fixed tests

246f740d · Amjad Almahairi · 32549fe2 · 246f740d
--- a/theano/sandbox/tests/test_weighted_select.py
+++ b/theano/sandbox/tests/test_weighted_select.py
 import numpy
+import theano
 from theano import config, function, tensor
 from theano.sandbox import multinomial
+from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
+import unittest
-def test_select_distinct():
+class test_OP(unittest.TestCase):
-    p = tensor.fmatrix()
-    u = tensor.fvector()
+    def test_select_distinct(self):
-    n = tensor.iscalar()
+        """
-    m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
+        Tests that WeightedSelectionFromUniform always selects distinct elements
+        """
-    f = function([p, u, n], m, allow_input_downcast=True)
+        p = tensor.fmatrix()
+        u = tensor.fvector()
-    n_elements = 1000
+        n = tensor.iscalar()
-    numpy.random.seed(12345)
+        m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
-    for i in [5, 10, 50, 100, 500]:
-        uni = numpy.random.rand(i).astype(config.floatX)
+        f = function([p, u, n], m, allow_input_downcast=True)
+        n_elements = 1000
+        numpy.random.seed(12345)
+        for i in [5, 10, 50, 100, 500, n_elements]:
+            uni = numpy.random.rand(i).astype(config.floatX)
+            pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
+            pvals /= pvals.sum(1)
+            res = f(pvals, uni, i)
+            res = numpy.squeeze(res)
+            assert len(res) == i
+            assert numpy.all(numpy.in1d(numpy.unique(res), res)), res
+    def test_fail_select_alot(self):
+        """
+        Tests that WeightedSelectionFromUniform fails when asked to sample more
+        elements than the actual number of elements
+        """
+        p = tensor.fmatrix()
+        u = tensor.fvector()
+        n = tensor.iscalar()
+        m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
+        f = function([p, u, n], m, allow_input_downcast=True)
+        n_elements = 100
+        n_selected = 200
+        numpy.random.seed(12345)
+        uni = numpy.random.rand(n_selected).astype(config.floatX)
        pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
        pvals /= pvals.sum(1)
-        res = f(pvals, uni, i)
+        self.assertRaises(ValueError, f, pvals, uni, n_selected)
-        res = numpy.squeeze(res)
-        assert len(res) == i
+    def test_select_proportional_to_weight(self):
-        assert numpy.all(numpy.in1d(numpy.unique(res), res)), res
+        """
+        Tests that WeightedSelectionFromUniform selects elements, on average,
-def test_select_all():
+        proportional to the their probabilities
-    p = tensor.fmatrix()
+        """
-    u = tensor.fvector()
+        p = tensor.fmatrix()
-    n = tensor.iscalar()
+        u = tensor.fvector()
-    m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
+        n = tensor.iscalar()
+        m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
-    f = function([p, u, n], m, allow_input_downcast=True)
+        f = function([p, u, n], m, allow_input_downcast=True)
-    n_elements = 1000
-    numpy.random.seed(12345)
+        n_elements = 100
-    for _ in range(100):
+        n_selected = 10
-        uni = numpy.random.rand(n_elements).astype(config.floatX)
+        mean_rtol = 0.04
+        numpy.random.seed(12345)
        pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
        pvals /= pvals.sum(1)
-        res = f(pvals, uni, n_elements)
+        avg_pvals = numpy.zeros((n_elements,))
-        res = numpy.squeeze(res)
-        assert len(res) == n_elements
+        for rep in range(1000):
-        assert numpy.all(numpy.in1d(numpy.unique(res), res)), res
+            uni = numpy.random.rand(n_selected).astype(config.floatX)
+            res = f(pvals, uni, n_selected)
+            res = numpy.squeeze(res)
-def test_select_proportional_to_weight():
+            avg_pvals[res] += 1
-    p = tensor.fmatrix()
+        avg_pvals /= avg_pvals.sum()
-    u = tensor.fvector()
+        assert numpy.mean(abs(avg_pvals - pvals)) < mean_rtol
-    n = tensor.iscalar()
-    m = multinomial.WeightedSelectionFromUniform('auto')(p, u, n)
+class test_function(unittest.TestCase):
-    f = function([p, u, n], m, allow_input_downcast=True)
+    def test_select_distinct(self):
-    n_elements = 100
+        """
-    n_selected = 10
+        Tests that weighted_selection always selects distinct elements
-    mean_rtol = 0.04
+        """
-    numpy.random.seed(12345)
+        th_rng = RandomStreams(12345)
-    pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
-    pvals /= pvals.sum(1)
+        p = tensor.fmatrix()
-    avg_pvals = numpy.zeros((n_elements,))
+        n = tensor.iscalar()
+        m = th_rng.weighted_selection(pvals=p, n=n)
-    for rep in range(1000):
-        uni = numpy.random.rand(n_selected).astype(config.floatX)
+        f = function([p, n], m, allow_input_downcast=True)
-        res = f(pvals, uni, n_selected)
-        res = numpy.squeeze(res)
+        n_elements = 1000
-        # print res
+        numpy.random.seed(12345)
-        avg_pvals[res] += 1
+        for i in [5, 10, 50, 100, 500, n_elements]:
-    avg_pvals /= avg_pvals.sum()
+            pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
+            pvals /= pvals.sum(1)
-    print avg_pvals
+            res = f(pvals, i)
-    print numpy.squeeze(pvals)
+            res = numpy.squeeze(res)
-    assert numpy.mean(abs(avg_pvals - pvals)) < mean_rtol
+            assert len(res) == i
+            assert numpy.all(numpy.in1d(numpy.unique(res), res)), res
+    def test_fail_select_alot(self):
+        """
+        Tests that weighted_selection fails when asked to sample more
+        elements than the actual number of elements
+        """
+        th_rng = RandomStreams(12345)
+        p = tensor.fmatrix()
+        n = tensor.iscalar()
+        m = th_rng.weighted_selection(pvals=p, n=n)
+        f = function([p, n], m, allow_input_downcast=True)
+        n_elements = 100
+        n_selected = 200
+        numpy.random.seed(12345)
+        pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
+        pvals /= pvals.sum(1)
+        self.assertRaises(ValueError, f, pvals, n_selected)
+    def test_select_proportional_to_weight(self):
+        """
+        Tests that weighted_selection selects elements, on average,
+        proportional to the their probabilities
+        """
+        th_rng = RandomStreams(12345)
+        p = tensor.fmatrix()
+        n = tensor.iscalar()
+        m = th_rng.weighted_selection(pvals=p, n=n)
+        f = function([p, n], m, allow_input_downcast=True)
+        n_elements = 100
+        n_selected = 10
+        mean_rtol = 0.04
+        numpy.random.seed(12345)
+        pvals = numpy.random.randint(1,100,(1,n_elements)).astype(config.floatX)
+        pvals /= pvals.sum(1)
+        avg_pvals = numpy.zeros((n_elements,))
+        for rep in range(1000):
+            res = f(pvals, n_selected)
+            res = numpy.squeeze(res)
+            avg_pvals[res] += 1
+        avg_pvals /= avg_pvals.sum()
+        assert numpy.mean(abs(avg_pvals - pvals)) < mean_rtol