Merge pull request #192 from yfukai/refine_logic

Refine logic for outlier filter

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "m2stitch"
-version = "0.5.2"
+version = "0.6.0"
 description = "M2Stitch"
 authors = ["Yohsuke Fukai <[email protected]>"]
 license = "MIT"

src/m2stitch/_stage_model.py

@@ -1,16 +1,16 @@
 import itertools
+from typing import Callable
 from typing import Tuple
 
 import numpy as np
 import pandas as pd
-from sklearn.base import BaseEstimator
 
 from ._typing_utils import Float
 from ._typing_utils import Int
 
 
 def compute_image_overlap2(
-    grid: pd.DataFrame, direction: str, sizeY: Int, sizeX: Int, predictor: BaseEstimator
+    grid: pd.DataFrame, direction: str, sizeY: Int, sizeX: Int, predictor: Callable
 ) -> Tuple[Float, ...]:
     """Compute the value of the image overlap.
 
@@ -42,7 +42,7 @@ def compute_image_overlap2(
         ]
     )
     translation = translation[:, np.all(np.isfinite(translation), axis=0)]
-    c = predictor.fit_predict(translation.T)
+    c = predictor(translation.T)
     res = np.median(translation[:, c == 1], axis=1)
     assert len(res) == 2
     return tuple(res)

src/m2stitch/_typing_utils.py

@@ -7,6 +7,7 @@
 NumArray = npt.NDArray[Any]
 FloatArray = npt.NDArray[np.float_]
 IntArray = npt.NDArray[np.int_]
+BoolArray = npt.NDArray[np.bool_]
 
 Int = Union[int, np.int_]
 Float = Union[float, np.float_]

src/m2stitch/stitching.py

@@ -1,6 +1,7 @@
 """This module provides microscope image stitching with the algorithm by MIST."""
 import itertools
 import warnings
+from dataclasses import dataclass
 from typing import Any
 from typing import Optional
 from typing import Sequence
@@ -23,10 +24,24 @@
 from ._translation_computation import interpret_translation
 from ._translation_computation import multi_peak_max
 from ._translation_computation import pcm
+from ._typing_utils import BoolArray
 from ._typing_utils import Float
 from ._typing_utils import NumArray
 
 
+@dataclass
+class ElipticEnvelopPredictor:
+    contamination: float
+    epsilon: float
+    random_seed: int
+
+    def __call__(self, X: NumArray) -> BoolArray:
+        ee = EllipticEnvelope(contamination=self.contamination)
+        rng = np.random.default_rng(self.random_seed)
+        X = rng.normal(size=X.shape) * self.epsilon + X
+        return ee.fit_predict(X) > 0
+
+
 def stitch_images(
     images: Union[Sequence[NumArray], NumArray],
     rows: Optional[Sequence[Any]] = None,
@@ -173,10 +188,10 @@ def get_lims(dimension, size):
             for j, key in enumerate(["ncc", "y", "x"]):
                 grid.loc[i2, f"{direction}_{key}_first"] = max_peak[j]
 
-    predictor = EllipticEnvelope(contamination=0.4)
     # TODO make threshold adjustable
     assert np.any(grid["top_ncc_first"] > 0.5), "there is no good top pair"
     assert np.any(grid["left_ncc_first"] > 0.5), "there is no good left pair"
+    predictor = ElipticEnvelopPredictor(contamination=0.4, epsilon=0.01, random_seed=0)
     left_displacement = compute_image_overlap2(
         grid[grid["left_ncc_first"] > 0.5], "left", sizeY, sizeX, predictor
     )