From b61ac1fbd639f482b7ecb661da47254695139299 Mon Sep 17 00:00:00 2001
From: Avik Basu <3485425+ab93@users.noreply.github.com>
Date: Mon, 27 Mar 2023 16:37:42 -0700
Subject: [PATCH] fix: Tanhscaler nan output for constant feature (#153)

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>
---
 numalogic/preprocess/transformer.py  | 18 +++++++++++++++---
 pyproject.toml                       |  2 +-
 tests/preprocess/test_transformer.py | 20 ++++++++++++++++++++
 3 files changed, 36 insertions(+), 4 deletions(-)

diff --git a/numalogic/preprocess/transformer.py b/numalogic/preprocess/transformer.py
index c2e79256..0fb70175 100644
--- a/numalogic/preprocess/transformer.py
+++ b/numalogic/preprocess/transformer.py
@@ -54,7 +54,7 @@ def inverse_transform(self, X) -> ArrayLike:
 
 class TanhScaler(OneToOneFeatureMixin, TransformerMixin):
     r"""
-    Tanh Estimator applies tanh normalization to the Z-score,
+    Tanh Estimator applies column-wise tanh normalization to the Z-score,
     and scales the values between 0 and 1.
 
     After scaling, the data has a mean of 0.5.
@@ -63,21 +63,29 @@ class TanhScaler(OneToOneFeatureMixin, TransformerMixin):
     Higher the value, the linear portion of the curve will have a higher slope
     but will reach the asymptote (flatten out) earlier.
 
+    Args:
+        coeff: float value determining the spread of the scores
+        eps: minimum value below which the feature will be treated as constant.
+             In order to avoid division by zero or a very small number,
+             standard deviation will be set as 1 for that feature.
+
     References:
         Nandakumar, Jain, Ross. 2005. Score Normalization in
         Multimodal Biometric Systems, Pattern Recognition 38, 2270-2285.
         https://web.cse.msu.edu/~rossarun/pubs/RossScoreNormalization_PR05.pdf
     """
-    __slots__ = ("_coeff", "_std", "_mean")
+    __slots__ = ("_coeff", "_std", "_mean", "_eps")
 
-    def __init__(self, coeff: float = 0.2):
+    def __init__(self, coeff: float = 0.2, eps: float = 1e-10):
         self._coeff = coeff
         self._std = None
         self._mean = None
+        self._eps = eps
 
     def fit(self, x: npt.NDArray[float]) -> Self:
         self._mean = np.mean(x, axis=0)
         self._std = np.std(x, axis=0)
+        self._check_if_constant(x)
         return self
 
     def transform(self, x: npt.NDArray[float]) -> npt.NDArray[float]:
@@ -86,3 +94,7 @@ def transform(self, x: npt.NDArray[float]) -> npt.NDArray[float]:
 
     def fit_transform(self, x: npt.NDArray[float], y=None, **_) -> npt.NDArray[float]:
         return self.fit(x).transform(x)
+
+    def _check_if_constant(self, x: npt.NDArray[float]) -> None:
+        delta = np.max(x, axis=0) - np.min(x, axis=0)
+        self._std[delta < self._eps] = 1.0
diff --git a/pyproject.toml b/pyproject.toml
index e6b5b6c2..83c8be6d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "numalogic"
-version = "0.3.6"
+version = "0.3.7"
 description = "Collection of operational Machine Learning models and tools."
 authors = ["Numalogic Developers"]
 packages = [{ include = "numalogic" }]
diff --git a/tests/preprocess/test_transformer.py b/tests/preprocess/test_transformer.py
index ac1fe094..6ef61b4a 100644
--- a/tests/preprocess/test_transformer.py
+++ b/tests/preprocess/test_transformer.py
@@ -42,6 +42,26 @@ def test_tanh_scaler_2(self):
         assert_array_less(x_scaled, np.ones_like(x_scaled))
         assert_array_less(np.zeros_like(x_scaled), x_scaled)
 
+    def test_tanh_scaler_3(self):
+        x = np.random.randn(5, 3)
+        x[:, 1] = np.zeros(5)
+
+        scaler = TanhScaler()
+
+        x_scaled = scaler.fit_transform(x)
+        self.assertFalse(np.isnan(x_scaled[:, 1]).all())
+        assert_array_less(x_scaled, np.ones_like(x_scaled))
+        assert_array_less(np.zeros_like(x_scaled), x_scaled)
+
+    def test_tanh_scaler_nan(self):
+        x = np.random.randn(5, 3)
+        x[:, 1] = np.zeros(5)
+
+        scaler = TanhScaler(eps=0.0)
+
+        x_scaled = scaler.fit_transform(x)
+        self.assertTrue(np.isnan(x_scaled[:, 1]).all())
+
 
 if __name__ == "__main__":
     unittest.main()