Fix lint formatting and mypy checks

.github/workflows/ci.yml

@@ -7,8 +7,6 @@ on:
     branches: [dev, master]
 
 env:
-  CXX: g++-9
-  CC: gcc-9
   # See coveralls-python - Github Actions support:
   # https://github.com/TheKevJames/coveralls-python/blob/master/docs/usage/configuration.rst#github-actions-support
   GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -31,6 +29,7 @@ jobs:
       run: |
         python -m pip install --upgrade pip wheel setuptools
         pip install ruff black mypy nbstripout nbformat
+        pip install -e .
     - name: Lint
       run: |
         make lint

Makefile

@@ -21,7 +21,7 @@ lint: FORCE
 	ruff check .
 	black --check *.py pyro examples tests scripts profiler
 	python scripts/update_headers.py --check
-	mypy --install-types --non-interactive pyro scripts tests
+	mypy --install-types --non-interactive --warn-unused-ignores pyro scripts tests
 
 license: FORCE
 	python scripts/update_headers.py
@@ -83,7 +83,7 @@ test-jit: FORCE
 		-k JIT=True | tee -a jit.log
 
 test-funsor: lint FORCE
-	pytest -vx -n auto --stage funsor
+	pytest -vx -n auto -p no:benchmark --stage funsor
 
 clean: FORCE
 	git clean -dfx -e pyro_ppl.egg-info

docs/requirements.txt

@@ -2,14 +2,14 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-sphinx==4.2.0
-sphinx-rtd-theme==1.0.0
+sphinx>=4.2.0
+sphinx-rtd-theme>=1.0.0
 graphviz>=0.8
 numpy>=1.7
 observations>=0.1.4
 opt_einsum>=2.3.2
 pyro-api>=0.1.1
 tqdm>=4.36
-funsor[torch]
+funsor[torch] @ git+https://github.com/pyro-ppl/funsor.git@8a8ca81fcc61733ad82e163028c7ff78420df9f0
 setuptools
 sphinx_copybutton

examples/air/main.py

@@ -9,7 +9,6 @@
 Processing Systems. 2016.
 """
 
-
 import argparse
 import math
 import os
@@ -18,7 +17,6 @@
 
 import numpy as np
 import torch
-import visdom
 from air import AIR, latents_to_tensor
 from viz import draw_many, tensor_to_objs
 
@@ -204,6 +202,8 @@ def z_pres_prior_p(opt_step, time_step):
 
     # Viz sample from prior.
     if args.viz:
+        import visdom
+
         vis = visdom.Visdom(env=args.visdom_env)
         z, x = air.prior(5, z_pres_prior_p=partial(z_pres_prior_p, 0))
         vis.images(draw_many(x, tensor_to_objs(latents_to_tensor(z))))

examples/baseball.py

@@ -178,14 +178,12 @@ def train_test_split(pd_dataframe):
     Training data - 45 initial at-bats and hits for each player.
     Validation data - Full season at-bats and hits for each player.
     """
-    device = torch.Tensor().device
     train_data = torch.tensor(
-        pd_dataframe[["At-Bats", "Hits"]].values, dtype=torch.float, device=device
+        pd_dataframe[["At-Bats", "Hits"]].values, dtype=torch.float
     )
     test_data = torch.tensor(
         pd_dataframe[["SeasonAt-Bats", "SeasonHits"]].values,
         dtype=torch.float,
-        device=device,
     )
     first_name = pd_dataframe["FirstName"].values
     last_name = pd_dataframe["LastName"].values

examples/contrib/cevae/synthetic.py

@@ -13,6 +13,7 @@
     http://papers.nips.cc/paper/7223-causal-effect-inference-with-deep-latent-variable-models.pdf
     https://github.com/AMLab-Amsterdam/CEVAE
 """
+
 import argparse
 import logging
 

examples/contrib/funsor/hmm.py

@@ -42,6 +42,7 @@
 Fritz Obermeyer, Eli Bingham, Martin Jankowiak, Justin Chiu,
 Neeraj Pradhan, Alexander Rush, Noah Goodman. https://arxiv.org/abs/1902.03210
 """
+
 import argparse
 import functools
 import logging

examples/hmm.py

@@ -35,6 +35,7 @@
 Fritz Obermeyer, Eli Bingham, Martin Jankowiak, Justin Chiu,
 Neeraj Pradhan, Alexander Rush, Noah Goodman. https://arxiv.org/abs/1902.03210
 """
+
 import argparse
 import logging
 import sys

examples/lda.py

@@ -19,6 +19,7 @@
     "Pathwise gradients beyond the reparametrization trick"
     https://arxiv.org/pdf/1806.01851.pdf
 """
+
 import argparse
 import functools
 import logging

examples/rsa/schelling.py

@@ -10,6 +10,7 @@
 
 Taken from: http://forestdb.org/models/schelling.html
 """
+
 import argparse
 
 import torch

examples/rsa/schelling_false.py

@@ -11,6 +11,7 @@
 
 Taken from: http://forestdb.org/models/schelling-falsebelief.html
 """
+
 import argparse
 
 import torch

examples/vae/ss_vae_M2.py

@@ -8,7 +8,6 @@
 from utils.custom_mlp import MLP, Exp
 from utils.mnist_cached import MNISTCached, mkdir_p, setup_data_loaders
 from utils.vae_plots import mnist_test_tsne_ssvae, plot_conditional_samples_ssvae
-from visdom import Visdom
 
 import pyro
 import pyro.distributions as dist
@@ -246,10 +245,10 @@ def run_inference_for_epoch(data_loaders, losses, periodic_interval_batches):
 
         # extract the corresponding batch
         if is_supervised:
-            (xs, ys) = next(sup_iter)
+            xs, ys = next(sup_iter)
             ctr_sup += 1
         else:
-            (xs, ys) = next(unsup_iter)
+            xs, ys = next(unsup_iter)
 
         # run the inference for each loss with supervised or un-supervised
         # data as arguments
@@ -306,6 +305,8 @@ def main(args):
 
     viz = None
     if args.visualize:
+        from visdom import Visdom
+
         viz = Visdom()
         mkdir_p("./vae_results")
 

examples/vae/utils/mnist_cached.py

@@ -182,9 +182,9 @@ def target_transform(y):
                 self.data = MNISTCached.train_data_unsup
 
                 # making sure that the unsupervised labels are not available to inference
-                self.targets = (
-                    torch.Tensor(MNISTCached.train_labels_unsup.shape[0]).view(-1, 1)
-                ) * np.nan
+                self.targets = torch.full(
+                    (MNISTCached.train_labels_unsup.shape[0], 1), torch.nan
+                )
             else:
                 self.data, self.targets = (
                     MNISTCached.data_valid,

examples/vae/vae.py

@@ -6,7 +6,6 @@
 import numpy as np
 import torch
 import torch.nn as nn
-import visdom
 from utils.mnist_cached import MNISTCached as MNIST
 from utils.mnist_cached import setup_data_loaders
 from utils.vae_plots import mnist_test_tsne, plot_llk, plot_vae_samples
@@ -145,6 +144,8 @@ def main(args):
 
     # setup visdom for visualization
     if args.visdom_flag:
+        import visdom
+
         vis = visdom.Visdom()
 
     train_elbo = {}

pyro/__init__.py

@@ -32,7 +32,7 @@
 
 # Get the __version__ string from the auto-generated _version.py file, if exists.
 try:
-    from pyro._version import __version__  # type: ignore
+    from pyro._version import __version__
 except ImportError:
     __version__ = version_prefix
 

pyro/contrib/autoname/__init__.py

@@ -5,6 +5,7 @@
 The :mod:`pyro.contrib.autoname` module provides tools for automatically
 generating unique, semantically meaningful names for sample sites.
 """
+
 from pyro.contrib.autoname import named
 from pyro.contrib.autoname.autoname import autoname, sample
 from pyro.contrib.autoname.scoping import name_count, scope

pyro/contrib/autoname/named.py

@@ -49,6 +49,7 @@
 
 Authors: Fritz Obermeyer, Alexander Rush
 """
+
 import functools
 
 import pyro

pyro/contrib/autoname/scoping.py

@@ -6,6 +6,7 @@
 :func:`pyro.contrib.autoname.scope`, a tool for automatically appending
 a semantically meaningful prefix to names of sample sites.
 """
+
 import functools
 
 from pyro.poutine.messenger import Messenger

pyro/contrib/cevae/__init__.py

@@ -21,6 +21,7 @@
     | http://papers.nips.cc/paper/7223-causal-effect-inference-with-deep-latent-variable-models.pdf
     | https://github.com/AMLab-Amsterdam/CEVAE
 """
+
 import logging
 
 import torch

pyro/contrib/epidemiology/models.py

@@ -1277,14 +1277,9 @@ def transition(self, params, state, t):
     key=lambda name, vals=locals(): vals[name].__init__.__code__.co_firstlineno
 )
 __doc__ = "\n\n".join(
-    [
-        """
+    ["""
     {}
     ----------------------------------------------------------------
     .. autoclass:: pyro.contrib.epidemiology.models.{}
-    """.format(
-            re.sub("([A-Z][a-z]+)", r"\1 ", _name[:-5]), _name
-        )
-        for _name in __all__
-    ]
+    """.format(re.sub("([A-Z][a-z]+)", r"\1 ", _name[:-5]), _name) for _name in __all__]
 )

pyro/contrib/examples/polyphonic_data_loader.py

@@ -109,7 +109,7 @@ def load_data(dataset):
                 torch.Tensor
             )
             dset[k]["sequence_lengths"] = v["sequence_lengths"].to(
-                device=torch.Tensor().device
+                device=torch.get_default_device()
             )
     return dset
 

pyro/contrib/funsor/handlers/enum_messenger.py

@@ -5,6 +5,7 @@
 This file contains reimplementations of some of Pyro's core enumeration machinery,
 which should eventually be drop-in replacements for the current versions.
 """
+
 import functools
 import math
 from collections import OrderedDict

pyro/contrib/gp/kernels/__init__.py

@@ -52,9 +52,7 @@
 ]
 
 # Create sphinx documentation.
-__doc__ = "\n\n".join(
-    [
-        """
+__doc__ = "\n\n".join(["""
     {0}
     ----------------------------------------------------------------
     .. autoclass:: pyro.contrib.gp.kernels.{0}
@@ -63,9 +61,4 @@
         :special-members: __call__
         :show-inheritance:
         :member-order: bysource
-    """.format(
-            _name
-        )
-        for _name in __all__
-    ]
-)
+    """.format(_name) for _name in __all__])

pyro/contrib/gp/likelihoods/__init__.py

@@ -17,9 +17,7 @@
 
 
 # Create sphinx documentation.
-__doc__ = "\n\n".join(
-    [
-        """
+__doc__ = "\n\n".join(["""
     {0}
     ----------------------------------------------------------------
     .. autoclass:: pyro.contrib.gp.likelihoods.{0}
@@ -28,9 +26,4 @@
         :special-members: __call__
         :show-inheritance:
         :member-order: bysource
-    """.format(
-            _name
-        )
-        for _name in __all__
-    ]
-)
+    """.format(_name) for _name in __all__])

pyro/contrib/minipyro.py

@@ -14,6 +14,7 @@
 An accompanying example that makes use of this implementation can be
 found at examples/minipyro.py.
 """
+
 import random
 import warnings
 import weakref

pyro/contrib/timeseries/__init__.py

@@ -5,6 +5,7 @@
 The :mod:`pyro.contrib.timeseries` module provides a collection of Bayesian time series
 models useful for forecasting applications.
 """
+
 from pyro.contrib.timeseries.base import TimeSeriesModel
 from pyro.contrib.timeseries.gp import (
     DependentMaternGP,

pyro/distributions/spanning_tree.py

@@ -131,12 +131,7 @@ def log_partition_function(self):
         shift = 0.5 * log_diag
         laplacian = torch.eye(V) - (logits - shift - shift[:, None]).exp()
         truncated = laplacian[:-1, :-1]
-        try:
-            import gpytorch
-
-            log_det = gpytorch.lazy.NonLazyTensor(truncated).logdet()
-        except ImportError:
-            log_det = torch.linalg.cholesky(truncated).diag().log().sum() * 2
+        log_det = torch.linalg.cholesky(truncated).diag().log().sum() * 2
         return log_det + log_diag[:-1].sum()
 
     def log_prob(self, edges):

pyro/distributions/stable_log_prob.py

@@ -20,17 +20,17 @@ def create_integrator(num_points):
     from scipy.special import roots_legendre
 
     roots, weights = roots_legendre(num_points)
-    roots = torch.Tensor(roots).double()
-    weights = torch.Tensor(weights).double()
+    roots = torch.as_tensor(roots)
+    weights = torch.as_tensor(weights)
     log_weights = weights.log()
     half_roots = roots * 0.5
 
     def integrate(fn, domain):
         sl = [slice(None)] + (len(domain.shape) - 1) * [None]
-        half_roots_sl = half_roots[sl]
+        half_roots_sl = half_roots[tuple(sl)]
         value = domain[0] * (0.5 - half_roots_sl) + domain[1] * (0.5 + half_roots_sl)
         return (
-            torch.logsumexp(fn(value) + log_weights[sl], dim=0)
+            torch.logsumexp(fn(value) + log_weights[tuple(sl)], dim=0)
             + ((domain[1] - domain[0]) / 2).log()
         )