fix: C contiguous error for exp mov avg tx

Signed-off-by: Avik Basu <[email protected]>

numalogic/config/factory.py

@@ -52,7 +52,7 @@ class PreprocessFactory(_ObjectFactory):
  DifferenceTransform,
  FlattenVector,
  PercentileScaler,
- ExpMovingAverage
+ ExpMovingAverage,
  )
 
  _CLS_MAP: ClassVar[dict] = {
@@ -68,7 +68,7 @@ class PreprocessFactory(_ObjectFactory):
  "DifferenceTransform": DifferenceTransform,
  "FlattenVector": FlattenVector,
  "PercentileScaler": PercentileScaler,
- "ExpMovingAverage": ExpMovingAverage
+ "ExpMovingAverage": ExpMovingAverage,
  }
 
  def get_pipeline_instance(self, objs_info: list[ModelInfo]):
@@ -88,7 +88,11 @@ class PostprocessFactory(_ObjectFactory):
 
  from numalogic.transforms import TanhNorm, ExpMovingAverage, SigmoidNorm
 
- _CLS_MAP: ClassVar[dict] = {"TanhNorm": TanhNorm, "ExpMovingAverage": ExpMovingAverage, "SigmoidNorm": SigmoidNorm}
+ _CLS_MAP: ClassVar[dict] = {
+ "TanhNorm": TanhNorm,
+ "ExpMovingAverage": ExpMovingAverage,
+ "SigmoidNorm": SigmoidNorm,
+ }
 
 
 class ThresholdFactory(_ObjectFactory):

numalogic/transforms/__init__.py

@@ -39,5 +39,5 @@
  "DifferenceTransform",
  "FlattenVector",
  "PercentileScaler",
- "SigmoidNorm"
+ "SigmoidNorm",
 ]

numalogic/transforms/_movavg.py

@@ -111,5 +111,6 @@ def transform(self, input_: npt.NDArray[float], **__) -> npt.NDArray[float]:
  ------
  InvalidDataShapeError: if input array is not single featured
  """
- x_df = pd.DataFrame(input_)
- return x_df.ewm(alpha=self.alpha).mean().to_numpy(dtype=np.float32)
+ x_df = pd.DataFrame(input_, dtype=np.float32, copy=True)
+ x_smoothed = x_df.ewm(alpha=self.alpha).mean().to_numpy(dtype=np.float32)
+ return np.ascontiguousarray(x_smoothed, dtype=np.float32)

numalogic/transforms/_postprocess.py

@@ -51,10 +51,10 @@ def transform(self, input_: npt.NDArray[float], **__) -> npt.NDArray[float]:
 
 
 class SigmoidNorm(StatelessTransformer):
- def __init__(self, scale_factor: float = 10., smooth_factor: float = 0.5):
+ def __init__(self, scale_factor: float = 10.0, smooth_factor: float = 0.5):
  super().__init__()
  self.scale_factor = scale_factor
  self.smooth_factor = smooth_factor
 
  def transform(self, x: npt.NDArray[float], **__) -> npt.NDArray[float]:
- return self.scale_factor / (1.0 + np.exp(5 - (self.smooth_factor * x)))
+ return self.scale_factor / (1.0 + np.exp(5 - (self.smooth_factor * x)))

numalogic/transforms/_scaler.py

@@ -89,7 +89,9 @@ class PercentileScaler(BaseTransformer):
  Default is None.
  """
 
- def __init__(self, max_percentile: float = 99, min_percentile: Optional[float] = None, eps: float = 1e-2):
+ def __init__(
+ self, max_percentile: float = 99, min_percentile: Optional[float] = None, eps: float = 1e-2
+ ):
  self._max_px = max_percentile
  self._min_px = min_percentile
  self.tx = MinMaxScaler()
@@ -120,8 +122,9 @@ def fit(self, x: npt.NDArray[float]) -> Self:
  for idx, _range in enumerate(p_ranges):
  if _range <= self._eps:
  LOGGER.warning(
- "Max and Min percentile difference is less than "
- "epsilon: %s for column %s", self._eps, idx
+ "Max and Min percentile difference is less than " "epsilon: %s for column %s",
+ self._eps,
+ idx,
  )
  data_max_px[idx] = data_max[idx]
 

tests/transforms/test_postprocess.py

@@ -4,7 +4,6 @@
 from sklearn.pipeline import make_pipeline
 
 from numalogic.transforms import tanh_norm, TanhNorm, ExpMovingAverage, expmov_avg_aggregator
-from numalogic.tools.exceptions import InvalidDataShapeError
 
 
 class TestPostprocess(unittest.TestCase):
@@ -22,21 +21,17 @@ def test_tanh_norm_clf(self):
  self.assertTupleEqual(arr.shape, scores.shape)
  self.assertAlmostEqual(np.sum(scores), 39.52, places=2)
 
- def test_exp_mov_avg_estimator_01(self):
+ def test_exp_mov_avg_estimator(self):
+ beta = 0.9
  arr = np.arange(1, 11).reshape(-1, 1)
- clf = ExpMovingAverage(0.9)
+ clf = ExpMovingAverage(beta)
  out = clf.fit_transform(arr)
- self.assertTupleEqual(arr.shape, out.shape)
- self.assertAlmostEqual(expmov_avg_aggregator(arr, 0.9), out[-1].item(), places=3)
 
- def test_exp_mov_avg_estimator_02(self):
- arr = np.arange(1, 11).reshape(-1, 1)
- clf = ExpMovingAverage(0.9, bias_correction=False)
- out = clf.fit_transform(arr)
+ expected = expmov_avg_aggregator(arr, beta)
+
  self.assertTupleEqual(arr.shape, out.shape)
- self.assertAlmostEqual(
- expmov_avg_aggregator(arr, 0.9, bias_correction=False), out[-1].item(), places=3
- )
+ self.assertAlmostEqual(expected, out[-1].item(), places=2)
+ self.assertTrue(out.data.c_contiguous)
 
  def test_exp_mov_avg_estimator_err(self):
  with self.assertRaises(ValueError):
@@ -48,14 +43,6 @@ def test_exp_mov_avg_estimator_err(self):
  with self.assertRaises(ValueError):
  ExpMovingAverage(1.0)
 
- with self.assertRaises(InvalidDataShapeError):
- clf = ExpMovingAverage(0.1)
- clf.fit_transform(np.arange(10).reshape(1, -1))
-
- with self.assertRaises(InvalidDataShapeError):
- clf = ExpMovingAverage(0.1)
- clf.fit_transform(np.arange(12).reshape((2, 3, -1)))
-
  def test_exp_mov_avg_agg(self):
  arr = np.arange(1, 11)
  val = expmov_avg_aggregator(arr, 0.9)