Make optimizer classes extend abc.ABC and UserList

tests/gof/test_optdb.py

@@ -8,6 +8,9 @@ class TestDB:
         class Opt(opt.GlobalOptimizer):  # inheritance buys __hash__
             name = "blah"
 
+            def apply(self, fgraph):
+                pass
+
         db = DB()
         db.register("a", Opt())
 

theano/compile/mode.py

@@ -146,6 +146,9 @@ class AddFeatureOptimizer(gof.GlobalOptimizer):
         super().add_requirements(fgraph)
         fgraph.attach_feature(self.feature)
 
+    def apply(self, fgraph):
+        pass
+
 
 class PrintCurrentFunctionGraph(gof.GlobalOptimizer):
     """

theano/gof/opt.py

@@ -3,6 +3,7 @@ Defines the base class for optimizations as well as a certain
 amount of useful generic optimization tools.
 
 """
+import abc
 import contextlib
 import copy
 import inspect
@@ -12,7 +13,7 @@ import sys
 import time
 import traceback
 import warnings
-from collections import OrderedDict, defaultdict, deque
+from collections import OrderedDict, UserList, defaultdict, deque
 from collections.abc import Iterable
 from functools import reduce
 
@@ -43,7 +44,7 @@ class LocalMetaOptimizerSkipAssertionError(AssertionError):
     """
 
 
-class GlobalOptimizer:
+class GlobalOptimizer(abc.ABC):
     """
 
     A L{GlobalOptimizer} can be applied to an L{FunctionGraph} to transform it.
@@ -52,22 +53,7 @@ class GlobalOptimizer:
 
     """
 
-    def __hash__(self):
-        if not hasattr(self, "_optimizer_idx"):
-            self._optimizer_idx = _optimizer_idx[0]
-            _optimizer_idx[0] += 1
-        return self._optimizer_idx
-
-    def __eq__(self, other):
-        # added to override the  __eq__ implementation that may be inherited
-        # in subclasses from other bases.
-        return id(self) == id(other)
-
-    def __ne__(self, other):
-        # added to override the  __ne__ implementation that may be inherited
-        # in subclasses from other bases.
-        return id(self) != id(other)
-
+    @abc.abstractmethod
     def apply(self, fgraph):
         """
 
@@ -77,6 +63,7 @@ class GlobalOptimizer:
         L{InstanceFinder}, it can do so in its L{add_requirements} method.
 
         """
+        raise NotImplementedError()
 
     def optimize(self, fgraph, *args, **kwargs):
         """
@@ -108,6 +95,7 @@ class GlobalOptimizer:
           etc.
 
         """
+        pass
 
     def print_summary(self, stream=sys.stdout, level=0, depth=-1):
         name = getattr(self, "name", None)
@@ -124,14 +112,23 @@ class GlobalOptimizer:
                 " optimizer return profiling information."
             )
 
+    def __hash__(self):
+        if not hasattr(self, "_optimizer_idx"):
+            self._optimizer_idx = _optimizer_idx[0]
+            _optimizer_idx[0] += 1
+        return self._optimizer_idx
+
 
 class FromFunctionOptimizer(GlobalOptimizer):
     """A `GlobalOptimizer` constructed from a given function."""
 
     def __init__(self, fn, requirements=()):
-        self.apply = fn
+        self.fn = fn
         self.requirements = requirements
 
+    def apply(self, *args, **kwargs):
+        return self.fn(*args, **kwargs)
+
     def add_requirements(self, fgraph):
         for req in self.requirements:
             req(fgraph)
@@ -168,7 +165,7 @@ def inplace_optimizer(f):
     return rval
 
 
-class SeqOptimizer(GlobalOptimizer, list):
+class SeqOptimizer(GlobalOptimizer, UserList):
     """A `GlobalOptimizer` that applies a list of optimizers sequentially."""
 
     @staticmethod
@@ -198,7 +195,9 @@ class SeqOptimizer(GlobalOptimizer, list):
         """
         if len(opts) == 1 and isinstance(opts[0], (list, tuple)):
             opts = opts[0]
-        self[:] = opts
+
+        super().__init__(opts)
+
         self.failure_callback = kw.pop("failure_callback", None)
         assert len(kw) == 0
 
@@ -234,7 +233,7 @@ class SeqOptimizer(GlobalOptimizer, list):
             {},
         )
         try:
-            for optimizer in self:
+            for optimizer in self.data:
                 try:
                     nb_nodes_before = len(fgraph.apply_nodes)
                     t0 = time.time()
@@ -283,11 +282,8 @@ class SeqOptimizer(GlobalOptimizer, list):
             )
         return self.pre_profile
 
-    def __str__(self):
-        return f"SeqOpt({list.__str__(self)})"
-
     def __repr__(self):
-        return list.__repr__(self)
+        return f"SeqOpt({self.data})"
 
     def print_summary(self, stream=sys.stdout, level=0, depth=-1):
         name = getattr(self, "name", None)
@@ -297,7 +293,7 @@ class SeqOptimizer(GlobalOptimizer, list):
         # This way, -1 will do all depth
         if depth != 0:
             depth -= 1
-            for opt in self:
+            for opt in self.data:
                 opt.print_summary(stream, level=(level + 2), depth=depth)
 
     @staticmethod
@@ -1093,19 +1089,8 @@ def pre_constant_merge(vars):
     return list(map(recursive_merge, vars))
 
 
-########################
-#   Local Optimizers   #
-########################
-
-
-class LocalOptimizer:
-    """
-    A class for node-based optimizations.
-
-    Instances should implement the transform function,
-    and be passed to configure a fgraph-based Optimizer instance.
-
-    """
+class LocalOptimizer(abc.ABC):
+    """A node-based optimizer."""
 
     def __hash__(self):
         if not hasattr(self, "_optimizer_idx"):
@@ -1122,7 +1107,8 @@ class LocalOptimizer:
         """
         return None
 
-    def transform(self, node):
+    @abc.abstractmethod
+    def transform(self, node, *args, **kwargs):
         """
         Transform a subgraph whose output is `node`.
 
@@ -1142,7 +1128,7 @@ class LocalOptimizer:
 
         """
 
-        raise utils.MethodNotDefined("transform", type(self), self.__class__.__name__)
+        raise NotImplementedError()
 
     def add_requirements(self, fgraph):
         """
@@ -1150,8 +1136,7 @@ class LocalOptimizer:
         fgraph, this is the place to do it.
 
         """
-        # Added by default
-        # fgraph.attach_feature(toolbox.ReplaceValidate())
+        pass
 
     def print_summary(self, stream=sys.stdout, level=0, depth=-1):
         print(f"{' ' * level}{self.__class_.__name__} id={id(self)}", file=stream)
@@ -1277,16 +1262,16 @@ class LocalMetaOptimizer(LocalOptimizer):
 
 
 class FromFunctionLocalOptimizer(LocalOptimizer):
-    """
-    WRITEME
-
-    """
+    """An optimizer constructed from a given function."""
 
     def __init__(self, fn, tracks=None, requirements=()):
-        self.transform = fn
+        self.fn = fn
         self._tracks = tracks
         self.requirements = requirements
 
+    def transform(self, *args, **kwargs):
+        return self.fn(*args, **kwargs)
+
     def add_requirements(self, fgraph):
         for req in self.requirements:
             req(fgraph)