diff --git a/spatio_temporal_dataset/coordinates/abstract_coordinates.py b/spatio_temporal_dataset/coordinates/abstract_coordinates.py
index 0c55f3dd2338d0d5b88cdc07ebc1e86c37234151..7035c2f1217b943ca69f6f417888e6f9b466c1b3 100644
--- a/spatio_temporal_dataset/coordinates/abstract_coordinates.py
+++ b/spatio_temporal_dataset/coordinates/abstract_coordinates.py
@@ -27,7 +27,8 @@ class AbstractCoordinates(object):
# Coordinates columns
COORDINATES_NAMES = COORDINATE_SPATIAL_NAMES + [COORDINATE_T]
- def __init__(self, df_coord: pd.DataFrame, slicer_class: type, s_split_spatial: pd.Series = None, s_split_temporal: pd.Series = None):
+ def __init__(self, df_coord: pd.DataFrame, slicer_class: type, s_split_spatial: pd.Series = None,
+ s_split_temporal: pd.Series = None):
self.df_all_coordinates = df_coord # type: pd.DataFrame
self.s_split_spatial = s_split_spatial # type: pd.Series
self.s_split_temporal = s_split_temporal # type: pd.Series
@@ -37,23 +38,26 @@ class AbstractCoordinates(object):
# ClassMethod constructor
+ @classmethod
+ def from_df(cls, df: pd.DataFrame):
+ pass
+
@classmethod
def from_df_and_slicer(cls, df: pd.DataFrame, slicer_class: type, train_split_ratio: float = None):
- """
- train_split_ratio is shared between the spatial part of the data, and the temporal part
- """
+ # train_split_ratio is shared between the spatial part of the data, and the temporal part
+
# All the index should be unique
assert len(set(df.index)) == len(df)
# Create a spatial split
if slicer_class in [SpatialSlicer, SpatioTemporalSlicer]:
- s_split_spatial = s_split_from_df(df, cls.COORDINATE_X, cls.SPATIAL_SPLIT, train_split_ratio, concat=False)
+ s_split_spatial = s_split_from_df(df, cls.COORDINATE_X, cls.SPATIAL_SPLIT, train_split_ratio, True)
else:
s_split_spatial = None
# Create a temporal split
if slicer_class in [TemporalSlicer, SpatioTemporalSlicer]:
- s_split_temporal = s_split_from_df(df, cls.COORDINATE_T, cls.TEMPORAL_SPLIT, train_split_ratio, concat=True)
+ s_split_temporal = s_split_from_df(df, cls.COORDINATE_T, cls.TEMPORAL_SPLIT, train_split_ratio, False)
else:
s_split_temporal = None
@@ -71,18 +75,6 @@ class AbstractCoordinates(object):
df.set_index(index_column_name, inplace=True)
return cls.from_df(df)
- @classmethod
- def from_nb_points(cls, nb_points: int, train_split_ratio: float = None, **kwargs):
- # Call the default class method from csv
- coordinates = cls.from_csv() # type: AbstractCoordinates
- # Check that nb_points asked is not superior to the number of coordinates
- nb_coordinates = len(coordinates)
- if nb_points > nb_coordinates:
- raise Exception('Nb coordinates in csv: {} < Nb points desired: {}'.format(nb_coordinates, nb_points))
- # Sample randomly nb_points coordinates
- df_sample = pd.DataFrame.sample(coordinates.df_merged, n=nb_points)
- return cls.from_df(df=df_sample, train_split_ratio=train_split_ratio)
-
@property
def index(self):
return self.df_all_coordinates.index
diff --git a/spatio_temporal_dataset/coordinates/spatial_coordinates/abstract_spatial_coordinates.py b/spatio_temporal_dataset/coordinates/spatial_coordinates/abstract_spatial_coordinates.py
index 50a4367546f076707a00ab9766ad592bfe18db20..f78456454fb9e747f4c22a10e6368aef08bfcc4e 100644
--- a/spatio_temporal_dataset/coordinates/spatial_coordinates/abstract_spatial_coordinates.py
+++ b/spatio_temporal_dataset/coordinates/spatial_coordinates/abstract_spatial_coordinates.py
@@ -11,3 +11,15 @@ class AbstractSpatialCoordinates(AbstractCoordinates):
assert cls.COORDINATE_X in df.columns
assert cls.COORDINATE_T not in df.columns
return super().from_df_and_slicer(df, SpatialSlicer, train_split_ratio)
+
+ @classmethod
+ def from_nb_points(cls, nb_points: int, train_split_ratio: float = None, **kwargs):
+ # Call the default class method from csv
+ coordinates = cls.from_csv() # type: AbstractCoordinates
+ # Check that nb_points asked is not superior to the number of coordinates
+ nb_coordinates = len(coordinates)
+ if nb_points > nb_coordinates:
+ raise Exception('Nb coordinates in csv: {} < Nb points desired: {}'.format(nb_coordinates, nb_points))
+ # Sample randomly nb_points coordinates
+ df_sample = pd.DataFrame.sample(coordinates.df_merged, n=nb_points)
+ return cls.from_df(df=df_sample, train_split_ratio=train_split_ratio)
diff --git a/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/abstract_spatio_temporal_coordinates.py b/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/abstract_spatio_temporal_coordinates.py
index 7074c5440b42d2406423ebf72cb3d8d00d3c0dc7..651eb34970b6a3a17f6ef6a3d87e5962e7489a75 100644
--- a/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/abstract_spatio_temporal_coordinates.py
+++ b/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/abstract_spatio_temporal_coordinates.py
@@ -10,4 +10,8 @@ class AbstractSpatioTemporalCoordinates(AbstractCoordinates):
def from_df(cls, df: pd.DataFrame, train_split_ratio: float = None):
assert cls.COORDINATE_T in df.columns
assert cls.COORDINATE_X in df.columns
+ # Assert that the time steps are in the good order with respect to the coordinates
+ nb_points = len(set(df[cls.COORDINATE_X]))
+ first_time_step_for_all_points = df.iloc[:nb_points][cls.COORDINATE_T]
+ assert len(set(first_time_step_for_all_points)) == 1
return super().from_df_and_slicer(df, SpatioTemporalSlicer, train_split_ratio)
\ No newline at end of file
diff --git a/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/generated_spatio_temporal_coordinates.py b/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/generated_spatio_temporal_coordinates.py
index f2b4b648ccce456ed9025fac0297959ca15936fe..fd1aa1f374f4b6c1f00fb4cb2e5d75867aa98e11 100644
--- a/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/generated_spatio_temporal_coordinates.py
+++ b/spatio_temporal_dataset/coordinates/spatio_temporal_coordinates/generated_spatio_temporal_coordinates.py
@@ -4,22 +4,24 @@ from spatio_temporal_dataset.coordinates.abstract_coordinates import AbstractCoo
from spatio_temporal_dataset.coordinates.spatial_coordinates.coordinates_1D import UniformSpatialCoordinates
from spatio_temporal_dataset.coordinates.spatio_temporal_coordinates.abstract_spatio_temporal_coordinates import \
AbstractSpatioTemporalCoordinates
+from spatio_temporal_dataset.coordinates.temporal_coordinates.generated_temporal_coordinates import \
+ ConsecutiveTemporalCoordinates
from spatio_temporal_dataset.slicer.spatio_temporal_slicer import SpatioTemporalSlicer
class UniformSpatioTemporalCoordinates(AbstractSpatioTemporalCoordinates):
@classmethod
- def from_nb_points(cls, nb_points, train_split_ratio: float = None, nb_time_steps=1, max_radius=1.0):
- assert isinstance(nb_time_steps, int) and nb_time_steps >= 1
- df_spatial = UniformSpatialCoordinates.df_spatial(nb_points)
+ def from_nb_points_and_nb_steps(cls, nb_points, nb_steps, train_split_ratio: float = None):
+ assert isinstance(nb_steps, int) and nb_steps >= 1
+ df_spatial = UniformSpatialCoordinates.df_spatial(nb_points=nb_points)
+ # df_temporal = ConsecutiveTemporalCoordinates.df_temporal(nb_temporal_steps=nb_temporal_steps)
df_time_steps = []
- for t in range(nb_time_steps):
+ for t in range(nb_steps):
df_time_step = df_spatial.copy()
df_time_step[cls.COORDINATE_T] = t
df_time_steps.append(df_time_step)
df_time_steps = pd.concat(df_time_steps, ignore_index=True)
- print(df_time_steps)
return cls.from_df(df=df_time_steps, train_split_ratio=train_split_ratio)
diff --git a/spatio_temporal_dataset/coordinates/temporal_coordinates/generated_temporal_coordinates.py b/spatio_temporal_dataset/coordinates/temporal_coordinates/generated_temporal_coordinates.py
index 749cfc27692e0fee218d15f3775cead9e44787a1..518459725907c7a457d49bde000f787b8d7e9efd 100644
--- a/spatio_temporal_dataset/coordinates/temporal_coordinates/generated_temporal_coordinates.py
+++ b/spatio_temporal_dataset/coordinates/temporal_coordinates/generated_temporal_coordinates.py
@@ -4,7 +4,7 @@ from spatio_temporal_dataset.coordinates.temporal_coordinates.abstract_temporal_
AbstractTemporalCoordinates
-class TemporalCoordinates(AbstractTemporalCoordinates):
+class ConsecutiveTemporalCoordinates(AbstractTemporalCoordinates):
pass
@classmethod
diff --git a/spatio_temporal_dataset/slicer/split.py b/spatio_temporal_dataset/slicer/split.py
index 6e33c57f2e77bc34e4146d9ab783e79de84e27f5..564657e0905957d7a699cb834cad75106859eef2 100644
--- a/spatio_temporal_dataset/slicer/split.py
+++ b/spatio_temporal_dataset/slicer/split.py
@@ -1,4 +1,5 @@
from enum import Enum
+from typing import Union
import pandas as pd
@@ -44,7 +45,7 @@ def small_s_split_from_ratio(index: pd.Index, train_split_ratio):
return s
-def s_split_from_df(df: pd.DataFrame, column, split_column, train_split_ratio, concat):
+def s_split_from_df(df: pd.DataFrame, column, split_column, train_split_ratio, spatial_split) -> Union[None, pd.Series]:
df = df.copy() # type: pd.DataFrame
# Extract the index
if train_split_ratio is None:
@@ -55,17 +56,17 @@ def s_split_from_df(df: pd.DataFrame, column, split_column, train_split_ratio, c
raise Exception('A split has already been defined')
else:
serie = df.drop_duplicates(subset=[column], keep='first')[column]
-
assert len(df) % len(serie) == 0
multiplication_factor = len(df) // len(serie)
small_s_split = small_s_split_from_ratio(serie.index, train_split_ratio)
- if concat:
+ if spatial_split:
+ # concatenation for spatial_split
s_split = pd.concat([small_s_split for _ in range(multiplication_factor)], ignore_index=True).copy()
else:
- # dilatjon
- s_split = pd.Series(None, index=df.infer_objects())
+ # dilatjon for the temporal split
+ s_split = pd.Series(None, index=df.index)
for i in range(len(s_split)):
- s_split.iloc[i] = small_s_split.iloc[i % len(small_s_split)]
+ s_split.iloc[i] = small_s_split.iloc[i // multiplication_factor]
s_split.index = df.index
return s_split
diff --git a/test/test_spatio_temporal_dataset/test_coordinates.py b/test/test_spatio_temporal_dataset/test_coordinates.py
index a82080ef8f3aefc5867798dcc77c6c5f96db52d5..a95abdad6569564ff309c5b81932793673c84a53 100644
--- a/test/test_spatio_temporal_dataset/test_coordinates.py
+++ b/test/test_spatio_temporal_dataset/test_coordinates.py
@@ -9,7 +9,8 @@ from spatio_temporal_dataset.coordinates.spatial_coordinates.alps_station_2D_coo
AlpsStation2DCoordinatesBetweenZeroAndOne
from spatio_temporal_dataset.coordinates.spatial_coordinates.alps_station_3D_coordinates import \
AlpsStation3DCoordinatesWithAnisotropy
-from spatio_temporal_dataset.coordinates.spatial_coordinates.generated_spatial_coordinates import CircleSpatialCoordinates
+from spatio_temporal_dataset.coordinates.spatial_coordinates.generated_spatial_coordinates import \
+ CircleSpatialCoordinates
from spatio_temporal_dataset.slicer.spatio_temporal_slicer import SpatioTemporalSlicer
@@ -40,7 +41,7 @@ class TestSpatialCoordinates(unittest.TestCase):
class SpatioTemporalCoordinates(unittest.TestCase):
nb_points = 4
- nb_times_steps = 2
+ nb_steps = 2
def tearDown(self):
c = Counter([len(self.coordinates.df_coordinates(split)) for split in SpatioTemporalSlicer.SPLITS])
@@ -48,9 +49,9 @@ class SpatioTemporalCoordinates(unittest.TestCase):
self.assertTrue(good_count)
def test_temporal_circle(self):
- self.coordinates = UniformSpatioTemporalCoordinates.from_nb_points(nb_points=self.nb_points,
- nb_time_steps=self.nb_times_steps,
- train_split_ratio=0.5)
+ self.coordinates = UniformSpatioTemporalCoordinates.from_nb_points_and_nb_steps(nb_points=self.nb_points,
+ nb_steps=self.nb_steps,
+ train_split_ratio=0.5)
# def test_temporal_alps(self):
# pass
diff --git a/test/test_spatio_temporal_dataset/test_slicer.py b/test/test_spatio_temporal_dataset/test_slicer.py
index c41d4f252acb3b56ac79e0d822142228fbb0af56..6551bee311dbaf5280218a6066df5b3bc75642e3 100644
--- a/test/test_spatio_temporal_dataset/test_slicer.py
+++ b/test/test_spatio_temporal_dataset/test_slicer.py
@@ -17,8 +17,8 @@ class TestSlicerForDataset(unittest.TestCase):
super().__init__(methodName)
self.dataset = None
- nb_spatial_points = 2
- nb_temporal_steps = 2
+ nb_points = 2
+ nb_steps = 2
nb_obs = 2
@property
@@ -48,10 +48,10 @@ class TestSlicerForSpatialDataset(TestSlicerForDataset):
@property
def complete_shape(self):
- return self.nb_spatial_points, self.nb_obs
+ return self.nb_points, self.nb_obs
def load_datasets(self, train_split_ratio):
- coordinates_list = load_test_1D_and_2D_spatial_coordinates(nb_points=self.nb_spatial_points,
+ coordinates_list = load_test_1D_and_2D_spatial_coordinates(nb_points=self.nb_points,
train_split_ratio=train_split_ratio)
dataset_list = [FullSimulatedDataset.from_double_sampling(nb_obs=self.nb_obs,
margin_model=ConstantMarginModel(
@@ -63,7 +63,7 @@ class TestSlicerForSpatialDataset(TestSlicerForDataset):
def test_spatial_slicer_for_spatial_dataset(self):
train_split_ratio_to_observation_shape = {
None: self.complete_shape,
- 0.5: (self.nb_spatial_points // 2, self.nb_obs),
+ 0.5: (self.nb_points // 2, self.nb_obs),
}
self.check_shapes(train_split_ratio_to_observation_shape)
@@ -72,10 +72,10 @@ class TestSlicerForTemporalDataset(TestSlicerForDataset):
@property
def complete_shape(self):
- return self.nb_temporal_steps, self.nb_obs
+ return self.nb_steps, self.nb_obs
def load_datasets(self, train_split_ratio):
- coordinates_list = load_test_temporal_coordinates(nb_temporal_steps=self.nb_temporal_steps,
+ coordinates_list = load_test_temporal_coordinates(nb_steps=self.nb_steps,
train_split_ratio=train_split_ratio)
dataset_list = [FullSimulatedDataset.from_double_sampling(nb_obs=self.nb_obs,
margin_model=ConstantMarginModel(
@@ -87,33 +87,34 @@ class TestSlicerForTemporalDataset(TestSlicerForDataset):
def test_temporal_slicer_for_temporal_dataset(self):
ind_tuple_to_observation_shape = {
None: self.complete_shape,
- 0.5: (self.nb_temporal_steps // 2, self.nb_obs),
+ 0.5: (self.nb_steps // 2, self.nb_obs),
}
self.check_shapes(ind_tuple_to_observation_shape)
-# class TestSlicerForSpatioTemporalDataset(TestSlicerForDataset):
-#
-# def complete_shape(self):
-# return self.nb_spatial_points * self.nb_temporal_points, self.nb_obs
-#
-# def load_datasets(self, train_split_ratio):
-# coordinates_list = load_test_spatiotemporal_coordinates(nb_points=self.nb_spatial_points,
-# train_split_ratio=train_split_ratio,
-# nb_time_steps=self.nb_temporal_points)
-# dataset_list = [FullSimulatedDataset.from_double_sampling(nb_obs=self.nb_obs,
-# margin_model=ConstantMarginModel(
-# coordinates=coordinates),
-# coordinates=coordinates, max_stable_model=Smith())
-# for coordinates in coordinates_list]
-# return dataset_list
-#
-# def test_spatiotemporal_slicer_for_spatio_temporal_dataset(self):
-# ind_tuple_to_observation_shape = {
-# None: self.complete_shape,
-# 0.5: (1, 1),
-# }
-# self.check_shapes(ind_tuple_to_observation_shape)
+class TestSlicerForSpatioTemporalDataset(TestSlicerForDataset):
+
+ @property
+ def complete_shape(self):
+ return self.nb_points * self.nb_steps, self.nb_obs
+
+ def load_datasets(self, train_split_ratio):
+ coordinates_list = load_test_spatiotemporal_coordinates(nb_points=self.nb_points,
+ nb_steps=self.nb_steps,
+ train_split_ratio=train_split_ratio)
+ dataset_list = [FullSimulatedDataset.from_double_sampling(nb_obs=self.nb_obs,
+ margin_model=ConstantMarginModel(
+ coordinates=coordinates),
+ coordinates=coordinates, max_stable_model=Smith())
+ for coordinates in coordinates_list]
+ return dataset_list
+
+ def test_spatiotemporal_slicer_for_spatio_temporal_dataset(self):
+ ind_tuple_to_observation_shape = {
+ None: self.complete_shape,
+ 0.5: (self.nb_steps * self.nb_points // 4, self.nb_obs),
+ }
+ self.check_shapes(ind_tuple_to_observation_shape)
if __name__ == '__main__':
diff --git a/test/test_utils.py b/test/test_utils.py
index 00b74491b9390317842379b11710d081f66f661d..b9badf698f90457131e0760cea155b24c303ca39 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -14,7 +14,7 @@ from spatio_temporal_dataset.coordinates.spatial_coordinates.generated_spatial_c
from spatio_temporal_dataset.coordinates.spatio_temporal_coordinates.generated_spatio_temporal_coordinates import \
UniformSpatioTemporalCoordinates
from spatio_temporal_dataset.coordinates.spatial_coordinates.coordinates_1D import UniformSpatialCoordinates
-from spatio_temporal_dataset.coordinates.temporal_coordinates.generated_temporal_coordinates import TemporalCoordinates
+from spatio_temporal_dataset.coordinates.temporal_coordinates.generated_temporal_coordinates import ConsecutiveTemporalCoordinates
"""
Common objects to load for the test.
@@ -25,7 +25,7 @@ In this case, unit test (at least on the constructor) must be ensured in the tes
TEST_MAX_STABLE_MODEL = [Smith, BrownResnick, Schlather, Geometric, ExtremalT, ISchlather]
TEST_1D_AND_2D_SPATIAL_COORDINATES = [UniformSpatialCoordinates, CircleSpatialCoordinates]
TEST_3D_SPATIAL_COORDINATES = [AlpsStation3DCoordinatesWithAnisotropy]
-TEST_TEMPORAL_COORDINATES = [TemporalCoordinates]
+TEST_TEMPORAL_COORDINATES = [ConsecutiveTemporalCoordinates]
TEST_SPATIO_TEMPORAL_COORDINATES = [UniformSpatioTemporalCoordinates]
TEST_MARGIN_TYPES = [ConstantMarginModel, LinearAllParametersAllDimsMarginModel][:]
TEST_MAX_STABLE_ESTIMATOR = [MaxStableEstimator]
@@ -71,12 +71,12 @@ def load_test_3D_spatial_coordinates(nb_points):
return load_test_spatial_coordinates(nb_points, TEST_3D_SPATIAL_COORDINATES)
-def load_test_temporal_coordinates(nb_temporal_steps, train_split_ratio=None):
- return [coordinate_class.from_nb_temporal_steps(nb_temporal_steps, train_split_ratio) for coordinate_class in
+def load_test_temporal_coordinates(nb_steps, train_split_ratio=None):
+ return [coordinate_class.from_nb_temporal_steps(nb_steps, train_split_ratio) for coordinate_class in
TEST_TEMPORAL_COORDINATES]
-def load_test_spatiotemporal_coordinates(nb_points, train_split_ratio=None, nb_time_steps=None):
- return [coordinate_class.from_nb_points(nb_points=nb_points, train_split_ratio=train_split_ratio,
- nb_time_steps=nb_time_steps)
+def load_test_spatiotemporal_coordinates(nb_points, nb_steps, train_split_ratio=None):
+ return [coordinate_class.from_nb_points_and_nb_steps(nb_points=nb_points, nb_steps=nb_steps,
+ train_split_ratio=train_split_ratio)
for coordinate_class in TEST_SPATIO_TEMPORAL_COORDINATES]