Change theano imports to aesara

.github/workflows/python-tests.yml

@@ -48,6 +48,7 @@ jobs:
     strategy:
       matrix:
         python-version: [3.7]
+        pymc3-version: [stable, dev]
 
     steps:
     - uses: actions/checkout@v2
@@ -60,6 +61,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
+        if [[ "${{ matrix.pymc3-version }}" != "stable" ]]; then pip install "pymc3 @ git+https://github.com/pymc-devs/pymc3.git@master"; fi
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
     - name: Test with pytest
       run: |

pymc3_hmm/distributions.py

@@ -1,15 +1,25 @@
 import warnings
 
 import numpy as np
+
+try:  # pragma: no cover
+    import aesara
+    import aesara.tensor as at
+    from aesara.graph.op import get_test_value
+    from aesara.graph.utils import TestValueError
+    from aesara.scalar import upcast
+    from aesara.tensor.extra_ops import broadcast_to as at_broadcast_to
+except ImportError:  # pragma: no cover
+    import theano as aesara
+    import theano.tensor as at
+    from theano.graph.op import get_test_value
+    from theano.graph.utils import TestValueError
+    from theano.scalar import upcast
+    from theano.tensor.extra_ops import broadcast_to as at_broadcast_to
+
 import pymc3 as pm
-import theano
-import theano.tensor as tt
 from pymc3.distributions.distribution import _DrawValuesContext, draw_values
 from pymc3.distributions.mixture import _conversion_map, all_discrete
-from theano.graph.op import get_test_value
-from theano.graph.utils import TestValueError
-from theano.scalar import upcast
-from theano.tensor.extra_ops import broadcast_to as tt_broadcast_to
 
 from pymc3_hmm.utils import tt_broadcast_arrays, tt_expand_dims, vsearchsorted
 
@@ -73,7 +83,7 @@ def distribution_subset_args(dist, shape, idx, point=None):
         else:
             x = point[param]
 
-        bcast_res = tt_broadcast_to(x, shape)
+        bcast_res = at_broadcast_to(x, shape)
 
         res.append(bcast_res[idx])
 
@@ -121,7 +131,7 @@ def __init__(self, comp_dists, states, *args, **kwargs):
             equal to the size of `comp_dists`.
 
         """
-        self.states = tt.as_tensor_variable(pm.intX(states))
+        self.states = at.as_tensor_variable(pm.intX(states))
 
         if len(comp_dists) > 31:
             warnings.warn(
@@ -147,7 +157,7 @@ def __init__(self, comp_dists, states, *args, **kwargs):
                 bcast_means = tt_broadcast_arrays(
                     *([self.states] + [d.mean.astype(dtype) for d in self.comp_dists])
                 )
-                self.mean = tt.choose(self.states, bcast_means[1:])
+                self.mean = at.choose(self.states, bcast_means[1:])
 
                 if "mean" not in defaults:
                     defaults.append("mean")
@@ -159,7 +169,7 @@ def __init__(self, comp_dists, states, *args, **kwargs):
             bcast_modes = tt_broadcast_arrays(
                 *([self.states] + [d.mode.astype(dtype) for d in self.comp_dists])
             )
-            self.mode = tt.choose(self.states, bcast_modes[1:])
+            self.mode = at.choose(self.states, bcast_modes[1:])
 
             if "mode" not in defaults:
                 defaults.append("mode")
@@ -172,15 +182,15 @@ def __init__(self, comp_dists, states, *args, **kwargs):
     def logp(self, obs):
         """Return the scalar Theano log-likelihood at a point."""
 
-        obs_tt = tt.as_tensor_variable(obs)
+        obs_tt = at.as_tensor_variable(obs)
 
-        logp_val = tt.alloc(-np.inf, *obs.shape)
+        logp_val = at.alloc(-np.inf, *obs.shape)
 
         for i, dist in enumerate(self.comp_dists):
-            i_mask = tt.eq(self.states, i)
+            i_mask = at.eq(self.states, i)
             obs_i = obs_tt[i_mask]
             subset_dist = dist.dist(*distribution_subset_args(dist, obs.shape, i_mask))
-            logp_val = tt.set_subtensor(logp_val[i_mask], subset_dist.logp(obs_i))
+            logp_val = at.set_subtensor(logp_val[i_mask], subset_dist.logp(obs_i))
 
         return logp_val
 
@@ -265,8 +275,8 @@ def __init__(self, mu=None, states=None, **kwargs):
             A vector of integer 0-1 states that indicate which component of
             the mixture is active at each point/time.
         """
-        self.mu = tt.as_tensor_variable(pm.floatX(mu))
-        self.states = tt.as_tensor_variable(states)
+        self.mu = at.as_tensor_variable(pm.floatX(mu))
+        self.states = at.as_tensor_variable(states)
 
         super().__init__([pm.Constant.dist(0), pm.Poisson.dist(mu)], states, **kwargs)
 
@@ -298,16 +308,16 @@ def __init__(self, Gammas, gamma_0, shape, **kwargs):
             Shape of the state sequence.  The last dimension is `N`, i.e. the
             length of the state sequence(s).
         """
-        self.gamma_0 = tt.as_tensor_variable(pm.floatX(gamma_0))
+        self.gamma_0 = at.as_tensor_variable(pm.floatX(gamma_0))
 
         assert Gammas.ndim >= 3
 
-        self.Gammas = tt.as_tensor_variable(pm.floatX(Gammas))
+        self.Gammas = at.as_tensor_variable(pm.floatX(Gammas))
 
         shape = np.atleast_1d(shape)
 
-        dtype = _conversion_map[theano.config.floatX]
-        self.mode = tt.zeros(tuple(shape), dtype=dtype)
+        dtype = _conversion_map[aesara.config.floatX]
+        self.mode = np.zeros(tuple(shape), dtype=dtype)
 
         super().__init__(shape=shape, **kwargs)
 
@@ -330,44 +340,44 @@ def logp(self, states):
 
         """  # noqa: E501
 
-        Gammas = tt.shape_padleft(self.Gammas, states.ndim - (self.Gammas.ndim - 2))
+        Gammas = at.shape_padleft(self.Gammas, states.ndim - (self.Gammas.ndim - 2))
 
         # Multiply the initial state probabilities by the first transition
         # matrix by to get the marginal probability for state `S_1`.
         # The integral that produces the marginal is essentially
         # `gamma_0.dot(Gammas[0])`
         Gamma_1 = Gammas[..., 0:1, :, :]
         gamma_0 = tt_expand_dims(self.gamma_0, (-3, -1))
-        P_S_1 = tt.sum(gamma_0 * Gamma_1, axis=-2)
+        P_S_1 = at.sum(gamma_0 * Gamma_1, axis=-2)
 
         # The `tt.switch`s allow us to broadcast the indexing operation when
         # the replication dimensions of `states` and `Gammas` don't match
         # (e.g. `states.shape[0] > Gammas.shape[0]`)
         S_1_slices = tuple(
             slice(
-                tt.switch(tt.eq(P_S_1.shape[i], 1), 0, 0),
-                tt.switch(tt.eq(P_S_1.shape[i], 1), 1, d),
+                at.switch(at.eq(P_S_1.shape[i], 1), 0, 0),
+                at.switch(at.eq(P_S_1.shape[i], 1), 1, d),
             )
             for i, d in enumerate(states.shape)
         )
-        S_1_slices = (tuple(tt.ogrid[S_1_slices]) if S_1_slices else tuple()) + (
+        S_1_slices = (tuple(at.ogrid[S_1_slices]) if S_1_slices else tuple()) + (
             states[..., 0:1],
         )
-        logp_S_1 = tt.log(P_S_1[S_1_slices]).sum(axis=-1)
+        logp_S_1 = at.log(P_S_1[S_1_slices]).sum(axis=-1)
 
         # These are slices for the extra dimensions--including the state
         # sequence dimension (e.g. "time")--along which which we need to index
         # the transition matrix rows using the "observed" `states`.
         trans_slices = tuple(
             slice(
-                tt.switch(
-                    tt.eq(Gammas.shape[i], 1), 0, 1 if i == states.ndim - 1 else 0
+                at.switch(
+                    at.eq(Gammas.shape[i], 1), 0, 1 if i == states.ndim - 1 else 0
                 ),
-                tt.switch(tt.eq(Gammas.shape[i], 1), 1, d),
+                at.switch(at.eq(Gammas.shape[i], 1), 1, d),
             )
             for i, d in enumerate(states.shape)
         )
-        trans_slices = (tuple(tt.ogrid[trans_slices]) if trans_slices else tuple()) + (
+        trans_slices = (tuple(at.ogrid[trans_slices]) if trans_slices else tuple()) + (
             states[..., :-1],
         )
 
@@ -376,12 +386,12 @@ def logp(self, states):
         P_S_2T = Gammas[trans_slices]
 
         obs_slices = tuple(slice(None, d) for d in P_S_2T.shape[:-1])
-        obs_slices = (tuple(tt.ogrid[obs_slices]) if obs_slices else tuple()) + (
+        obs_slices = (tuple(at.ogrid[obs_slices]) if obs_slices else tuple()) + (
             states[..., 1:],
         )
-        logp_S_1T = tt.log(P_S_2T[obs_slices])
+        logp_S_1T = at.log(P_S_2T[obs_slices])
 
-        res = logp_S_1 + tt.sum(logp_S_1T, axis=-1)
+        res = logp_S_1 + at.sum(logp_S_1T, axis=-1)
         res.name = "DiscreteMarkovChain_logp"
 
         return res

pymc3_hmm/step_methods.py

@@ -1,21 +1,36 @@
 from itertools import chain
 
 import numpy as np
+
+try:  # pragma: no cover
+    import aesara.scalar as aes
+    import aesara.tensor as at
+    from aesara.compile import optdb
+    from aesara.graph.basic import Variable, graph_inputs
+    from aesara.graph.fg import FunctionGraph
+    from aesara.graph.op import get_test_value as test_value
+    from aesara.graph.opt import OpRemove, pre_greedy_local_optimizer
+    from aesara.graph.optdb import Query
+    from aesara.tensor.elemwise import DimShuffle, Elemwise
+    from aesara.tensor.subtensor import AdvancedIncSubtensor1
+    from aesara.tensor.var import TensorConstant
+except ImportError:  # pragma: no cover
+    import theano.scalar as aes
+    import theano.tensor as at
+    from theano.compile import optdb
+    from theano.graph.basic import Variable, graph_inputs
+    from theano.graph.fg import FunctionGraph
+    from theano.graph.op import get_test_value as test_value
+    from theano.graph.opt import OpRemove, pre_greedy_local_optimizer
+    from theano.graph.optdb import Query
+    from theano.tensor.elemwise import DimShuffle, Elemwise
+    from theano.tensor.subtensor import AdvancedIncSubtensor1
+    from theano.tensor.var import TensorConstant
+
 import pymc3 as pm
-import theano.scalar as ts
-import theano.tensor as tt
 from pymc3.distributions.distribution import draw_values
 from pymc3.step_methods.arraystep import ArrayStep, BlockedStep, Competence
 from pymc3.util import get_untransformed_name
-from theano.compile import optdb
-from theano.graph.basic import Variable, graph_inputs
-from theano.graph.fg import FunctionGraph
-from theano.graph.op import get_test_value as test_value
-from theano.graph.opt import OpRemove, pre_greedy_local_optimizer
-from theano.graph.optdb import Query
-from theano.tensor.elemwise import DimShuffle, Elemwise
-from theano.tensor.subtensor import AdvancedIncSubtensor1
-from theano.tensor.var import TensorConstant
 
 from pymc3_hmm.distributions import DiscreteMarkovChain, SwitchingProcess
 from pymc3_hmm.utils import compute_trans_freqs
@@ -159,15 +174,15 @@ def __init__(self, vars, values=None, model=None):
                 for comp_dist in dependent_rv.distribution.comp_dists:
                     comp_logps.append(comp_dist.logp(dependent_rv))
 
-                comp_logp_stacked = tt.stack(comp_logps)
+                comp_logp_stacked = at.stack(comp_logps)
             else:
                 raise TypeError(
                     "This sampler only supports `SwitchingProcess` observations"
                 )
 
             dep_comps_logp_stacked.append(comp_logp_stacked)
 
-        comp_logp_stacked = tt.sum(dep_comps_logp_stacked, axis=0)
+        comp_logp_stacked = at.sum(dep_comps_logp_stacked, axis=0)
 
         (M,) = draw_values([var.distribution.gamma_0.shape[-1]], point=model.test_point)
         N = model.test_point[var.name].shape[-1]
@@ -326,9 +341,9 @@ def _set_row_mappings(self, Gamma, dir_priors, model):
         Gamma = pre_greedy_local_optimizer(
             FunctionGraph([], []),
             [
-                OpRemove(Elemwise(ts.Cast(ts.float32))),
-                OpRemove(Elemwise(ts.Cast(ts.float64))),
-                OpRemove(Elemwise(ts.identity)),
+                OpRemove(Elemwise(aes.Cast(aes.float32))),
+                OpRemove(Elemwise(aes.Cast(aes.float64))),
+                OpRemove(Elemwise(aes.identity)),
             ],
             Gamma,
         )
@@ -352,7 +367,7 @@ def _set_row_mappings(self, Gamma, dir_priors, model):
 
         Gamma_Join = Gamma_DimShuffle.inputs[0].owner
 
-        if not (isinstance(Gamma_Join.op, tt.basic.Join)):
+        if not (isinstance(Gamma_Join.op, at.basic.Join)):
             raise TypeError(
                 "The transition matrix should be comprised of stacked row vectors"
             )

pymc3_hmm/utils.py

@@ -3,14 +3,22 @@
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-import theano.tensor as tt
 from matplotlib import cm
 from matplotlib.axes import Axes
 from matplotlib.colors import Colormap
 from scipy.special import logsumexp
-from theano.tensor.extra_ops import broadcast_shape
-from theano.tensor.extra_ops import broadcast_to as tt_broadcast_to
-from theano.tensor.var import TensorVariable
+
+try:  # pragma: no cover
+    import aesara.tensor as at
+    from aesara.tensor.extra_ops import broadcast_shape
+    from aesara.tensor.extra_ops import broadcast_to as at_broadcast_to
+    from aesara.tensor.var import TensorVariable
+except ImportError:  # pragma: no cover
+    import theano.tensor as at
+    from theano.tensor.extra_ops import broadcast_shape
+    from theano.tensor.extra_ops import broadcast_to as at_broadcast_to
+    from theano.tensor.var import TensorVariable
+
 
 vsearchsorted = np.vectorize(np.searchsorted, otypes=[int], signature="(n),()->()")
 
@@ -30,8 +38,8 @@ def compute_steady_state(P):
 
     P = P[0]
     N_states = P.shape[-1]
-    Lam = (tt.eye(N_states) - P + tt.ones((N_states, N_states))).T
-    u = tt.slinalg.solve(Lam, tt.ones((N_states,)))
+    Lam = (at.eye(N_states) - P + at.ones((N_states, N_states))).T
+    u = at.slinalg.solve(Lam, at.ones((N_states,)))
     return u
 
 
@@ -81,15 +89,15 @@ def compute_trans_freqs(states, N_states, counts_only=False):
 
 def tt_logsumexp(x, axis=None, keepdims=False):
     """Construct a Theano graph for a log-sum-exp calculation."""
-    x_max_ = tt.max(x, axis=axis, keepdims=True)
+    x_max_ = at.max(x, axis=axis, keepdims=True)
 
     if x_max_.ndim > 0:
-        x_max_ = tt.set_subtensor(x_max_[tt.isinf(x_max_)], 0.0)
-    elif tt.isinf(x_max_):
-        x_max_ = tt.as_tensor(0.0)
+        x_max_ = at.set_subtensor(x_max_[at.isinf(x_max_)], 0.0)
+    elif at.isinf(x_max_):
+        x_max_ = at.as_tensor(0.0)
 
-    res = tt.sum(tt.exp(x - x_max_), axis=axis, keepdims=keepdims)
-    res = tt.log(res)
+    res = at.sum(at.exp(x - x_max_), axis=axis, keepdims=keepdims)
+    res = at.log(res)
 
     if not keepdims:
         # SciPy uses the `axis` keyword here, but Theano doesn't support that.
@@ -179,7 +187,7 @@ def tt_broadcast_arrays(*args: TensorVariable):
 
     """
     bcast_shape = broadcast_shape(*args)
-    return tuple(tt_broadcast_to(a, bcast_shape) for a in args)
+    return tuple(at_broadcast_to(a, bcast_shape) for a in args)
 
 
 def multilogit_inv(ys):
@@ -202,7 +210,7 @@ def multilogit_inv(ys):
         lib = np
         lib_logsumexp = logsumexp
     else:
-        lib = tt
+        lib = at
         lib_logsumexp = tt_logsumexp
 
     # exp_ys = lib.exp(ys)

setup.cfg

@@ -14,8 +14,8 @@ add-ignore = D100,D101,D102,D103,D104,D105,D106,D107,D202
 convention = numpy
 
 [tool:pytest]
-python_files=test*.py
-testpaths=tests
+python_files = test_*.py
+testpaths = tests
 
 [coverage:run]
 omit =