Add unit tests

.appveyor.yml

@@ -33,5 +33,5 @@ install:
 # Not a .NET project, we build in the install step instead
 build: false
 test_script:
-  - cd examples
-  - python -c "import msdnet"
+  - cd tests
+  - python traintest.py -v

.gitignore

@@ -7,6 +7,10 @@ examples/*.txt
 examples/*.checkpoint
 examples/*/*.tiff
 examples/*/*/*.tiff
+tests/*.h5
+tests/*.png
+tests/*.txt
+tests/*.checkpoint
 .vscode/
 
 

.travis.yml

@@ -13,5 +13,5 @@ install:
 
 script:
   - python setup.py install
-  - cd examples
-  - python -c "import msdnet"
+  - cd tests
+  - python traintest.py -v

tests/traintest.py

@@ -0,0 +1,63 @@
+import unittest
+import msdnet
+import numpy as np
+
+class TestTraining(unittest.TestCase):
+
+    def test_regr_training(self):
+        dil = msdnet.dilations.IncrementDilations(3)
+        n = msdnet.network.MSDNet(10, dil, 1, 1, gpu=False)
+        n.initialize()
+        tgt_im = np.zeros((1,128,128), dtype=np.float32)
+        tgt_im[:,32:96,32:96] = 1
+        inp_im = (tgt_im + np.random.normal(size=tgt_im.shape)).astype(np.float32)
+        d = msdnet.data.ArrayDataPoint(inp_im, tgt_im)
+        bprov = msdnet.data.BatchProvider([d,],1)
+        val = msdnet.validate.MSEValidation([d,])
+        t = msdnet.train.AdamAlgorithm(n)
+        n.normalizeinout([d,])
+        log1 = msdnet.loggers.ConsoleLogger()
+        log2 = msdnet.loggers.FileLogger('test_log_regr.txt')
+        log3 = msdnet.loggers.ImageLogger('test_log_regr')
+        for i in range(10):
+            t.step(n, bprov.getbatch())
+            t.to_file('test_regr_params.h5')
+            val.to_file('test_regr_params.h5')
+            n.to_file('test_regr_params.h5')
+            val.validate(n)
+            log1.log(val)
+            log2.log(val)
+            log3.log(val)
+        n_load = msdnet.network.MSDNet.from_file('test_regr_params.h5')
+        n_load.forward(inp_im)
+
+    def test_segm_training(self):
+        dil = msdnet.dilations.IncrementDilations(3)
+        n = msdnet.network.SegmentationMSDNet(10, dil, 1, 2, gpu=False)
+        n.initialize()
+        tgt_im = np.zeros((1,128,128), dtype=np.uint8)
+        tgt_im[:,32:96,32:96] = 1
+        inp_im = (tgt_im + np.random.normal(size=tgt_im.shape)).astype(np.float32)
+        d = msdnet.data.ArrayDataPoint(inp_im, tgt_im)
+        d = msdnet.data.OneHotDataPoint(d,[0,1])
+        bprov = msdnet.data.BatchProvider([d,],1)
+        val = msdnet.validate.MSEValidation([d,])
+        t = msdnet.train.AdamAlgorithm(n)
+        n.normalizeinout([d,])
+        log1 = msdnet.loggers.ConsoleLogger()
+        log2 = msdnet.loggers.FileLogger('test_log_segm.txt')
+        log3 = msdnet.loggers.ImageLabelLogger('test_log_segm')
+        for i in range(10):
+            t.step(n, bprov.getbatch())
+            t.to_file('test_segm_params.h5')
+            val.to_file('test_segm_params.h5')
+            n.to_file('test_segm_params.h5')
+            val.validate(n)
+            log1.log(val)
+            log2.log(val)
+            log3.log(val)
+        n_load = msdnet.network.MSDNet.from_file('test_segm_params.h5')
+        n_load.forward(inp_im)
+
+if __name__ == '__main__':
+    unittest.main()