Feature/optional dependencies

.readthedocs.yaml

@@ -14,4 +14,4 @@ python:
   install:
   - method: pip
     path: .
-    extra_requirements: [docs]
+    extra_requirements: [docs, neural]

docs/index.rst

@@ -32,7 +32,13 @@ Install ``OTT`` from `PyPI <https://pypi.org/project/ott-jax/>`_ as:
 
     pip install ott-jax
 
-or with ``conda`` via `conda-forge`_ as:
+or with the :mod:`neural OT solvers <ott.solvers.nn>` dependencies:
+
+.. code-block:: bash
+
+    pip install 'ott-jax[neural]'
+
+or using `conda`_ as:
 
 .. code-block:: bash
 
@@ -139,4 +145,4 @@ Packages
 .. _auto-vectorization (VMAP): https://jax.readthedocs.io/en/latest/jax.html#vectorization-vmap
 .. _automatic: https://jax.readthedocs.io/en/latest/jax.html#automatic-differentiation
 .. _implicit: https://jax.readthedocs.io/en/latest/_autosummary/jax.custom_jvp.html#jax.custom_jvp
-.. _conda-forge: https://anaconda.org/conda-forge/ott-jax
+.. _conda: https://anaconda.org/conda-forge/ott-jax

docs/spelling/misc.txt

@@ -1,3 +1,4 @@
+alg
 arg
 args
 cond

docs/spelling/technical.txt

@@ -19,6 +19,7 @@ Kullback
 Leibler
 Mahalanobis
 Monge
+Moreau
 SGD
 Schur
 Seidel
@@ -73,6 +74,7 @@ jax
 jit
 jitting
 linearization
+linearized
 logit
 macOS
 methylation
@@ -95,6 +97,7 @@ preconditioner
 preprocess
 preprocessing
 proteome
+prox
 quantile
 quantiles
 quantizes
@@ -105,6 +108,7 @@ renormalize
 reproducibility
 rescale
 rescaled
+rescaling
 reweighted
 reweighting
 reweightings

pyproject.toml

@@ -15,12 +15,10 @@ authors = [
 dependencies = [
     "jax>=0.1.67",
     "jaxlib>=0.1.47",
-    "jaxopt>=0.5.5",
-    # numba/numpy compatibility issue in JAXOPT.
+    "jaxopt>=0.8",
+    "lineax>=0.0.1; python_version >= '3.9'",
+    # https://github.com/numba/numba/issues/9130
     "numpy>=1.18.4, <1.25.0",
-    "flax>=0.6.6",
-    "optax>=0.1.1",
-    "lineax>=0.0.1; python_version >= '3.9'"
 ]
 keywords = [
     "optimal transport",
@@ -59,6 +57,10 @@ Documentation = "https://ott-jax.readthedocs.io"
 Changelog = "https://github.com/ott-jax/ott/releases"
 
 [project.optional-dependencies]
+neural = [
+    "flax>=0.6.6",
+    "optax>=0.1.1",
+]
 dev = [
     "pre-commit>=2.16.0",
     "tox>=4",
@@ -180,7 +182,7 @@ legacy_tox_ini = """
     skip_missing_interpreters = true
 
     [testenv]
-    extras = test
+    extras = test,neural
     pass_env = CUDA_*,PYTEST_*,CI
     commands_pre =
         gpu: python -I -m pip install "jax[cuda]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
@@ -200,7 +202,7 @@ legacy_tox_ini = """
     [testenv:lint-docs]
     description = Lint the documentation.
     deps =
-    extras = docs
+    extras = docs,neural
     ignore_errors = true
     allowlist_externals = make
     pass_env = PYENCHANT_LIBRARY_PATH
@@ -214,7 +216,7 @@ legacy_tox_ini = """
     description = Build the documentation.
     use_develop = true
     deps =
-    extras = docs
+    extras = docs,neural
     allowlist_externals = make
     changedir = {tox_root}{/}docs
     commands =

src/ott/initializers/__init__.py

@@ -11,4 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from . import linear, nn, quadratic
+import contextlib
+
+from . import linear, quadratic
+
+with contextlib.suppress(ImportError):
+  from . import nn
+del contextlib

src/ott/solvers/__init__.py

@@ -11,4 +11,10 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from . import linear, nn, quadratic
+import contextlib
+
+from . import linear, quadratic, was_solver
+
+with contextlib.suppress(ImportError):
+  from . import nn
+del contextlib

src/ott/solvers/linear/sinkhorn.py

@@ -298,7 +298,7 @@ class SinkhornOutput(NamedTuple):
       ``max_iterations // inner_iterations`` where those were the parameters
       passed on to the :class:`ott.solvers.linear.sinkhorn.Sinkhorn` solver.
       For each entry indexed at ``i``, ``errors[i]`` can be either a real
-      nonnegative value (meaning the algorithm recorded that error at the
+      non-negative value (meaning the algorithm recorded that error at the
       ``i * inner_iterations`` iteration), a ``jnp.inf`` value (meaning the
       algorithm computed that iteration but did not compute its error, because,
       for instance, ``i < min_iterations // inner_iterations``), or a ``-1``,

src/ott/tools/__init__.py

@@ -11,12 +11,17 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import contextlib
+
 from . import (
     gaussian_mixture,
     k_means,
-    map_estimator,
     plot,
     segment_sinkhorn,
     sinkhorn_divergence,
     soft_sort,
 )
+
+with contextlib.suppress(ImportError):
+  from . import map_estimator
+del contextlib

src/ott/tools/gaussian_mixture/fit_gmm_pair.py

@@ -83,7 +83,6 @@
 
 import jax
 import jax.numpy as jnp
-import optax
 
 from ott.tools.gaussian_mixture import (
     fit_gmm,
@@ -235,6 +234,7 @@ def get_m_step_fn(learning_rate: float, objective_fn, jit: bool):
   Returns:
     A function that performs the M-step of EM.
   """
+  import optax
 
   def _m_step_fn(
       pair: gaussian_mixture_pair.GaussianMixturePair,

src/ott/tools/map_estimator.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 import collections
 import functools
 from typing import Any, Callable, Dict, Iterator, Optional, Tuple
@@ -22,7 +21,7 @@
 from flax.core import frozen_dict
 from flax.training import train_state
 
-from ott.solvers.nn.models import ModelBase
+from ott.solvers.nn import models
 
 __all__ = ["MapEstimator"]
 
@@ -65,7 +64,7 @@ class MapEstimator:
   def __init__(
       self,
       dim_data: int,
-      model: ModelBase,
+      model: models.ModelBase,
       optimizer: Optional[optax.OptState] = None,
       fitting_loss: Optional[Callable[[jnp.ndarray, jnp.ndarray],
                                       float]] = None,
@@ -94,7 +93,7 @@ def __init__(
   def setup(
       self,
       dim_data: int,
-      neural_net: ModelBase,
+      neural_net: models.ModelBase,
       optimizer: optax.OptState,
   ):
     """Setup all components required to train the network."""