module svm

solverpy_learn.builder.svm

deconflict(xs: spmatrix, ys: ndarray) -> tuple[spmatrix, ndarray]

Find conflicting positive and negative samples and remove the negative ones.

Source code in packages/solverpy-learn/src/solverpy_learn/builder/svm.py

def deconflict(xs: "spmatrix", ys: "ndarray") -> tuple["spmatrix", "ndarray"]:
   "Find conflicting positive and negative samples and remove the negative ones."
   assert xs.shape[0] == ys.shape[0]
   logger.info("Looking up conflicting samples.")
   logger.debug("building samples map")
   dups = defaultdict(list)
   for i in range(xs.shape[0]):
      row = xs.getrow(i)
      key = (tuple(row.indices), tuple(row.data))
      dups[key].append(i)
   logger.debug("marking conflicting negative samples")
   todel = set()
   for ids in dups.values():
      if len(ids) < 2:
         continue
      td = []
      onepos = False
      for i in ids:
         if ys[i] == 0:
            td.append(i)
         else:
            onepos = True
      if onepos:
         todel.update(td)
   logger.debug("deleting marked rows")
   keep = [i for i in range(xs.shape[0]) if i not in todel]
   xs0 = xs[keep]  # type: ignore
   ys0 = ys[keep]
   logger.info("\n".join([
      "Data shape difference:",
      f"\t{xs.shape} --> {xs0.shape}",
      f"\t{ys.shape} --> {ys0.shape}",
   ]))
   return (xs0, ys0)