added separate file for decorators tests

Author: tariqdaouda

Mariana/tests/decorators_tests.py

@@ -0,0 +1,55 @@
+import unittest
+
+import Mariana.layers as ML
+import Mariana.initializations as MI
+import Mariana.decorators as MD
+import Mariana.costs as MC
+import Mariana.regularizations as MR
+import Mariana.scenari as MS
+import Mariana.activations as MA
+
+import theano.tensor as tt
+import numpy
+
+class DecoratorTests(unittest.TestCase):
+
+	def setUp(self) :
+		pass
+
+	def tearDown(self) :
+		pass
+
+	# @unittest.skip("skipping")
+	def test_batch_norm(self) :
+		import theano, numpy
+		
+		def batchnorm(W, b, data) :
+			return numpy.asarray( W * ( (data-numpy.mean(data)) / numpy.std(data) ) + b, dtype= theano.config.floatX)
+
+		data = numpy.random.randn(1, 100).astype(theano.config.floatX)
+		
+		inp = ML.Input(100, 'inp', decorators=[MD.BatchNormalization()])
+		
+		model = inp.network
+		m1 = numpy.mean( model.propagate(inp, inp=data)["outputs"])
+		m2 = numpy.mean( batchnorm(inp.batchnorm_W.get_value(), inp.batchnorm_b.get_value(), data) )
+
+		epsilon = 1e-6
+		self.assertTrue ( (m1 - m2) < epsilon )
+	
+	# @unittest.skip("skipping")
+	def test_mask(self) :
+		import theano, numpy
+		
+		inp = ML.Input(100, 'inp', decorators=[MD.Mask(mask = numpy.zeros(100))])
+		model = inp.network
+
+		data = numpy.random.randn(1, 100).astype(theano.config.floatX)
+		out = model.propagate(inp, inp=data)["outputs"]
+		
+		self.assertEqual(sum(out[0]), 0)
+
+if __name__ == '__main__' :
+	import Mariana.settings as MSET
+	MSET.VERBOSE = False
+	unittest.main()

Mariana/tests/tests.py

@@ -244,24 +244,6 @@ def makeDataset(nbExamples, size, patternSize) :
 
 		self.assertTrue(res["score"] < 0.1)
 
-	# @unittest.skip("skipping")
-	def test_batch_norm(self) :
-		import theano, numpy
-		
-		def batchnorm(W, b, data) :
-			return numpy.asarray( W * ( (data-numpy.mean(data)) / numpy.std(data) ) + b, dtype= theano.config.floatX)
-
-		data = numpy.random.randn(1, 100).astype(theano.config.floatX)
-		
-		inp = ML.Input(100, 'inp', decorators=[MD.BatchNormalization()])
-		
-		model = inp.network
-		m1 = numpy.mean( model.propagate(inp, inp=data)["outputs"])
-		m2 = numpy.mean( batchnorm(inp.batchnorm_W.get_value(), inp.batchnorm_b.get_value(), data) )
-
-		epsilon = 1e-6
-		self.assertTrue ( (m1 - m2) < epsilon )
-
 if __name__ == '__main__' :
 	import Mariana.settings as MSET
 	MSET.VERBOSE = False