tests and strengthened conversions

agrinet/tests/testMetrics.py

@@ -0,0 +1,52 @@
+import sys
+import os
+
+# fixes "ModuleNotFoundError: No module named 'utils'"
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
+
+# flake8: noqa
+import unittest
+import numpy as np
+import pandas as pd
+from utils.Metrics import CGANMetrics
+
+
+class TestCGANMetrics(unittest.TestCase):
+    def setUp(self):
+        self.cgan_metrics = CGANMetrics()
+
+    def tearDown(self):
+        del self.cgan_metrics
+
+    def test_psnr(self):
+        x = np.random.randint(0, 255, size=(256, 256, 3)).astype(np.float32)
+        y = np.random.randint(0, 255, size=(256, 256, 3)).astype(np.float32)
+        psnr_score = self.cgan_metrics.psnr(x, y)
+        self.assertIsInstance(psnr_score, float)
+
+    def test_mmd(self):
+        x = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        y = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        mmd_value = self.cgan_metrics.mmd(x, y)
+        self.assertIsInstance(mmd_value, float)
+
+    def test_update(self):
+        disc_out = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        gen_out = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        truth_in = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        truth_out = np.random.randn(16, 256, 256, 3).astype(np.float32)
+        self.cgan_metrics.update(disc_out, gen_out, truth_in, truth_out)
+        self.assertEqual(len(self.cgan_metrics.results), 1)
+
+    def test_get_metric(self):
+        self.cgan_metrics.results = pd.DataFrame(
+            {"MMD": [0.5, 0.6, 0.7], "PSNR": [20, 25, 30]}
+        )
+        mmd_mean = self.cgan_metrics.get_metric("MMD")
+        psnr_mean = self.cgan_metrics.get_metric("PSNR")
+        self.assertAlmostEqual(mmd_mean, 0.6, places=2)
+        self.assertAlmostEqual(psnr_mean, 25, places=2)
+
+
+if __name__ == "__main__":
+    unittest.main()

agrinet/utils/Metrics.py

@@ -23,15 +23,15 @@ def update(self, disc_out, gen_out, truth_in, truth_out):
 
     def psnr(self, x, y):
         """Compute the Peak Signal to Noise Ratio between two images"""
-        x = x.numpy()
-        y = y.numpy()
+        x = x.numpy() if hasattr(x, "numpy") else x
+        y = y.numpy() if hasattr(y, "numpy") else y
         mse = np.mean((x - y) ** 2)
         return 20 * np.log10(255) - 10 * np.log10(mse)
 
     def mmd(self, x, y):
         """Compute the Maximum Mean Discrepancy between two sets of samples"""
-        x = x.numpy()
-        y = y.numpy()
+        x = x.numpy() if hasattr(x, "numpy") else x
+        y = y.numpy() if hasattr(y, "numpy") else y
         x_flat = np.reshape(x, (x.shape[0], -1))
         y_flat = np.reshape(y, (y.shape[0], -1))