Test fixes for CUDA (part 3 / n) (#2000)

* Refactor float32 test helpers, fix test_step

* Adjust precision for CPU float32

* Revert some of the extra floatX

* Move fxarray to floatX_array in theanof

* Fix floatX in test_stats for CUDA

* Fixes for test_starting on CUDA

* Fix lint

Author: kyleabeauchamp

pymc3/tests/helpers.py

@@ -1,9 +1,7 @@
 from logging.handlers import BufferingHandler
 import numpy.random as nr
-import numpy as np
 from theano.sandbox.rng_mrg import MRG_RandomStreams
 from ..theanof import set_tt_rng, tt_rng
-from pymc3.theanof import floatX
 import theano
 
 
@@ -22,6 +20,7 @@ def setup_method(self):
     def teardown_method(self):
         set_tt_rng(self.old_tt_rng)
 
+
 class LoggingHandler(BufferingHandler):
     def __init__(self, matcher):
         # BufferingHandler takes a "capacity" argument
@@ -48,6 +47,7 @@ def matches(self, **kwargs):
                 break
         return result
 
+
 class Matcher(object):
 
     _partial_matches = ('msg', 'message')
@@ -86,7 +86,3 @@ def select_by_precision(float64, float32):
     """Helper function to choose reasonable decimal cutoffs for different floatX modes."""
     decimal = float64 if theano.config.floatX == "float64" else float32
     return decimal
-
-
-def fxarray(x):
-    return floatX(np.array(x))

pymc3/tests/models.py

@@ -3,8 +3,10 @@
 import pymc3 as pm
 from itertools import product
 import theano.tensor as tt
+import theano
 from theano.compile.ops import as_op
-from .helpers import fxarray
+from pymc3.theanof import floatX_array
+
 
 def simple_model():
     mu = -2.1
@@ -16,8 +18,8 @@ def simple_model():
 
 
 def simple_categorical():
-    p = fxarray([0.1, 0.2, 0.3, 0.4])
-    v = fxarray([0.0, 1.0, 2.0, 3.0])
+    p = floatX_array([0.1, 0.2, 0.3, 0.4])
+    v = floatX_array([0.0, 1.0, 2.0, 3.0])
     with Model() as model:
         Categorical('x', p, shape=3, testval=[1, 2, 3])
 
@@ -36,14 +38,16 @@ def multidimensional_model():
 
 
 def simple_arbitrary_det():
-    @as_op(itypes=[tt.dscalar], otypes=[tt.dscalar])
+    scalar_type = tt.dscalar if theano.config.floatX == "float64" else tt.fscalar
+
+    @as_op(itypes=[scalar_type], otypes=[scalar_type])
     def arbitrary_det(value):
         return value
 
     with Model() as model:
         a = Normal('a')
         b = arbitrary_det(a)
-        Normal('obs', mu=b.astype('float64'), observed=fxarray([1, 3, 5]))
+        Normal('obs', mu=b.astype('float64'), observed=floatX_array([1, 3, 5]))
 
     return model.test_point, model
 
@@ -66,15 +70,15 @@ def simple_2model():
 
 
 def mv_simple():
-    mu = fxarray([-.1, .5, 1.1])
-    p = fxarray([
+    mu = floatX_array([-.1, .5, 1.1])
+    p = floatX_array([
         [2., 0, 0],
         [.05, .1, 0],
         [1., -0.05, 5.5]])
     tau = np.dot(p, p.T)
     with pm.Model() as model:
         pm.MvNormal('x', tt.constant(mu), tau=tt.constant(tau),
-                    shape=3, testval=fxarray([.1, 1., .8]))
+                    shape=3, testval=floatX_array([.1, 1., .8]))
     H = tau
     C = np.linalg.inv(H)
     return model.test_point, model, (mu, C)
@@ -83,7 +87,7 @@ def mv_simple():
 def mv_simple_discrete():
     d = 2
     n = 5
-    p = fxarray([.15, .85])
+    p = floatX_array([.15, .85])
     with pm.Model() as model:
         pm.Multinomial('x', n, tt.constant(p), shape=d, testval=np.array([1, 4]))
         mu = n * p
@@ -106,7 +110,7 @@ def mv_prior_simple():
     K = pm.gp.cov.ExpQuad(1, 1)(X).eval()
     L = np.linalg.cholesky(K)
     K_noise = K + noise * np.eye(n)
-    obs = fxarray([-0.1, 0.5, 1.1])
+    obs = floatX_array([-0.1, 0.5, 1.1])
 
     # Posterior mean
     L_noise = np.linalg.cholesky(K_noise)

pymc3/tests/test_starting.py

@@ -3,19 +3,21 @@
 from pymc3.tuning import starting
 from pymc3 import Model, Uniform, Normal, Beta, Binomial, find_MAP, Point
 from .models import simple_model, non_normal, exponential_beta, simple_arbitrary_det
+from .helpers import select_by_precision
+
 
 def test_accuracy_normal():
     _, model, (mu, _) = simple_model()
     with model:
         newstart = find_MAP(Point(x=[-10.5, 100.5]))
-        close_to(newstart['x'], [mu, mu], 1e-5)
+        close_to(newstart['x'], [mu, mu], select_by_precision(float64=1e-5, float32=1E-4))
 
 
 def test_accuracy_non_normal():
     _, model, (mu, _) = non_normal(4)
     with model:
         newstart = find_MAP(Point(x=[.5, .01, .95, .99]))
-        close_to(newstart['x'], mu, 1e-5)
+        close_to(newstart['x'], mu, select_by_precision(float64=1e-5, float32=1E-4))
 
 
 def test_errors():

pymc3/tests/test_stats.py

@@ -7,6 +7,7 @@
 from ..tests import backend_fixtures as bf
 from ..backends import ndarray
 from ..stats import df_summary, autocorr, hpd, mc_error, quantiles, make_indices
+from ..theanof import floatX_array
 from numpy.random import random, normal
 from numpy.testing import assert_equal, assert_almost_equal, assert_array_almost_equal
 from scipy import stats as st
@@ -116,7 +117,7 @@ def test_summary_0d_variable_model(self):
         mu = -2.1
         tau = 1.3
         with Model() as model:
-            Normal('x', mu, tau, testval=.1)
+            Normal('x', mu, tau, testval=floatX_array(.1))
             step = Metropolis(model.vars, np.diag([1.]), blocked=True)
             trace = pm.sample(100, step=step)
         pm.summary(trace)
@@ -125,7 +126,7 @@ def test_summary_1d_variable_model(self):
         mu = -2.1
         tau = 1.3
         with Model() as model:
-            Normal('x', mu, tau, shape=2, testval=[.1, .1])
+            Normal('x', mu, tau, shape=2, testval=floatX_array([.1, .1]))
             step = Metropolis(model.vars, np.diag([1.]), blocked=True)
             trace = pm.sample(100, step=step)
         pm.summary(trace)
@@ -135,7 +136,7 @@ def test_summary_2d_variable_model(self):
         tau = 1.3
         with Model() as model:
             Normal('x', mu, tau, shape=(2, 2),
-                   testval=np.tile(.1, (2, 2)))
+                   testval=floatX_array(np.tile(.1, (2, 2))))
             step = Metropolis(model.vars, np.diag([1.]), blocked=True)
             trace = pm.sample(100, step=step)
         pm.summary(trace)

pymc3/theanof.py

@@ -409,3 +409,7 @@ def __init__(self, multiplier):
 
     def grad(self, args, g_outs):
         return [self.multiplier * g_out for g_out in g_outs]
+
+
+def floatX_array(x):
+    return floatX(np.array(x))