Assign Analysis Windows by Study Day

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

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

1# Reference solution for the yamaa benchmark adam-advs-windows (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)910# SDTM VS staged as the ADaM input; rename to the analysis names the11# rest of the solution works with.12raw <- read_csv(13  "/app/input/vs.csv",14  col_types = cols(15    VSSEQ = col_integer(), VISITNUM = col_double(), VSDTC = col_date(),16    VSDY = col_integer(), VSSTRESN = col_double(), .default = col_character()17  )18) |>19  rename(PARAMCD = VSTESTCD, ADT = VSDTC, ADY = VSDY, AVAL = VSSTRESN)2021# Windows follow the study day: screening before day 0, baseline on day22# 1, weeks 2 and 4 on days 2-21 and 22-42, and post-treatment from day23# 43 on with no upper bound. No study day means no window.24framed <- raw |>25  mutate(26    AVISIT = case_when(27      is.na(ADY) ~ NA_character_,28      ADY < 0 ~ "SCREENING",29      ADY < 2 ~ "BASELINE",30      ADY < 22 ~ "WEEK 2",31      ADY < 43 ~ "WEEK 4",32      .default = "POST-TREATMENT"33    ),34    AVISITN = case_when(35      AVISIT %in% "SCREENING" ~ -1L,36      AVISIT %in% "BASELINE" ~ 0L,37      AVISIT %in% "WEEK 2" ~ 2L,38      AVISIT %in% "WEEK 4" ~ 4L,39      AVISIT %in% "POST-TREATMENT" ~ 99L40    )41  )4243# SV lists every planned visit of every subject, including those that did44# not take place, and the unscheduled visits.45sv <- read_csv(46  "/app/input/sv.csv",47  col_types = cols(VISITNUM = col_double(), .default = col_character())48)4950# Each planned SCREENING, BASELINE, WEEK 2, or WEEK 4 visit gets one SYSBP51# expected record when no collected record's study day falls in the52# analysis window of that name: the planned visit name and number and the53# window, and no date, study day, or value. No other visit gets one.54# Expected records continue the subject's sequence numbering after the55# highest collected VSSEQ, in VISITNUM order.56windows <- c("SCREENING" = -1L, "BASELINE" = 0L, "WEEK 2" = 2L, "WEEK 4" = 4L)5758top <- framed |>59  group_by(STUDYID, USUBJID) |>60  summarise(top_seq = max(VSSEQ), .groups = "drop")6162expected <- sv |>63  filter(VISIT %in% names(windows)) |>64  mutate(65    PARAMCD = "SYSBP", AVISIT = VISIT, AVISITN = unname(windows[VISIT])66  ) |>67  anti_join(68    filter(framed, !is.na(AVISIT)),69    by = c("STUDYID", "USUBJID", "PARAMCD", "AVISIT")70  ) |>71  left_join(top, by = c("STUDYID", "USUBJID")) |>72  group_by(STUDYID, USUBJID, PARAMCD) |>73  arrange(VISITNUM, .by_group = TRUE) |>74  mutate(75    VSSEQ = coalesce(top_seq, 0L) + row_number(),76    ADT = as.Date(NA),77    ADY = NA_integer_,78    AVAL = NA_real_79  ) |>80  ungroup() |>81  select(82    STUDYID, USUBJID, PARAMCD, VSSEQ, VISIT, VISITNUM, ADT, ADY, AVAL,83    AVISIT, AVISITN84  )8586framed <- bind_rows(framed, expected)8788# The earliest record by study day in each study, subject, parameter,89# and visit; the lower sequence number breaks a same-day tie. Expected90# records have no study day and never take the flag.91flagged <- framed |>92  filter(!is.na(AVISIT), !is.na(ADY)) |>93  arrange(ADY, VSSEQ) |>94  distinct(STUDYID, USUBJID, PARAMCD, AVISIT, .keep_all = TRUE) |>95  select(STUDYID, USUBJID, PARAMCD, AVISIT, VSSEQ) |>96  mutate(ANL01FL = "Y")9798advs <- framed |>99  left_join(flagged, by = c("STUDYID", "USUBJID", "PARAMCD", "AVISIT", "VSSEQ")) |>100  arrange(STUDYID, USUBJID, VSSEQ) |>101  select(102    STUDYID, USUBJID, PARAMCD, VSSEQ, VISIT, VISITNUM, ADT, ADY, AVAL,103    AVISIT, AVISITN, ANL01FL104  )105106dir.create("/app/output", showWarnings = FALSE)107write_csv(advs, "/app/output/advs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-advs-windows (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_DIR = "/app/input"11OUTPUT_PATH = "/app/output/advs.csv"1213COLUMNS = [14    "STUDYID", "USUBJID", "PARAMCD", "VSSEQ", "VISIT", "VISITNUM",15    "ADT", "ADY", "AVAL", "AVISIT", "AVISITN", "ANL01FL",16]1718AVISITN = {19    "SCREENING": -1,20    "BASELINE": 0,21    "WEEK 2": 2,22    "WEEK 4": 4,23    "POST-TREATMENT": 99,24}2526EXPECTED_VISITS = ["SCREENING", "BASELINE", "WEEK 2", "WEEK 4"]272829def window_for(ady):30    """Analysis visit whose window holds the study day, or None."""31    if ady is None:32        return None33    if ady < 0:34        return "SCREENING"35    if ady == 1:36        return "BASELINE"37    if 2 <= ady <= 21:38        return "WEEK 2"39    if 22 <= ady <= 42:40        return "WEEK 4"41    if ady >= 43:42        return "POST-TREATMENT"43    return None444546def to_int(text):47    text = (text or "").strip()48    if text == "":49        return None50    return int(float(text))515253def to_num_text(text):54    text = (text or "").strip()55    if text == "":56        return ""57    value = float(text)58    if value == int(value):59        return str(int(value))60    return repr(value)616263def read_csv(path):64    with open(path, newline="") as handle:65        return list(csv.DictReader(handle))666768def main():69    sv_rows = read_csv(INPUT_DIR + "/sv.csv")70    vs_rows = read_csv(INPUT_DIR + "/vs.csv")7172    # Planned visits per subject from SV (visit name -> visit number).73    planned = OrderedDict()74    for row in sv_rows:75        subject = row["USUBJID"]76        planned.setdefault(subject, OrderedDict())77        if row["VISIT"] in EXPECTED_VISITS:78            planned[subject][row["VISIT"]] = to_int(row["VISITNUM"])7980    records = []81    subjects = OrderedDict()  # subject -> {"max_seq": int, "windows": set}82    for row in vs_rows:83        subject = row["USUBJID"]84        ady = to_int(row["VSDY"])85        avisit = window_for(ady)86        info = subjects.setdefault(subject, {"max_seq": 0, "windows": set()})87        seq = to_int(row["VSSEQ"])88        info["max_seq"] = max(info["max_seq"], seq)89        if avisit is not None:90            info["windows"].add(avisit)91        records.append(OrderedDict([92            ("STUDYID", row["STUDYID"]),93            ("USUBJID", subject),94            ("PARAMCD", row["VSTESTCD"]),95            ("VSSEQ", seq),96            ("VISIT", row["VISIT"]),97            ("VISITNUM", to_int(row["VISITNUM"])),98            ("ADT", (row["VSDTC"] or "").strip()),99            ("ADY", ady),100            ("AVAL", to_num_text(row["VSSTRESN"])),101            ("AVISIT", avisit or ""),102            ("AVISITN", AVISITN[avisit] if avisit else None),103            ("ANL01FL", ""),104            ("_expected", False),105        ]))106107    # ANL01FL: earliest by study day, lower VSSEQ breaks a tie, per108    # study/subject/parameter/analysis visit. Records with no study day109    # (hence no window) never take the flag.110    best = {}111    for rec in records:112        if rec["ADY"] is None or rec["AVISIT"] == "":113            continue114        key = (rec["STUDYID"], rec["USUBJID"], rec["PARAMCD"], rec["AVISIT"])115        rank = (rec["ADY"], rec["VSSEQ"])116        if key not in best or rank < best[key][0]:117            best[key] = (rank, rec)118    for _, rec in best.values():119        rec["ANL01FL"] = "Y"120121    # Expected records for planned visits whose window holds no measured122    # study day for the subject.123    expected_by_subject = OrderedDict()124    for subject, visits in planned.items():125        info = subjects.get(subject, {"max_seq": 0, "windows": set()})126        missing = [127            (num, name) for name, num in visits.items()128            if name not in info["windows"]129        ]130        missing.sort(key=lambda item: (item[0] is None, item[0]))131        expected = []132        for place, (num, name) in enumerate(missing, start=1):133            expected.append(OrderedDict([134                ("STUDYID", next(135                    r["STUDYID"] for r in sv_rows if r["USUBJID"] == subject)),136                ("USUBJID", subject),137                ("PARAMCD", "SYSBP"),138                ("VSSEQ", info["max_seq"] + place),139                ("VISIT", name),140                ("VISITNUM", num),141                ("ADT", ""),142                ("ADY", None),143                ("AVAL", ""),144                ("AVISIT", name),145                ("AVISITN", AVISITN[name]),146                ("ANL01FL", ""),147                ("_expected", True),148            ]))149        expected_by_subject[subject] = expected150151    # Emit collected records in input order, with each subject's expected152    # records (VSSEQ continuing the subject's sequence) after them.153    last_index = {}154    for i, rec in enumerate(records):155        last_index[rec["USUBJID"]] = i156    output = []157    for i, rec in enumerate(records):158        output.append(rec)159        if last_index[rec["USUBJID"]] == i:160            output.extend(expected_by_subject.get(rec["USUBJID"], []))161    for subject, expected in expected_by_subject.items():162        if subject not in last_index:163            output.extend(expected)164165    with open(OUTPUT_PATH, "w", newline="") as handle:166        writer = csv.writer(handle)167        writer.writerow(COLUMNS)168        for rec in output:169            writer.writerow([170                "" if rec[col] is None else rec[col] for col in COLUMNS171            ])172173174if __name__ == "__main__":175    main()

1/* opensas reference solution for the yamaa benchmark adam-advs-windows.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data sv;6  length STUDYID USUBJID VISIT SVPRESP SVOCCUR SVSTDTC $1024;7  infile "/app/input/sv.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. VISITNUM VISIT : $char1024. SVPRESP : $char1024. SVOCCUR : $char1024. SVSTDTC : $char1024.;9run;1011data vs;12  length STUDYID USUBJID VSTESTCD VISIT VSDTC VSDY $1024;13  infile "/app/input/vs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. VSTESTCD : $char1024. VSSEQ VISIT : $char1024. VISITNUM VSDTC : $char1024. VSDY : $char1024. VSSTRESN;15run;1617data collected;18  length PARAMCD $16 ADT $10 AVISIT $20 ANL01FL $1;19  set vs;20  PARAMCD=VSTESTCD; ADT=VSDTC; ADY=input(VSDY,best32.); AVAL=VSSTRESN;21  if not missing(ADY) then do;22    if ADY<0 then do; AVISIT='SCREENING'; AVISITN=-1; end;23    else if ADY=1 then do; AVISIT='BASELINE'; AVISITN=0; end;24    else if ADY>=2 and ADY<=21 then do; AVISIT='WEEK 2'; AVISITN=2; end;25    else if ADY>=22 and ADY<=42 then do; AVISIT='WEEK 4'; AVISITN=4; end;26    else if ADY>=43 then do; AVISIT='POST-TREATMENT'; AVISITN=99; end;27  end;28run;29proc sort data=collected; by STUDYID USUBJID PARAMCD AVISITN ADY VSSEQ; run;30data flagged; set collected; by STUDYID USUBJID PARAMCD AVISITN; if first.AVISITN and not missing(AVISITN) then ANL01FL='Y'; run;31proc sql;32  create table maxima as select STUDYID,USUBJID,max(VSSEQ) as maximum from collected group by STUDYID,USUBJID;33  create table absent as select a.*,b.maximum from sv as a left join maxima as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID34  where a.VISIT in ('SCREENING','BASELINE','WEEK 2','WEEK 4') and not exists(select * from collected as c where a.STUDYID=c.STUDYID and a.USUBJID=c.USUBJID and c.PARAMCD='SYSBP' and a.VISIT=c.AVISIT);35quit;36proc sort data=absent; by STUDYID USUBJID VISITNUM; run;37data expected;38  length PARAMCD $16 ADT $10 AVISIT $20 ANL01FL $1;39  set absent; by STUDYID USUBJID;40  if first.USUBJID then counter=coalesce(maximum,0); counter+1; VSSEQ=counter;41  PARAMCD='SYSBP'; AVISIT=VISIT;42  select(VISIT); when('SCREENING') AVISITN=-1; when('BASELINE') AVISITN=0; when('WEEK 2') AVISITN=2; when('WEEK 4') AVISITN=4; otherwise; end;43  ADY=.; AVAL=.;44run;45data result; set flagged expected; run;4647proc sql;48  create table final as select STUDYID, USUBJID, PARAMCD, VSSEQ, VISIT, VISITNUM, ADT, ADY, AVAL, AVISIT, AVISITN, ANL01FL from result;49quit;50proc export data=final outfile="/app/output/advs.csv" dbms=csv replace; run;