fix: close retest findings N1-N5 (pipeline exit code, cpads path, xgbst inference, verifier regex)

N2: execute_pipeline now tracks per-module failures, prints "FAILED" (not
"Done") for a module that errored, and returns 1 when any module failed —
previously it always printed "completed successfully" and returned 0, so CI
and humans could not tell a 7/23 run from a 23/23 one.

N1: the Python bridge resolves --cpads-csv (and --output-dir) to absolute
paths before launching Rscript with cwd=output_dir; the R side's fragile
10-level parent-directory walk is removed in favour of resolve-or-error. A
relative --cpads-csv with --output-dir outside the repo silently killed all
16 R-backed modules.

N4: drop the `is.integer(y)` clause from xgbst task inference — count
outcomes are integers and must resolve to regression, not be auto-routed to
a binary classifier. Binary is already caught by `all(y %in% c(0,1))`.

N3: the gbm fallback hands bernoulli a numeric {0,1}, not a factor (matching
the xgboost branch), so classification actually fits for users without
xgboost (Suggests-only).

N5: anchor the r-squared verifier pattern so bare `r2` no longer matches the
substring in column names like `factor2`.

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Vansh Singh Ruhela (rootcoder007) <vsruhela@proton.me>

Author: rootcoder007

r-package/morie/R/modules.R

@@ -75,22 +75,18 @@ morie_list_morie_modules <- function() {
 }
 
 .resolve_cpads_csv <- function(cpads_csv) {
+  # The Python bridge passes an absolute path (see modules.py, finding N1);
+  # a direct Rscript caller is responsible for a path valid from its own cwd.
+  # No parent-directory walk: it only ever "worked" when the output dir
+  # happened to sit inside the repo, and silently surprised everyone else.
   if (file.exists(cpads_csv)) {
     return(normalizePath(cpads_csv, mustWork = TRUE))
   }
-  current <- normalizePath(getwd(), winslash = "/", mustWork = TRUE)
-  for (i in seq_len(10L)) {
-    candidate <- file.path(current, cpads_csv)
-    if (file.exists(candidate)) {
-      return(normalizePath(candidate, mustWork = TRUE))
-    }
-    parent <- dirname(current)
-    if (identical(parent, current)) {
-      break
-    }
-    current <- parent
-  }
-  stop("CPADS CSV not found: ", cpads_csv, call. = FALSE)
+  stop(
+    "CPADS CSV not found: ", cpads_csv,
+    "\n  Pass an absolute path (or one valid from the current working directory).",
+    call. = FALSE
+  )
 }
 
 #' Canonicalize raw CPADS PUMF columns

r-package/morie/R/xgbst.R

@@ -34,7 +34,10 @@ morie_xgboost_objective <- function(x, y, n_estimators = 100L, learning_rate = 0
   if (is.null(dim(x))) x <- matrix(x, ncol = 1)
   x <- as.matrix(x)
   if (identical(task, "auto")) {
-    task <- if (is.factor(y) || all(y %in% c(0L, 1L)) || is.integer(y)) {
+    # A 0/1 vector (integer or double) is caught by the `%in%` test; do NOT
+    # treat every integer as classification — count outcomes are integers too
+    # and must resolve to regression (N4).
+    task <- if (is.factor(y) || all(y %in% c(0L, 1L))) {
       "classification"
     } else {
       "regression"
@@ -79,7 +82,9 @@ morie_xgboost_objective <- function(x, y, n_estimators = 100L, learning_rate = 0
     if (!requireNamespace("gbm", quietly = TRUE)) {
       stop("install 'xgboost' (preferred) or 'gbm' for morie_xgboost_objective")
     }
-    yv <- if (task == "classification") factor(y) else as.numeric(y)
+    # gbm's bernoulli requires numeric {0,1}, NOT a factor (N3). Use the same
+    # coercion as the xgboost branch above so the fallback actually fits.
+    yv <- if (task == "classification") as.numeric(as.factor(y)) - 1 else as.numeric(y)
     df <- as.data.frame(x)
     df$.y <- yv
     distribution <- if (task == "classification") "bernoulli" else "gaussian"

src/morie/inspector.py

@@ -43,7 +43,10 @@
     r"|^ate$|^att$|^atc$|point[_.\-]?est|post[_.\-]?mean)",
     re.IGNORECASE,
 )
-_R_SQUARED_PATTERNS = re.compile(r"(r[_.\-]?squared|r2|rsq|pseudo[_.\-]?r2|adj[_.\-]?r2)", re.IGNORECASE)
+_R_SQUARED_PATTERNS = re.compile(
+    r"(^|[_.\-])(r[_.\-]?squared|r2|rsq|pseudo[_.\-]?r2|adj[_.\-]?r2)($|[_.\-])",
+    re.IGNORECASE,
+)  # anchored so bare `r2` no longer matches the substring in `factor2`/`predictor2` (N5)
 _SAMPLE_SIZE_PATTERNS = re.compile(r"(^n$|^n[_.\-]obs|sample[_.\-]?size|^nobs$|n_total)", re.IGNORECASE)
 _AIC_BIC_PATTERNS = re.compile(r"(^aic$|^bic$)", re.IGNORECASE)
 

src/morie/modules.py

@@ -417,8 +417,12 @@ def _run_r_module(
     if output_dir is None:
         _tmp_ctx = tempfile.TemporaryDirectory(prefix=f"morie-{module_name}-")
         output_dir = Path(_tmp_ctx.name)
-    output_dir = Path(output_dir)
+    output_dir = Path(output_dir).expanduser().resolve()
     output_dir.mkdir(parents=True, exist_ok=True)
+    # Rscript runs with cwd=output_dir, so any relative path we forward would
+    # resolve against the OUTPUT dir, not the caller's cwd. Make cpads_csv
+    # absolute here so it works regardless of where output lands (N1).
+    cpads_csv = Path(cpads_csv).expanduser().resolve()
     try:
         if not _R_MODULE_SHIM.exists():
             raise RuntimeError(

src/morie/runner.py

@@ -93,6 +93,7 @@ def execute_pipeline(
     except ImportError:
         pass
 
+    failed: dict[str, str] = {}
     for idx, module_name in enumerate(selected, start=1):
         if _pbar:
             _pbar.desc = f"Running: {module_name}"
@@ -105,12 +106,15 @@ def execute_pipeline(
                 dataset_key=dataset_key,
                 output_dir=output_dir,
             )
+            status = "Done"
         except Exception as exc:
+            failed[module_name] = str(exc)
             print(f"  ERROR in {module_name}: {exc}")
+            status = "FAILED"
         if _pbar:
             _pbar.update()
         else:
-            print(f"[{idx}/{total}] Done: {module_name}", flush=True)
+            print(f"[{idx}/{total}] {status}: {module_name}", flush=True)
 
     if _manager:
         _manager.stop()
@@ -123,7 +127,16 @@ def execute_pipeline(
         except Exception:  # pragma: no cover
             pass
 
-    print("Pipeline completed successfully.")
+    n_ok = len(results)
+    if failed:
+        print(
+            f"Pipeline completed {n_ok}/{total} "
+            f"({len(failed)} failed: {', '.join(failed)})."
+        )
+        if results:
+            print("Succeeded modules:", ", ".join(results.keys()))
+        return 1
+    print(f"Pipeline completed successfully ({n_ok}/{total}).")
     print("Completed modules:", ", ".join(results.keys()))
     return 0