fix conflict while pulling from remote

src/metrics/randomization/model_randomization.py

@@ -4,7 +4,7 @@
 
 from metrics.base import Metric
 from utils.explanations import Explanations
-from utils.functions.correlations import explanation_spearman_rank_correlation
+from utils.functions.correlations import kendall_rank_corr, spearman_rank_corr
 
 
 class ModelRandomizationMetric(Metric):
@@ -21,7 +21,9 @@ def __init__(self, correlation_measure: Union[Callable, str, None]="spearman",
         if isinstance(correlation_measure, str):
             assert correlation_measure in ["spearman"], f"Correlation measure {correlation_measure} is not implemented."
             if correlation_measure=="spearman":
-                correlation_measure=explanation_spearman_rank_correlation    
+                correlation_measure=spearman_rank_corr
+            elif correlation_measure=="kendall":
+                correlation_measure=kendall_rank_corr
         assert isinstance(Callable,correlation_measure) 
         self.correlation_measure=correlation_measure
 

src/utils/functions/correlations.py

@@ -0,0 +1,11 @@
+from torchmetrics.functional.regression import kendall_rank_corrcoef, spearman_corrcoef
+
+
+# torchmetrics wants the independent realizations to be the final dimension
+# we transpose inputs before passing so that it is straightforward to pass explanations
+# and use these funcitons in evaluation metrics
+def kendall_rank_corr(tensor1,tensor2):
+    return kendall_rank_corrcoef(tensor1.T,tensor2.T)
+
+def spearman_rank_corr(tensor1, tensor2):
+    return spearman_corrcoef(tensor1.T,tensor2.T)

tests/metrics/test_randomization_metrics.py

@@ -1,29 +1,62 @@
 import pytest
 import torch
 
-from metrics.randomization.mprt import MPRTMetric
+from utils.explanations import TensorExplanations
+from utils.functions.correlations import spearman_rank_corr
+from metrics.randomization.model_randomization import ModelRandomizationMetric
 
 
-@pytest.mark.randomization
+@pytest.mark.randomize
 @pytest.mark.parametrize(
     "model",
     [
         ("load_mnist_model"),
     ],
 )
-def parameter_randomization_test(model, request):
+def model_randomization_test(model, request):
     model1 = request.getfixturevalue(model)
     model2 = request.getfixturevalue(model)
     gen = torch.Generator()
     gen.manual_seed(42)
-    MPRTMetric._randomize_model(model2, gen)
+    ModelRandomizationMetric._randomize_model(model2, gen)
     for param1, param2 in zip(model1.parameters(), model2.parameters()):
         assert torch.norm(param1.data - param2.data) > 1e3  # norm of the difference in parameters should be significant
 
-
-@pytest.mark.parametrize()
-def model_randomization_test():
-    assert torch.__version__ == "2.0.0"
+@pytest.mark.randomize
+def reproducibility_test():
+    assert torch.__version__=="2.0.0"
     gen = torch.Generator()
     gen.manual_seed(42)
-    assert torch.all(torch.rand(5, generator=gen) == torch.Tensor([0.8823, 0.9150, 0.3829, 0.9593, 0.3904]))
+    assert torch.all(torch.rand(5,generator=gen)==torch.Tensor([0.8823, 0.9150, 0.3829, 0.9593, 0.3904]))
+
+@pytest.mark.randomize
+def kendall_metric_test():
+    def explain_fn(model):
+        xpl_tensor=torch.tensor([[1,2,3,4],[4,3,2,1]])
+        return TensorExplanations(xpl_tensor)
+
+    xpl_tensor=torch.tensor([[1,2,3,4],[1,2,3,4]])
+    metric=ModelRandomizationMetric(correlation_measure="kendall")
+    assert torch.all(metric["rank_correlations"]==torch.tensor([1.,-1.]))
+
+@pytest.mark.randomize
+@pytest.mark.parametrize(
+    "model",
+    [
+        ("load_mnist_model"),
+    ],
+)
+def spearman_metric_test(model,request):
+    def explain_fn(model):
+        xpl_tensor=torch.tensor([[1,2,3,4],[4,3,2,1]])
+        return TensorExplanations(xpl_tensor)
+
+    def corr_measure(tensor1,tensor2):
+        return spearman_rank_corr(tensor1,tensor2)
+    model=request.getfixturevalue(model)
+    xpl_tensor=torch.tensor([[1,2,3,4],[1,2,3,4]])
+    for corr_measure in ["spearman", "kendall", corr_measure]:
+        metric=ModelRandomizationMetric(correlation_measure=corr_measure)
+        metric=metric(
+            model,"0","",None,None,xpl_tensor,explain_fn,{})
+        assert torch.all(metric["rank_correlations"]==torch.tensor([1.,-1.]))