)66    }67    max(tv_map[prior$VISIT], na.rm = TRUE) + 0.0168  },69  .progress = FALSE70)71vs$EPOCH <- pmap_chr(72  list(vs$USUBJID, vs$VSDTC),73  function(usubjid, vsdtc) {74    day <- suppressWarnings(ymd(vsdtc))75    if (is.na(day)) {76      return(NA_character_)77    }78    elements <- se |> filter(USUBJID == usubjid)79    hits <- elements |>80      filter(81        suppressWarnings(ymd(SESTDTC)) <= day,82        day <= suppressWarnings(ymd(SEENDTC))83      ) |>84      arrange(suppressWarnings(ymd(SESTDTC)))85    if (nrow(hits) == 0) {86      return(NA_character_)87    }88    hits$EPOCH[nrow(hits)]89  },90  .progress = FALSE91)92vs <- vs |>93  arrange(USUBJID, VSSEQ) |>94  select(95    DOMAIN, STUDYID, USUBJID, VSSEQ, VSTESTCD, VSORRES, VSDTC, VISIT,96    VISITNUM, EPOCH, VSDY97  )9899dir.create("/app/output", showWarnings = FALSE)100write_csv(vs, "/app/output/vs.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-vs-study-day (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 datetime import date910INPUT_DIR = "/app/input"11OUTPUT_DIR = "/app/output"121314def read_csv(path):15    with open(path, newline="", encoding="utf-8") as f:16        return list(csv.DictReader(f))171819def parse_date(value):20    value = (value or "").strip()21    if not value:22        return None23    return date.fromisoformat(value)242526def format_number(value):27    if value is None:28        return ""29    if float(value).is_integer():30        return str(int(value))31    return f"{value:.2f}".rstrip("0").rstrip(".")323334def main():35    dm = read_csv(f"{INPUT_DIR}/dm.csv")36    se = read_csv(f"{INPUT_DIR}/se.csv")37    tv = read_csv(f"{INPUT_DIR}/tv.csv")38    vs_raw = read_csv(f"{INPUT_DIR}/vs_raw.csv")3940    ref_start = {}41    for row in dm:42        ref_start[row["USUBJID"]] = parse_date(row["RFSTDTC"])4344    elements = {}45    for row in se:46        elements.setdefault(row["USUBJID"], []).append(47            (row["EPOCH"], parse_date(row["SESTDTC"]), parse_date(row["SEENDTC"]))48        )4950    planned = {}51    for row in tv:52        planned[row["VISIT"]] = float(row["VISITNUM"])5354    def unplanned_visitnum(subject, coll_date):55        # Latest planned visit number collected for the subject before56        # the record's collection date, plus 0.01.57        if coll_date is None:58            return None59        best = None60        for row in vs_raw:61            if row["USUBJID"] != subject or row["VISIT"] not in planned:62                continue63            other_date = parse_date(row["VSDTC"])64            if other_date is not None and other_date < coll_date:65                num = planned[row["VISIT"]]66                if best is None or num > best:67                    best = num68        return None if best is None else best + 0.016970    out_rows = []71    for row in vs_raw:72        subject = row["USUBJID"]73        coll_date = parse_date(row["VSDTC"])7475        if row["VISIT"] in planned:76            visitnum = planned[row["VISIT"]]77        else:78            visitnum = unplanned_visitnum(subject, coll_date)7980        epoch = ""81        if coll_date is not None:82            chosen = None83            for name, start, end in elements.get(subject, []):84                if start is not None and end is not None and start <= coll_date <= end:85                    # A day shared by two elements belongs to the later one.86                    if chosen is None or start >= chosen[1]:87                        chosen = (name, start)88            if chosen is not None:89                epoch = chosen[0]9091        vsdy = ""92        ref = ref_start.get(subject)93        if coll_date is not None and ref is not None:94            diff = (coll_date - ref).days95            vsdy = str(diff + 1) if diff >= 0 else str(diff)9697        out_rows.append({98            "DOMAIN": "VS",99            "STUDYID": row["STUDYID"],100            "USUBJID": subject,101            "VSSEQ": row["VSSEQ"],102            "VSTESTCD": row["VSTESTCD"],103            "VSORRES": row["VSORRES"],104            "VSDTC": row["VSDTC"],105            "VISIT": row["VISIT"],106            "VISITNUM": format_number(visitnum),107            "EPOCH": epoch,108            "VSDY": vsdy,109        })110111    columns = ["DOMAIN", "STUDYID", "USUBJID", "VSSEQ", "VSTESTCD", "VSORRES",112               "VSDTC", "VISIT", "VISITNUM", "EPOCH", "VSDY"]113    with open(f"{OUTPUT_DIR}/vs.csv", "w", newline="", encoding="utf-8") as f:114        writer = csv.DictWriter(f, fieldnames=columns)115        writer.writeheader()116        writer.writerows(out_rows)117118119if __name__ == "__main__":120    main()

1/* opensas reference solution for the yamaa benchmark sdtm-vs-study-day.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data dm;6  length STUDYID USUBJID RFSTDTC $1024;7  infile "/app/input/dm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. RFSTDTC : $char1024.;9run;1011data se;12  length STUDYID USUBJID EPOCH SESTDTC SEENDTC $1024;13  infile "/app/input/se.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. EPOCH : $char1024. SESTDTC : $char1024. SEENDTC : $char1024.;15run;1617data tv;18  length VISIT $1024;19  infile "/app/input/tv.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;20  input VISIT : $char1024. VISITNUM;21run;2223data vs_raw;24  length STUDYID USUBJID VISIT VSTESTCD VSORRES VSDTC $1024;25  infile "/app/input/vs_raw.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;26  input STUDYID : $char1024. USUBJID : $char1024. VSSEQ VISIT : $char1024. VSTESTCD : $char1024. VSORRES : $char1024. VSDTC : $char1024.;27run;2829proc sql; create table collected as select a.*,b.VISITNUM,c.RFSTDTC from vs_raw a left join tv b on a.VISIT=b.VISIT left join dm c on a.STUDYID=c.STUDYID and a.USUBJID=c.USUBJID; quit;30data dated; set collected; NODATE=missing(VSDTC); run;31proc sort data=dated; by STUDYID USUBJID NODATE VSDTC VSSEQ; run;32data numbered; set dated; by STUDYID USUBJID; retain PRIORNUM; if first.USUBJID then PRIORNUM=.; if not missing(VISITNUM) and not missing(VSDTC) then PRIORNUM=VISITNUM; else if missing(VISITNUM) and not missing(VSDTC) and not missing(PRIORNUM) then VISITNUM=PRIORNUM+0.01; run;33proc sql; create table candidates as select a.*,b.EPOCH,b.SESTDTC from numbered a left join se b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VSDTC>=b.SESTDTC and a.VSDTC<=b.SEENDTC and not missing(a.VSDTC); quit;34proc sort data=candidates; by STUDYID USUBJID VSSEQ descending SESTDTC; run;35data result; length DOMAIN $2; set candidates; by STUDYID USUBJID VSSEQ; if first.VSSEQ; DOMAIN='VS'; VSDY=.; if lengthn(VSDTC)>=10 and lengthn(RFSTDTC)>=10 then do; VSDY=input(substr(VSDTC,1,10),yymmdd10.)-input(substr(RFSTDTC,1,10),yymmdd10.); if VSDY>=0 then VSDY+1; end; run;36proc sql;37  create table final as select DOMAIN, STUDYID, USUBJID, VSSEQ, VSTESTCD, VSORRES, VSDTC, VISIT, VISITNUM, EPOCH, VSDY from result;38quit;39proc export data=final outfile="/app/output/vs.csv" dbms=csv replace; run;