v0.9.95

README.md

@@ -38,15 +38,13 @@
 
 ## News
 
+**December 10**: v0.9.95 includes a new tuple miner, [BatchEasyHardMiner](https://kevinmusgrave.github.io/pytorch-metric-learning/miners/#batcheasyhardminer). [Release notes](https://github.com/KevinMusgrave/pytorch-metric-learning/releases/tag/v0.9.95)
+
 **November 6**: v0.9.94 has minor bug fixes and improvements. [Release notes](https://github.com/KevinMusgrave/pytorch-metric-learning/releases/tag/v0.9.94)
 
 **October 6**: v0.9.93 is a small update: 
  - ```get_random_triplet_indices``` has been optimized, so if you were using DistanceWeightedMiner, or if you ever set the ```triplets_per_anchor``` argument to something other than ```"all"``` anywhere in your code, it should run a lot faster now.
 
-**September 14**: v0.9.92 comes with new features ([release notes](https://github.com/KevinMusgrave/pytorch-metric-learning/releases/tag/v0.9.92)). The main highlights are:
-* utils.distributed for multiprocessing.
-* Improvements to CrossBatchMemory which make it usable in MoCo-style self-supervised training. Check out the [MoCo on CIFAR10](https://github.com/KevinMusgrave/pytorch-metric-learning/tree/master/examples#simple-examples) notebook to see how it works.
-
 ## Documentation
 - [**View the documentation here**](https://kevinmusgrave.github.io/pytorch-metric-learning/)
 - [**View the installation instructions here**](https://github.com/KevinMusgrave/pytorch-metric-learning#installation)
@@ -360,11 +358,12 @@ guide. Proceed to `pip install -e .[dev]` afterwards.
 Thanks to the contributors who made pull requests!
 
 #### Algorithm implementations + useful features
-- [AlenUbuntu](https://github.com/AlenUbuntu)
-	- [CircleLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#circleloss)
 - [marijnl](https://github.com/marijnl)
+	- [BatchEasyHardMiner](https://kevinmusgrave.github.io/pytorch-metric-learning/miners/#batcheasyhardminer)
     - [TwoStreamMetricLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/trainers/#twostreammetricloss)
     - [GlobalTwoStreamEmbeddingSpaceTester](https://kevinmusgrave.github.io/pytorch-metric-learning/testers/#globaltwostreamembeddingspacetester)
+- [AlenUbuntu](https://github.com/AlenUbuntu)
+	- [CircleLoss](https://kevinmusgrave.github.io/pytorch-metric-learning/losses/#circleloss)
 - [btseytlin](https://github.com/btseytlin)
     - ```get_nearest_neighbors``` in [InferenceModel](https://kevinmusgrave.github.io/pytorch-metric-learning/inference_models)
 - [JohnGiorgi](https://github.com/JohnGiorgi)
@@ -381,6 +380,8 @@ Thanks to the contributors who made pull requests!
 #### General improvements and bug fixes
 - [wconnell](https://github.com/wconnell)
 - [marijnl](https://github.com/marijnl)
+- [thinline72](https://github.com/thinline72)
+- [tpanum](https://github.com/tpanum)
 - [fralik](https://github.com/fralik)
 - [JoOkuma](https://github.com/JoOkuma)
 

setup.py

@@ -14,7 +14,7 @@
 
 setuptools.setup(
     name="pytorch-metric-learning",
-    version="0.9.95.dev2",
+    version="0.9.95",
     author="Kevin Musgrave",
     author_email="[email protected]",
     description="The easiest way to use deep metric learning in your application. Modular, flexible, and extensible. Written in PyTorch.",

src/pytorch_metric_learning/__init__.py

@@ -1 +1 @@
-__version__ = "0.9.95.dev2"
+__version__ = "0.9.95"

src/pytorch_metric_learning/miners/batch_easy_hard_miner.py

@@ -194,4 +194,4 @@ def set_stats(self, positive_dists, negative_dists):
     def get_func_for_stats(self, min_if_inverted):
         if min_if_inverted:
             return torch.min if self.distance.is_inverted else torch.max
-        return torch.max if self.distance.is_inverted else torch.min
+        return torch.max if self.distance.is_inverted else torch.min

tests/miners/test_batch_easy_hard_miner.py

@@ -32,9 +32,7 @@ def setUpClass(self):
                 "easiest_neg_pair": 3,
                 "hardest_neg_pair": 2,
                 "expected": {
-                    "correct_a": torch.LongTensor([0, 7, 8]).to(
-                        TEST_DEVICE
-                    ),
+                    "correct_a": torch.LongTensor([0, 7, 8]).to(TEST_DEVICE),
                     "correct_p": [
                         torch.LongTensor([1, 6, 6]).to(TEST_DEVICE),
                         torch.LongTensor([1, 8, 6]).to(TEST_DEVICE),
@@ -58,12 +56,8 @@ def setUpClass(self):
                 "easiest_neg_pair": 7,
                 "hardest_neg_pair": 4,
                 "expected": {
-                    "correct_a": torch.LongTensor([0, 1, 6, 7, 8]).to(
-                        TEST_DEVICE
-                    ),
-                    "correct_p": [
-                        torch.LongTensor([4, 4, 2, 2, 2]).to(TEST_DEVICE)
-                    ],
+                    "correct_a": torch.LongTensor([0, 1, 6, 7, 8]).to(TEST_DEVICE),
+                    "correct_p": [torch.LongTensor([4, 4, 2, 2, 2]).to(TEST_DEVICE)],
                     "correct_n": [
                         torch.LongTensor([5, 5, 1, 1, 1]).to(TEST_DEVICE),
                     ],
@@ -319,15 +313,11 @@ def test_strategy_assertion(self):
         )
         self.assertRaises(
             ValueError,
-            lambda: BatchEasyHardMiner(
-                pos_strategy="all", neg_strategy="semihard"
-            ),
+            lambda: BatchEasyHardMiner(pos_strategy="all", neg_strategy="semihard"),
         )
         self.assertRaises(
             ValueError,
-            lambda: BatchEasyHardMiner(
-                pos_strategy="semihard", neg_strategy="all"
-            ),
+            lambda: BatchEasyHardMiner(pos_strategy="semihard", neg_strategy="all"),
         )
 
     def helper(self, a1, p, a2, n, gt):
@@ -348,4 +338,4 @@ def tearDown(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main()

tests/miners/test_batch_hard_miner.py

@@ -134,4 +134,4 @@ def test_empty_output(self):
                 self.assertTrue(len(n) == 0)
                 self.assertTrue(miner.hardest_pos_pair == 0)
                 self.assertTrue(miner.hardest_neg_pair == 0)
-                self.assertTrue(miner.hardest_triplet == 0)
+                self.assertTrue(miner.hardest_triplet == 0)