* Test image used for unittests is now a pseudo-random image instead of a true...

* Test image used for unittests is now a pseudo-random image instead of a true random image to ensure comparable results.
* Fixed issue #6

 (ValueError: X has 6 features, but MaxAbsScaler is expecting 1 features as input.).
Signed-off-by: Daniel Scheffler <danschef@gfz-potsdam.de>

specclassify/misc.py

@@ -35,13 +35,22 @@ def normalize_endmembers_image(endmembers, image):
     em = endmembers.astype(np.float)
     im = image.astype(np.float)
 
+    # provide training values as 2D ROW (n samples x 1 feature),
+    # because normalization should be applied globally, not band-by-band
     allVals = np.hstack([em.flat, im.flat]).reshape(-1, 1)
 
     if allVals.min() < -1 or allVals.max() > 1:
         max_abs_scaler = MaxAbsScaler()
-        max_abs_scaler.fit_transform(allVals)
-        endmembers_norm = max_abs_scaler.transform(em)
-        image_norm = spectra2im(max_abs_scaler.transform(im2spectra(im)), tgt_rows=im.shape[0], tgt_cols=im.shape[1])
+        max_abs_scaler.fit(allVals)
+
+        endmembers_norm = \
+            max_abs_scaler \
+            .transform(em.reshape(-1, 1)) \
+            .reshape(em.shape)
+        image_norm = \
+            max_abs_scaler \
+            .transform(im.reshape(-1, 1)) \
+            .reshape(im.shape)
 
         return endmembers_norm, image_norm
 

tests/test_image_classifier.py

@@ -64,20 +64,23 @@ test_gA_pure_endmembers[:, :, :] = cluster_centers
 
 # compute test image
 def _get_testIm(shape, nodataVal):
+    # fix the random state to avoid different test images with each new test run (would be not comparable)
+    rand = np.random.RandomState(1234)
+
     # get array full of nodata values
     spec = np.full(shape, nodataVal).reshape(shape[0] * shape[1], shape[2])
 
     # fill array with random spectra taken from the cluster centers
     for i in range(cluster_centers.shape[0]):
-        spec[np.random.randint(spec.shape[0],
-                               size=spec.shape[0]//10), :] = \
+        spec[rand.randint(spec.shape[0],
+                          size=spec.shape[0]//10), :] = \
             cluster_centers[i, :]
 
     # fill remaining nodata positions with first spectrum
     spec[spec[:, 0] == nodataVal, :] = cluster_centers[0, :]
 
     # add 10% noise to each spectrum
-    noise_factor = np.random.normal(1, .1, spec.size).reshape(spec.shape)
+    noise_factor = rand.normal(1, .1, spec.size).reshape(spec.shape)
     spec_out = spec * noise_factor
 
     # return in desired image dimensions