Wasserstein distances (#500)

* refactor distance submodule

* cont

* implement wasserstein distances

* fix style

* install cython

* fix2

* add raise test

doc/api.rst

@@ -5,7 +5,7 @@ API reference
 .. automodule:: pyabc
 
 .. toctree::
-   :maxdepth: 2
+   :maxdepth: 3
 
    api_inference
    api_distance

doc/examples.rst

@@ -34,6 +34,7 @@ Potentially, further dependencies may be required.
    examples/adaptive_distances.ipynb
    examples/informative.ipynb
    examples/aggregated_distances.ipynb
+   examples/wasserstein.ipynb
    examples/external_simulators.ipynb
    examples/data_plots.ipynb
    examples/noise.ipynb
@@ -58,6 +59,7 @@ Download the examples as notebooks
 * :download:`Adaptive distances <examples/adaptive_distances.ipynb>`
 * :download:`Informative distances and summary statistics <examples/informative.ipynb>`
 * :download:`Aggregated distances <examples/aggregated_distances.ipynb>`
+* :download:`Wasserstein distances <examples/wasserstein.ipynb>`
 * :download:`External simulators <examples/external_simulators.ipynb>`
 * :download:`Data plots <examples/data_plots.ipynb>`
 * :download:`Measurement noise and exact inference <examples/noise.ipynb>`

doc/examples/informative.ipynb

@@ -31,13 +31,13 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "id": "df5d57e9-6819-4d6a-bdbb-8639f315602a",
    "metadata": {},
    "outputs": [],
    "source": [
     "# install if not done yet\n",
-    "!pip install pyabc --quiet"
+    "!pip install pyabc[plotly] --quiet"
    ]
   },
   {

doc/index.rst

@@ -70,7 +70,7 @@ If you use it in your work, you can cite the paper:
 
 
 .. toctree::
-   :maxdepth: 2
+   :maxdepth: 3
    :caption: API reference
 
    api

pyabc/__init__.py

@@ -39,6 +39,8 @@
     PercentileDistance,
     RangeEstimatorDistance,
     DistanceWithMeasureList,
+    WassersteinDistance,
+    SlicedWassersteinDistance,
     StochasticKernel,
     NormalKernel,
     IndependentNormalKernel,

pyabc/distance/__init__.py

@@ -1,10 +1,10 @@
 """
-Distance functions
-==================
+Distances
+=========
 
-Distance functions measure closeness of observed and sampled data. This
-module implements various commonly used distance functions for ABC, featuring
-a few advanced concepts.
+Distance functions or metrics measure closeness of observed and sampled data.
+This module implements various commonly used distance functions for ABC,
+featuring a few advanced concepts.
 
 For custom distance functions, either pass a plain function to ABCSMC, or
 subclass the pyabc.Distance class.
@@ -34,6 +34,10 @@
     AggregatedDistance,
     AdaptiveAggregatedDistance,
 )
+from .ot import (
+    WassersteinDistance,
+    SlicedWassersteinDistance,
+)
 from .scale import (
     median_absolute_deviation,
     mad,

pyabc/distance/distance.py

@@ -7,6 +7,7 @@
 import logging
 
 from ..population import Sample
+
 from .base import Distance
 
 
@@ -20,7 +21,6 @@ class DistanceWithMeasureList(Distance, ABC):
 
     Parameters
     ----------
-
     measures_to_use: Union[str, List[str]].
         * If set to "all", all measures are used. This is the default.
         * If a list is provided, the measures in the list are used.
@@ -167,13 +167,11 @@ def lower(parameter_list: List[float]):
 
         Parameters
         ----------
-
         parameter_list: List[float]
             List of values of a parameter.
 
         Returns
         -------
-
         lower_margin: float
             The lower margin of the range calculated from these parameters
         """
@@ -185,13 +183,11 @@ def upper(parameter_list: List[float]):
 
         Parameters
         ----------
-
         parameter_list: List[float]
             List of values of a parameter.
 
         Returns
         -------
-
         upper_margin: float
             The upper margin of the range calculated from these parameters
         """