Visit Study Day

1library(yamaa)23vs <- yamaa_domain("spec.yaml")$output4vs

1import yamaa23vs = yamaa.yamaa_domain("spec.yaml").output4vs

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 1.67from datetime import date8from pathlib import Path910import polars as pl111213def parse_date(text: str | None) -> date | None:14    if text is None or text == "":15        return None16    try:17        # Collection and element dates are ISO dates without times.18        return date.fromisoformat(text[:10])19    except ValueError:20        return None212223def study_day(ref: date | None, coll: date | None) -> int | None:24    if ref is None or coll is None:25        return None26    delta = (coll - ref).days27    return delta + 1 if delta >= 0 else delta282930dm = pl.read_csv("/app/input/dm.csv", infer_schema=False)31se = pl.read_csv("/app/input/se.csv", infer_schema=False)32tv = pl.read_csv("/app/input/tv.csv", infer_schema=False)33vs_raw = pl.read_csv("/app/input/vs_raw.csv", infer_schema=False).with_columns(34    pl.col("VSSEQ").cast(pl.Int64)35)3637tv_map = {row["VISIT"]: row["VISITNUM"] for row in tv.to_dicts()}38ref_map = {39    row["USUBJID"]: parse_date(row["RFSTDTC"]) for row in dm.to_dicts()40}41se_by_subject: dict[str, list[dict]] = {}42for row in se.to_dicts():43    se_by_subject.setdefault(row["USUBJID"], []).append(44        {45            "EPOCH": row["EPOCH"],46            "start": parse_date(row["SESTDTC"]),47            "end": parse_date(row["SEENDTC"]),48        }49    )5051# Latest planned visit number collected before each date, per subject, for52# unplanned labels such as UNSCHEDULED.53planned_by_subject: dict[str, list[tuple[date, float]]] = {}54for row in vs_raw.to_dicts():55    if row["VISIT"] in tv_map and row["VSDTC"] not in (None, ""):56        coll = parse_date(row["VSDTC"])57        if coll is not None:58            planned_by_subject.setdefault(row["USUBJID"], []).append(59                (coll, float(tv_map[row["VISIT"]]))60            )6162rows = []63for row in vs_raw.to_dicts():64    usubjid = row["USUBJID"]65    coll = parse_date(row["VSDTC"])66    # Planned visit number, or the latest planned number before the67    # collection date plus .01 for an unplanned label.68    if row["VISIT"] in tv_map:69        visitnum: float | None = float(tv_map[row["VISIT"]])70    else:71        candidates = [72            num for day, num in planned_by_subject.get(usubjid, []) if day < coll73        ] if coll is not None else []74        visitnum = max(candidates) + 0.01 if candidates else None75    # Element containing the collection date; a shared day belongs to the76    # later element.77    epoch = None78    if coll is not None:79        hits = [80            e81            for e in se_by_subject.get(usubjid, [])82            if e["start"] is not None83            and e["end"] is not None84            and e["start"] <= coll <= e["end"]85        ]86        if hits:87            hits.sort(key=lambda e: e["start"])88            epoch = hits[-1]["EPOCH"]89    vsdy = study_day(ref_map.get(usubjid), coll)90    rows.append(91        {92            "DOMAIN": "VS",93            "STUDYID": row["STUDYID"],94            "USUBJID": usubjid,95            "VSSEQ": row["VSSEQ"],96            "VSTESTCD": row["VSTESTCD"],97            "VSORRES": row["VSORRES"],98            "VSDTC": row["VSDTC"],99            "VISIT": row["VISIT"],100            "VISITNUM": visitnum,101            "EPOCH": epoch,102            "VSDY": vsdy,103        }104    )105106vs = pl.DataFrame(rows).sort(["USUBJID", "VSSEQ"]).select(107    "DOMAIN",108    "STUDYID",109    "USUBJID",110    "VSSEQ",111    "VSTESTCD",112    "VSORRES",113    "VSDTC",114    "VISIT",115    "VISITNUM",116    "EPOCH",117    "VSDY",118)119120Path("/app/output").mkdir(exist_ok=True)121vs.write_csv("/app/output/vs.csv")

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;