add cross validation data set

floriangriese · Sep 28, 2022 · 34a6ad2 · 34a6ad2
1 parent 9135455
commit 34a6ad2
Show file tree

Hide file tree

Showing 9 changed files with 63 additions and 14 deletions.
diff --git a/Example_firstgalaxydata.py b/Example_firstgalaxydata.py
@@ -2,21 +2,47 @@
 import torchvision.transforms as transforms
 
 
-
 if __name__ == "__main__":
     transformRGB = transforms.Compose(
         [transforms.ToTensor(),
          transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])])
 
+    # Basic usage of splitting train, valid and test
+
     data = FIRSTGalaxyData(root="./", selected_split="train", input_data_list=["galaxy_data_h5.h5"],
                            is_PIL=True, is_RGB=True, transform=transformRGB)
 
+    print(data)
+    data = FIRSTGalaxyData(root="./", selected_split="valid", input_data_list=["galaxy_data_h5.h5"],
+                           is_PIL=True, is_RGB=True, transform=transformRGB)
+
     print(data)
 
     data = FIRSTGalaxyData(root="./", selected_split="test", input_data_list=["galaxy_data_h5.h5"],
                            selected_classes=["FRI", "FRII", "Compact", "Bent"],
                            selected_catalogues=["Gendre", "MiraBest", "Capetti2017a", "Capetti2017b",
                                                 "Baldi2018", "Proctor_Tab1"],
                            is_PIL=True, is_RGB=True, transform=transformRGB)
+    print(data)
+
+    # Usage of 5-fold cross validation set
+
+    data = FIRSTGalaxyData(root="./", selected_split="train", input_data_list=["galaxy_data_crossvalid_0_h5.h5"],
+                           is_PIL=True, is_RGB=True, transform=transformRGB)
+
+    print(data)
+
+    data = FIRSTGalaxyData(root="./", selected_split="valid", input_data_list=["galaxy_data_crossvalid_0_h5.h5"],
+                           is_PIL=True, is_RGB=True, transform=transformRGB)
+
+    print(data)
+
+    data = FIRSTGalaxyData(root="./", selected_split="test", input_data_list=["galaxy_data_crossvalid_test_h5.h5"],
+                           selected_classes=["FRI", "FRII", "Compact", "Bent"],
+                           selected_catalogues=["Gendre", "MiraBest", "Capetti2017a", "Capetti2017b",
+                                                "Baldi2018", "Proctor_Tab1"],
+                           is_PIL=True, is_RGB=True, transform=transformRGB)
+    print(data)
+
+
 
-    print(data)
diff --git a/README.md b/README.md
@@ -13,9 +13,25 @@ Further, the following catalogues are included in this dataset:
 * Baldi 2018 [[7]](https://www.aanda.org/articles/aa/full_html/2018/01/aa31333-17/aa31333-17.html), [Table](https://www.aanda.org/articles/aa/full_html/2018/01/aa31333-17/T1.html)
 * Proctor [[8]](https://ui.adsabs.harvard.edu/abs/2011ApJS..194...31P/abstract), [Table](https://iopscience.iop.org/article/10.1088/0067-0049/194/2/31#apjs390184t1), data from Table 1 with label “WAT” and “NAT”
 
-When using the literature class definition of FRI, FRII, Compact and Bent as schown below, 
+Examples for the class definitions of FRI, FRII, Compact and Bent are shown below, 
 ![image](img/Classification_Scheme.png)
-the dataset contains the following number of samples per class.
+with the labels
+
+| classes    | Label |  
+| ----------- | ----------- |
+| FRI    | 0  |
+| FRII   | 1  | 
+| Compact| 2  | 
+| Bent   | 3  |
+
+The dataset has the following total number of samples per class.
+
+| classes/split     | FRI |   FRII |     Compact |    Bent |   Total     |
+| ----------- | ----------- |----------- |----------- |-----------       |-----------|
+| total   | 495        |924       |391       |348       |2158       |
+
+We provide two splitting options for the dataset. 
+The first splitting option (galaxy_data_h5.zip) provides three splittings in train, valid and test with the following number of sample per class.
 
 | classes/split     | FRI |   FRII |     Compact |    Bent |   Total     |
 | ----------- | ----------- |----------- |----------- |-----------       |-----------|
@@ -24,12 +40,14 @@ the dataset contains the following number of samples per class.
 | test   | 50        | 50       | 50       | 50      |200       |
 | total   | 495        |924       |391       |348       |2158       |
 
-| classes    | Label |  
-| ----------- | ----------- |
-| FRI    | 0  |
-| FRII   | 1  | 
-| Compact| 2  | 
-| Bent   | 3  |
+The second splitting option (galaxy_data_crossvalid_0_h5.zip to galaxy_data_crossvalid_4_h5.zip and galaxy_data_crossvalid_test_h5.zip) provides a 5-fold cross validation dataset with a larger test set.
+
+| classes/split     | FRI |   FRII |     Compact |    Bent |   Total     |
+| ----------- | ----------- |----------- |----------- |-----------       |-----------|
+| 5-fold cross train   | 316        | 659       | 232       | 198      |1405       |
+| 5-fold cross valid   | 79        | 165       | 59       | 50      |353       |
+| test   | 100        | 100       | 100       | 100      |400       |
+| total   | 495        |924       |391       |348       |2158       |
 
 # Installation usage with pytorch
 If you want to use the dataset via the dataset class `FIRSTGalaxyData` with pytorch, install the necessary packages with

diff --git a/firstgalaxydata.py b/firstgalaxydata.py
@@ -13,7 +13,7 @@
 
 class FIRSTGalaxyData(data.Dataset):
     """
-    FIRSTGalaxyData class provides FIRST images from various different data catalogs
+    FIRSTGalaxyData class provides FIRST images with labels from various different data catalogs
 
     Attributes
     ----------
@@ -39,12 +39,18 @@ class FIRSTGalaxyData(data.Dataset):
     show_coords():
         shows the coordinates of the images in a Aitoff projection
     __repr__()
-        presents import information aboout the dataset in the Repl
+        presents import information about the dataset in the Repl
      """
 
     urls = {
         "galaxy_data.zip": "https://syncandshare.desy.de/index.php/s/yWNQEoCxbpwxCWm/download",
-        "galaxy_data_h5.zip": "https://syncandshare.desy.de/index.php/s/9kLKJGxwARZdfiZ/download"
+        "galaxy_data_h5.zip": "https://syncandshare.desy.de/index.php/s/9kLKJGxwARZdfiZ/download",
+        "galaxy_data_crossvalid_0_h5.zip": "https://syncandshare.desy.de/index.php/s/QPjZBZE4WeCPDxg/download",
+        "galaxy_data_crossvalid_1_h5.zip": "https://syncandshare.desy.de/index.php/s/8iDGAAqxtdM2tj9/download",
+        "galaxy_data_crossvalid_2_h5.zip": "https://syncandshare.desy.de/index.php/s/j6gKn8HiMa8nB4J/download",
+        "galaxy_data_crossvalid_3_h5.zip": "https://syncandshare.desy.de/index.php/s/oSz6WdWtnZq4s8M/download",
+        "galaxy_data_crossvalid_4_h5.zip": "https://syncandshare.desy.de/index.php/s/qoNsik3sF39cRBG/download",
+        "galaxy_data_crossvalid_test_h5.zip": "https://syncandshare.desy.de/index.php/s/aM9eGcr8ydddCNM/download"
     }
 
     def __init__(self, root, input_data_list=None, selected_split="train", selected_classes=None,
@@ -161,7 +167,6 @@ def __getitem__(self, index):
             img = Image.fromarray(img, mode="L")
             if self.is_RGB:
                 img = img.convert("RGB")
-        # else...return numpy array directly
 
         if self.transform is not None:
             img = self.transform(img)

diff --git a/galaxy_data_crossvalid_0_h5.zip b/galaxy_data_crossvalid_0_h5.zip
diff --git a/galaxy_data_crossvalid_1_h5.zip b/galaxy_data_crossvalid_1_h5.zip
diff --git a/galaxy_data_crossvalid_2_h5.zip b/galaxy_data_crossvalid_2_h5.zip
diff --git a/galaxy_data_crossvalid_3_h5.zip b/galaxy_data_crossvalid_3_h5.zip
diff --git a/galaxy_data_crossvalid_4_h5.zip b/galaxy_data_crossvalid_4_h5.zip
diff --git a/galaxy_data_crossvalid_test_h5.zip b/galaxy_data_crossvalid_test_h5.zip