Arm Assignment

1library(yamaa)23dm <- yamaa_domain("spec.yaml")$output4dm

1import yamaa23dm = yamaa.yamaa_domain("spec.yaml").output4dm

1# Reference solution for the yamaa benchmark sdtm-dm-arms (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)9library(stringr)1011rand <- read_csv(12  "/app/input/rand.csv",13  col_types = cols(.default = col_character())14)15ex <- read_csv(16  "/app/input/ex.csv",17  col_types = cols(.default = col_character())18)1920# The planned arm follows randomization; the actual arm follows exposure.21# PLACEBO was the PBO arm and VITAMIN D3 the TRT arm.22exposure <- ex |>23  filter(!is.na(EXTRT) & EXTRT != "") |>24  distinct(STUDYID, USUBJID, .keep_all = TRUE) |>25  select(STUDYID, USUBJID, EXTRT)2627dm <- rand |>28  select(STUDYID, USUBJID, RANDCD, RAND, SCRNFL) |>29  full_join(exposure, by = c("STUDYID", "USUBJID")) |>30  mutate(31    RANDCD = na_if(str_trim(RANDCD), ""),32    RAND = na_if(str_trim(RAND), ""),33    EXTRT = na_if(str_trim(EXTRT), ""),34    SCRNFL = coalesce(str_trim(SCRNFL), ""),35    # Never randomized leaves the planned arm empty; never treated or a36    # treatment matching no planned arm leaves the actual arm empty.37    ARMCD = if_else(is.na(RANDCD), NA_character_, RANDCD),38    ARM = if_else(is.na(RANDCD), NA_character_, RAND),39    ACTARMCD = case_when(40      EXTRT == "PLACEBO" ~ "PBO",41      EXTRT == "VITAMIN D3" ~ "TRT",42      .default = NA_character_43    ),44    ACTARM = case_when(45      EXTRT == "PLACEBO" ~ "Placebo",46      EXTRT == "VITAMIN D3" ~ "Vitamin D3",47      .default = NA_character_48    ),49    # A treated subject carries the other planned arm when that is what50    # the exposure says; the reason for a blank arm depends on why both51    # are blank, or on never being treated.52    ARMNRS = case_when(53      is.na(ARM) & is.na(ACTARM) & SCRNFL == "Y" ~ "SCREEN FAILURE",54      is.na(ARM) & is.na(ACTARM) ~ "NOT ASSIGNED",55      !is.na(ARM) & is.na(ACTARM) & is.na(EXTRT) ~ "NOT TREATED",56      .default = NA_character_57    ),58    ACTARMUD = if_else(59      !is.na(EXTRT) & is.na(ACTARM), EXTRT, NA_character_, missing = NA_character_60    ),61    DOMAIN = "DM"62  ) |>63  arrange(USUBJID) |>64  select(65    DOMAIN, STUDYID, USUBJID, ARMCD, ARM,66    ACTARMCD, ACTARM, ARMNRS, ACTARMUD67  )6869dir.create("/app/output", showWarnings = FALSE)70write_csv(dm, "/app/output/dm.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-dm-arms (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 pl1011rand = pl.read_csv("/app/input/rand.csv", infer_schema=False)12ex = pl.read_csv("/app/input/ex.csv", infer_schema=False)1314# The planned arm follows randomization; the actual arm follows exposure.15# PLACEBO was the PBO arm and VITAMIN D3 the TRT arm.16planned = {"PLACEBO": ("PBO", "Placebo"), "VITAMIN D3": ("TRT", "Vitamin D3")}1718exposure = (19    ex.group_by(["STUDYID", "USUBJID"])20    .agg(EXTRT=pl.col("EXTRT").filter(pl.col("EXTRT").fill_null("") != "").first())21    .with_columns(EXTRT=pl.col("EXTRT").fill_null(""))22)2324subjects = (25    rand.select("STUDYID", "USUBJID", "RANDCD", "RAND", "SCRNFL")26    .join(exposure, on=["STUDYID", "USUBJID"], how="full", coalesce=True)27)2829rows = []30for record in subjects.to_dicts():31    studyid = record["STUDYID"]32    usubjid = record["USUBJID"]33    randcd = (record.get("RANDCD") or "").strip() or None34    randdesc = (record.get("RAND") or "").strip() or None35    extrt = (record.get("EXTRT") or "").strip() or None36    scrnfl = (record.get("SCRNFL") or "").strip()37    # Never randomized leaves the planned arm empty; never treated or a38    # treatment matching no planned arm leaves the actual arm empty.39    armcd = randcd if randcd else None40    arm = randdesc if randcd else None41    actcd, act = planned.get(extrt, (None, None))42    # A treated subject carries the other planned arm when that is what43    # the exposure says; the reason for a blank arm depends on why both44    # are blank, or on never being treated.45    if arm is None and act is None:46        armnrs = "SCREEN FAILURE" if scrnfl == "Y" else "NOT ASSIGNED"47    elif arm is not None and act is None and extrt is None:48        armnrs = "NOT TREATED"49    else:50        armnrs = None51    armud = extrt if (extrt is not None and extrt not in planned) else None52    rows.append(53        {54            "DOMAIN": "DM",55            "STUDYID": studyid,56            "USUBJID": usubjid,57            "ARMCD": armcd,58            "ARM": arm,59            "ACTARMCD": actcd,60            "ACTARM": act,61            "ARMNRS": armnrs,62            "ACTARMUD": armud,63        }64    )6566dm = pl.DataFrame(67    rows,68    schema={69        "DOMAIN": pl.String,70        "STUDYID": pl.String,71        "USUBJID": pl.String,72        "ARMCD": pl.String,73        "ARM": pl.String,74        "ACTARMCD": pl.String,75        "ACTARM": pl.String,76        "ARMNRS": pl.String,77        "ACTARMUD": pl.String,78    },79).sort("USUBJID")8081Path("/app/output").mkdir(exist_ok=True)82dm.write_csv("/app/output/dm.csv")

1/* opensas reference solution for the yamaa benchmark sdtm-dm-arms.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data ex;6  length STUDYID USUBJID EXTRT $1024;7  infile "/app/input/ex.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. EXTRT : $char1024.;9run;1011data rand;12  length STUDYID USUBJID RANDCD RAND SCRNFL $1024;13  infile "/app/input/rand.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. RANDCD : $char1024. RAND : $char1024. SCRNFL : $char1024.;15run;1617/* Keep the first nonmissing collected exposure per subject. */18data nonmissing_ex; set ex; if not missing(EXTRT); EXORDER=_N_; run;19proc sort data=nonmissing_ex; by STUDYID USUBJID EXORDER; run;20data exposure; set nonmissing_ex; by STUDYID USUBJID; if first.USUBJID; run;21proc sql; create table joined as select a.*,b.EXTRT from rand a left join exposure b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID; quit;22data result; set joined; length DOMAIN $2 ARMCD ARM ACTARMCD ACTARM ARMNRS ACTARMUD $1024; DOMAIN='DM'; ARMCD=RANDCD; ARM=RAND;23if EXTRT='PLACEBO' then do; ACTARMCD='PBO'; ACTARM='Placebo'; end;24else if EXTRT='VITAMIN D3' then do; ACTARMCD='TRT'; ACTARM='Vitamin D3'; end;25else if not missing(EXTRT) then ACTARMUD=EXTRT;26if missing(EXTRT) then do; if not missing(ARMCD) then ARMNRS='NOT TREATED'; else if SCRNFL='Y' then ARMNRS='SCREEN FAILURE'; else ARMNRS='NOT ASSIGNED'; end; run;27proc sql;28  create table final as select DOMAIN, STUDYID, USUBJID, ARMCD, ARM, ACTARMCD, ACTARM, ARMNRS, ACTARMUD from result;29quit;30proc export data=final outfile="/app/output/dm.csv" dbms=csv replace; run;