Fix tt-prefixed utility names

Author: brandonwillard

pymc3_hmm/utils.py

@@ -4,10 +4,10 @@
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
+import scipy.special as sp
 from matplotlib import cm
 from matplotlib.axes import Axes
 from matplotlib.colors import Colormap
-from scipy.special import logsumexp
 
 vsearchsorted = np.vectorize(np.searchsorted, otypes=[int], signature="(n),()->()")
 
@@ -76,7 +76,7 @@ def compute_trans_freqs(states, N_states, counts_only=False):
     return res
 
 
-def tt_logsumexp(x, axis=None, keepdims=False):
+def logsumexp(x, axis=None, keepdims=False):
     """Construct a Theano graph for a log-sum-exp calculation."""
     x_max_ = at.max(x, axis=axis, keepdims=True)
 
@@ -103,7 +103,7 @@ def tt_logsumexp(x, axis=None, keepdims=False):
     return res + x_max_
 
 
-def tt_logdotexp(A, b):
+def logdotexp(A, b):
     """Construct a Theano graph for a numerically stable log-scale dot product.
 
     The result is more or less equivalent to `tt.log(tt.exp(A).dot(tt.exp(b)))`
@@ -120,11 +120,11 @@ def tt_logdotexp(A, b):
         sqz = True
 
     b_bcast = b.dimshuffle(shape_b)
-    res = tt_logsumexp(A_bcast + b_bcast, axis=1)
+    res = logsumexp(A_bcast + b_bcast, axis=1)
     return res.squeeze() if sqz else res
 
 
-def logdotexp(A, b):
+def np_logdotexp(A, b):
     """Compute a numerically stable log-scale dot product of NumPy values.
 
     The result is more or less equivalent to `np.log(np.exp(A).dot(np.exp(b)))`
@@ -138,7 +138,7 @@ def logdotexp(A, b):
 
     A_bcast = np.expand_dims(A, -1)
 
-    res = logsumexp(A_bcast + b_bcast, axis=1)
+    res = sp.logsumexp(A_bcast + b_bcast, axis=1)
     return res.squeeze() if sqz else res
 
 
@@ -184,10 +184,10 @@ def multilogit_inv(ys):
     """
     if isinstance(ys, np.ndarray):
         lib = np
-        lib_logsumexp = logsumexp
+        lib_logsumexp = sp.logsumexp
     else:
         lib = at
-        lib_logsumexp = tt_logsumexp
+        lib_logsumexp = logsumexp
 
     # exp_ys = lib.exp(ys)
     # res = lib.concatenate([exp_ys, lib.ones(tuple(ys.shape)[:-1] + (1,))], axis=-1)

tests/test_utils.py

@@ -1,4 +1,3 @@
-import aesara
 import aesara.tensor as at
 import numpy as np
 import pytest
@@ -7,9 +6,9 @@
 from pymc3_hmm.utils import (
     compute_trans_freqs,
     logdotexp,
+    logsumexp,
     multilogit_inv,
-    tt_logdotexp,
-    tt_logsumexp,
+    np_logdotexp,
 )
 
 
@@ -35,62 +34,60 @@ def test_compute_trans_freqs():
 )
 def test_logsumexp(test_input):
     np_res = sp.special.logsumexp(test_input)
-    tt_res = tt_logsumexp(at.as_tensor_variable(test_input)).eval()
-    assert np.array_equal(np_res, tt_res)
+    at_res = logsumexp(at.as_tensor_variable(test_input)).eval()
+    assert np.array_equal(np_res, at_res)
 
 
-def test_logdotexp():
+def test_np_logdotexp():
     A = np.c_[[1.0, 2.0], [3.0, 4.0], [10.0, 20.0]]
     b = np.c_[[0.1], [0.2], [30.0]].T
 
-    test_res = logdotexp(np.log(A), np.log(b))
+    test_res = np_logdotexp(np.log(A), np.log(b))
     assert test_res.shape == (2, 1)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     b = np.r_[0.1, 0.2, 30.0]
-    test_res = logdotexp(np.log(A), np.log(b))
+    test_res = np_logdotexp(np.log(A), np.log(b))
     assert test_res.shape == (2,)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     A = np.c_[[1.0, 2.0], [10.0, 20.0]]
     b = np.c_[[0.1], [0.2]].T
-    test_res = logdotexp(np.log(A), np.log(b))
+    test_res = np_logdotexp(np.log(A), np.log(b))
     assert test_res.shape == (2, 1)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     b = np.r_[0.1, 0.2]
-    test_res = logdotexp(np.log(A), np.log(b))
+    test_res = np_logdotexp(np.log(A), np.log(b))
     assert test_res.shape == (2,)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
 
-def test_tt_logdotexp():
+def test_at_logdotexp():
 
     np.seterr(over="ignore", under="ignore")
 
-    aesara.config.compute_test_value = "warn"
-
     A = np.c_[[1.0, 2.0], [3.0, 4.0], [10.0, 20.0]]
     b = np.c_[[0.1], [0.2], [30.0]].T
     A_tt = at.as_tensor_variable(A)
     b_tt = at.as_tensor_variable(b)
-    test_res = tt_logdotexp(at.log(A_tt), at.log(b_tt)).eval()
+    test_res = logdotexp(at.log(A_tt), at.log(b_tt)).eval()
     assert test_res.shape == (2, 1)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     b = np.r_[0.1, 0.2, 30.0]
-    test_res = tt_logdotexp(at.log(A), at.log(b)).eval()
+    test_res = logdotexp(at.log(A), at.log(b)).eval()
     assert test_res.shape == (2,)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     A = np.c_[[1.0, 2.0], [10.0, 20.0]]
     b = np.c_[[0.1], [0.2]].T
-    test_res = tt_logdotexp(at.log(A), at.log(b)).eval()
+    test_res = logdotexp(at.log(A), at.log(b)).eval()
     assert test_res.shape == (2, 1)
     assert np.allclose(A.dot(b), np.exp(test_res))
 
     b = np.r_[0.1, 0.2]
-    test_res = tt_logdotexp(at.log(A), at.log(b)).eval()
+    test_res = logdotexp(at.log(A), at.log(b)).eval()
     assert test_res.shape == (2,)
     assert np.allclose(A.dot(b), np.exp(test_res))