Assign Analysis Windows by Study Day

1library(yamaa)23advs <- yamaa_domain("spec.yaml")$output4advs

1import yamaa23advs = yamaa.yamaa_domain("spec.yaml").output4advs

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 1.67from pathlib import Path89import polars as pl1011# SDTM VS staged as the ADaM input; rename to the analysis names the12# rest of the solution works with.13raw = (14    pl.read_csv("/app/input/vs.csv", infer_schema=False)15    .rename(16        {17            "VSTESTCD": "PARAMCD",18            "VSDTC": "ADT",19            "VSDY": "ADY",20            "VSSTRESN": "AVAL",21        }22    )23    .with_columns(24        pl.col("VSSEQ").cast(pl.Int64, strict=False),25        pl.col("VISITNUM").cast(pl.Float64, strict=False),26        pl.col("ADT").str.to_date(strict=False),27        pl.col("ADY").cast(pl.Int64, strict=False),28        pl.col("AVAL").cast(pl.Float64, strict=False),29    )30)313233def visit(day: int | None) -> tuple[str | None, int | None]:34    if day is None:35        return None, None36    if day < 0:37        return "SCREENING", -138    if day < 2:39        return "BASELINE", 040    if day < 22:41        return "WEEK 2", 242    if day < 43:43        return "WEEK 4", 444    return "POST-TREATMENT", 99454647framed = raw.with_columns(48    pl.col("ADY")49    .map_elements(lambda d: visit(d)[0], return_dtype=pl.String)50    .alias("AVISIT"),51    pl.col("ADY")52    .map_elements(lambda d: visit(d)[1], return_dtype=pl.Int64)53    .alias("AVISITN"),54)5556# SV lists every planned visit of every subject, including those that did57# not take place, and the unscheduled visits.58sv = pl.read_csv("/app/input/sv.csv", infer_schema=False).with_columns(59    pl.col("VISITNUM").cast(pl.Float64, strict=False)60)6162# Each planned SCREENING, BASELINE, WEEK 2, or WEEK 4 visit gets one SYSBP63# expected record when no collected record's study day falls in the64# analysis window of that name: the planned visit name and number and the65# window, and no date, study day, or value. No other visit gets one.66# Expected records continue the subject's sequence numbering after the67# highest collected VSSEQ, in VISITNUM order.68windows = {"SCREENING": -1, "BASELINE": 0, "WEEK 2": 2, "WEEK 4": 4}69keys = ["STUDYID", "USUBJID", "PARAMCD"]70covered = framed.select(*keys, "AVISIT").drop_nulls().unique()71top = framed.group_by("STUDYID", "USUBJID").agg(pl.col("VSSEQ").max().alias("TOP"))7273expected = (74    sv.filter(pl.col("VISIT").is_in(list(windows)))75    .with_columns(76        PARAMCD=pl.lit("SYSBP"),77        AVISIT=pl.col("VISIT"),78        AVISITN=pl.col("VISIT").replace_strict(windows, return_dtype=pl.Int64),79    )80    .join(covered, on=[*keys, "AVISIT"], how="anti")81    .join(top, on=["STUDYID", "USUBJID"], how="left")82    .sort([*keys, "VISITNUM"])83    .with_columns(84        VSSEQ=pl.col("TOP").fill_null(0)85        + pl.int_range(1, pl.len() + 1, dtype=pl.Int64).over(keys),86        ADT=pl.lit(None, dtype=pl.Date),87        ADY=pl.lit(None, dtype=pl.Int64),88        AVAL=pl.lit(None, dtype=pl.Float64),89    )90    .select(91        *keys, "VSSEQ", "VISIT", "VISITNUM", "ADT", "ADY", "AVAL",92        "AVISIT", "AVISITN",93    )94)95framed = pl.concat([framed, expected], how="vertical")9697# The earliest record by study day in each study, subject, parameter,98# and visit; the lower sequence number breaks a same-day tie. Expected99# records have no study day and never take the flag.100flagged = (101    framed.filter(pl.col("AVISIT").is_not_null() & pl.col("ADY").is_not_null())102    .sort(["ADY", "VSSEQ"])103    .unique(["STUDYID", "USUBJID", "PARAMCD", "AVISIT"], keep="first", maintain_order=True)104    .select("STUDYID", "USUBJID", "PARAMCD", "AVISIT", "VSSEQ")105    .with_columns(ANL01FL=pl.lit("Y"))106)107108advs = (109    framed.join(110        flagged,111        on=["STUDYID", "USUBJID", "PARAMCD", "AVISIT", "VSSEQ"],112        how="left",113    )114    .sort(["STUDYID", "USUBJID", "VSSEQ"])115    .select(116        "STUDYID", "USUBJID", "PARAMCD", "VSSEQ", "VISIT", "VISITNUM",117        "ADT", "ADY", "AVAL", "AVISIT", "AVISITN", "ANL01FL",118    )119)120121Path("/app/output").mkdir(parents=True, exist_ok=True)122advs.write_csv("/app/output/advs.csv")

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;