Added module for handling training/validation/test dataset split

pysegcnn/core/split.py

+# -*- coding: utf-8 -*-
+"""
+Created on Wed Jul 29 12:02:32 2020
+
+@author: Daniel
+"""
+# builtins
+import datetime
+
+# externals
+import numpy as np
+from torch.utils.data.dataset import Subset
+
+
+# function calculating number of samples in a dataset given a ratio
+def _ds_len(ds, ratio):
+    return int(np.round(len(ds) * ratio))
+
+
+# randomly split the tiles of a dataset across the training, validation and
+# test dataset
+# for each scene, the tiles can be distributed among the training, validation
+# and test set
+def random_tile_split(ds, tvratio, ttratio=1, seed=0):
+
+    # set the random seed for reproducibility
+    np.random.seed(seed)
+
+    # randomly permute indices to access dataset
+    indices = np.random.permutation(len(ds))
+
+    # length of the training and validation dataset
+    # number of samples: (ttratio * len(ds))
+    trav_len = _ds_len(indices, ttratio)
+    trav_indices = indices[:trav_len]
+
+    # length of the training dataset
+    # number of samples: (ttratio * tvratio * len(ds))
+    train_len = _ds_len(trav_indices, tvratio)
+    train_indices = trav_indices[:train_len]
+
+    # length of the validation dataset
+    # number of samples: (ttratio * (1- tvratio) * len(ds))
+    valid_indices = trav_indices[train_len:]
+
+    # length of the test dataset
+    # number of samples: ((1 - ttratio) * len(ds))
+    test_indices = indices[trav_len:]
+
+    # get the tiles of the scenes of each dataset
+    subsets = []
+    for dataset in [train_indices, valid_indices, test_indices]:
+
+        # build the subset: store the scenes
+        sbst = Subset(dataset=ds, indices=list(dataset))
+        sbst.scenes = [ds.scenes[i] for i in dataset]
+
+        # add to list of subsets
+        subsets.append(sbst)
+
+    # check if the splits are disjoint
+    assert pairwise_disjoint([s.indices for s in subsets])
+
+    return subsets
+
+
+# randomly split the tiles of a dataset across the training, validation and
+# test dataset
+# for each scene, all the tiles of the scene are included in either the
+# training set, the validation set or the test set, respectively
+def random_scene_split(ds, tvratio, ttratio=1, seed=0):
+
+    # set the random seed for reproducibility
+    np.random.seed(seed)
+
+    # get the names of the scenes and generate random permutation
+    scene_ids = np.random.permutation(np.unique([s['id'] for s in ds.scenes]))
+
+    # the training and validation scenes
+    # number of samples: (ttratio * nscenes)
+    trav_len = _ds_len(scene_ids, ttratio)
+    trav_scenes = scene_ids[:trav_len]
+
+    # the training scenes
+    # number of samples: (ttratio * tvratio * nscenes)
+    train_len = _ds_len(trav_scenes, tvratio)
+    train_scenes = trav_scenes[:train_len]
+
+    # the validation scenes
+    # number of samples: (ttratio * (1- tvratio) * nscenes)
+    valid_scenes = trav_scenes[train_len:]
+
+    # the test scenes
+    # number of samples:((1 - ttratio) * nscenes)
+    test_scenes = scene_ids[trav_len:]
+
+    # get the tiles of the scenes of each dataset
+    subsets = []
+    for dataset in [train_scenes, valid_scenes, test_scenes]:
+        # the indices of the scenes in the dataset
+        indices = []
+        tiles = []
+
+        # iterate over the scenes of the whole dataset
+        for i, scene in enumerate(ds.scenes):
+            if scene['id'] in dataset:
+                indices.append(i)
+                tiles.append(scene)
+
+        # build the subset: store scene ids
+        sbst = Subset(dataset=ds, indices=indices)
+        sbst.scenes = tiles
+        sbst.ids = dataset
+
+        # add to list of subsets
+        subsets.append(sbst)
+
+    # check if the splits are disjoint
+    assert pairwise_disjoint([s.indices for s in subsets])
+
+    return subsets
+
+
+# split the scenes of a dataset based on a date, useful for time series data
+# scenes before date build the training set, scenes after the date build the
+# validation set, the test set is empty
+def date_scene_split(ds, date, dateformat='%Y%m%d'):
+
+    # convert date to datetime object
+    date = datetime.datetime.strptime(date, dateformat)
+
+    # the training, validation and test scenes
+    train_scenes = {i: s for i, s in enumerate(ds.scenes) if s['date'] <= date}
+    valid_scenes = {i: s for i, s in enumerate(ds.scenes) if s['date'] > date}
+    test_scenes = {}
+
+    # build the training and test datasets
+    subsets = []
+    for scenes in [train_scenes, valid_scenes, test_scenes]:
+        # build the subset: store the scenes
+        sbst = Subset(dataset=ds, indices=list(scenes.keys()))
+        sbst.scenes = list(scenes.values())
+        sbst.ids = np.unique([s['id'] for s in scenes.values()])
+
+        # add to list of subsets
+        subsets.append(sbst)
+
+    # check if the splits are disjoint
+    assert pairwise_disjoint([s.indices for s in subsets])
+
+    return subsets
+
+
+def pairwise_disjoint(sets):
+    union = set().union(*sets)
+    n = sum(len(u) for u in sets)
+    return n == len(union)