PERF: delay numexpr

Author: TomAugspurger

pandas/core/computation/__init__.py

@@ -1,23 +0,0 @@
-
-import warnings
-from distutils.version import LooseVersion
-
-_NUMEXPR_INSTALLED = False
-_MIN_NUMEXPR_VERSION = "2.4.6"
-
-try:
-    import numexpr as ne
-    ver = ne.__version__
-    _NUMEXPR_INSTALLED = ver >= LooseVersion(_MIN_NUMEXPR_VERSION)
-
-    if not _NUMEXPR_INSTALLED:
-        warnings.warn(
-            "The installed version of numexpr {ver} is not supported "
-            "in pandas and will be not be used\nThe minimum supported "
-            "version is {min_ver}\n".format(
-                ver=ver, min_ver=_MIN_NUMEXPR_VERSION), UserWarning)
-
-except ImportError:  # pragma: no cover
-    pass
-
-__all__ = ['_NUMEXPR_INSTALLED']

pandas/core/computation/check.py

@@ -0,0 +1,22 @@
+import warnings
+from distutils.version import LooseVersion
+
+_NUMEXPR_INSTALLED = False
+_MIN_NUMEXPR_VERSION = "2.4.6"
+
+try:
+    import numexpr as ne
+    ver = ne.__version__
+    _NUMEXPR_INSTALLED = ver >= LooseVersion(_MIN_NUMEXPR_VERSION)
+
+    if not _NUMEXPR_INSTALLED:
+        warnings.warn(
+            "The installed version of numexpr {ver} is not supported "
+            "in pandas and will be not be used\nThe minimum supported "
+            "version is {min_ver}\n".format(
+                ver=ver, min_ver=_MIN_NUMEXPR_VERSION), UserWarning)
+
+except ImportError:  # pragma: no cover
+    pass
+
+__all__ = ['_NUMEXPR_INSTALLED']

pandas/core/computation/eval.py

@@ -5,8 +5,6 @@
 
 import tokenize
 from pandas.io.formats.printing import pprint_thing
-from pandas.core.computation import _NUMEXPR_INSTALLED
-from pandas.core.computation.expr import Expr, _parsers, tokenize_string
 from pandas.core.computation.scope import _ensure_scope
 from pandas.compat import string_types
 from pandas.core.computation.engines import _engines
@@ -32,6 +30,7 @@ def _check_engine(engine):
     string engine
 
     """
+    from pandas.core.computation.check import _NUMEXPR_INSTALLED
 
     if engine is None:
         if _NUMEXPR_INSTALLED:
@@ -69,6 +68,8 @@ def _check_parser(parser):
     KeyError
       * If an invalid parser is passed
     """
+    from pandas.core.computation.expr import _parsers
+
     if parser not in _parsers:
         raise KeyError('Invalid parser {parser!r} passed, valid parsers are'
                        ' {valid}'.format(parser=parser, valid=_parsers.keys()))
@@ -129,6 +130,8 @@ def _convert_expression(expr):
 
 
 def _check_for_locals(expr, stack_level, parser):
+    from pandas.core.computation.expr import tokenize_string
+
     at_top_of_stack = stack_level == 0
     not_pandas_parser = parser != 'pandas'
 
@@ -252,6 +255,7 @@ def eval(expr, parser='pandas', engine=None, truediv=True,
     pandas.DataFrame.query
     pandas.DataFrame.eval
     """
+    from pandas.core.computation.expr import Expr
 
     inplace = validate_bool_kwarg(inplace, "inplace")
 

pandas/core/computation/expressions.py

@@ -9,7 +9,7 @@
 import warnings
 import numpy as np
 from pandas.core.common import _values_from_object
-from pandas.core.computation import _NUMEXPR_INSTALLED
+from pandas.core.computation.check import _NUMEXPR_INSTALLED
 from pandas.core.config import get_option
 
 if _NUMEXPR_INSTALLED:

pandas/core/frame.py

@@ -76,9 +76,7 @@
                                    create_block_manager_from_blocks)
 from pandas.core.series import Series
 from pandas.core.categorical import Categorical
-import pandas.core.computation.expressions as expressions
 import pandas.core.algorithms as algorithms
-from pandas.core.computation.eval import eval as _eval
 from pandas.compat import (range, map, zip, lrange, lmap, lzip, StringIO, u,
                            OrderedDict, raise_with_traceback)
 from pandas import compat
@@ -2296,6 +2294,8 @@ def eval(self, expr, inplace=False, **kwargs):
         >>> df.eval('a + b')
         >>> df.eval('c = a + b')
         """
+        from pandas.core.computation.eval import eval as _eval
+
         inplace = validate_bool_kwarg(inplace, 'inplace')
         resolvers = kwargs.pop('resolvers', None)
         kwargs['level'] = kwargs.pop('level', 0) + 1
@@ -3845,6 +3845,7 @@ def _combine_const(self, other, func, raise_on_error=True, try_cast=True):
 
     def _compare_frame_evaluate(self, other, func, str_rep, try_cast=True):
 
+        import pandas.core.computation.expressions as expressions
         # unique
         if self.columns.is_unique:
 
@@ -3997,6 +3998,7 @@ def combine_first(self, other):
         -------
         combined : DataFrame
         """
+        import pandas.core.computation.expressions as expressions
 
         def combiner(x, y, needs_i8_conversion=False):
             x_values = x.values if hasattr(x, 'values') else x
@@ -4032,6 +4034,7 @@ def update(self, other, join='left', overwrite=True, filter_func=None,
             If True, will raise an error if the DataFrame and other both
             contain data in the same place.
         """
+        import pandas.core.computation.expressions as expressions
         # TODO: Support other joins
         if join != 'left':  # pragma: no cover
             raise NotImplementedError("Only left join is supported")

pandas/core/internals.py

@@ -66,7 +66,6 @@
 from pandas._libs.tslib import Timedelta
 from pandas._libs.lib import BlockPlacement
 
-import pandas.core.computation.expressions as expressions
 from pandas.util._decorators import cache_readonly
 from pandas.util._validators import validate_bool_kwarg
 from pandas import compat
@@ -1395,6 +1394,8 @@ def where(self, other, cond, align=True, raise_on_error=True,
         -------
         a new block(s), the result of the func
         """
+        import pandas.core.computation.expressions as expressions
+
         values = self.values
         orig_other = other
         if transpose:

pandas/core/ops.py

@@ -16,7 +16,6 @@
 
 from pandas import compat
 from pandas.util._decorators import Appender
-import pandas.core.computation.expressions as expressions
 
 from pandas.compat import bind_method
 import pandas.core.missing as missing
@@ -668,8 +667,9 @@ def _arith_method_SERIES(op, name, str_rep, fill_zeros=None, default_axis=None,
     Wrapper function for Series arithmetic operations, to avoid
     code duplication.
     """
-
     def na_op(x, y):
+        import pandas.core.computation.expressions as expressions
+
         try:
             result = expressions.evaluate(op, str_rep, x, y,
                                           raise_on_error=True, **eval_kwargs)
@@ -1193,6 +1193,8 @@ def to_series(right):
 def _arith_method_FRAME(op, name, str_rep=None, default_axis='columns',
                         fill_zeros=None, **eval_kwargs):
     def na_op(x, y):
+        import pandas.core.computation.expressions as expressions
+
         try:
             result = expressions.evaluate(op, str_rep, x, y,
                                           raise_on_error=True, **eval_kwargs)
@@ -1349,6 +1351,8 @@ def _arith_method_PANEL(op, name, str_rep=None, fill_zeros=None,
     # copied from Series na_op above, but without unnecessary branch for
     # non-scalar
     def na_op(x, y):
+        import pandas.core.computation.expressions as expressions
+
         try:
             result = expressions.evaluate(op, str_rep, x, y,
                                           raise_on_error=True, **eval_kwargs)
@@ -1378,6 +1382,8 @@ def f(self, other):
 
 def _comp_method_PANEL(op, name, str_rep=None, masker=False):
     def na_op(x, y):
+        import pandas.core.computation.expressions as expressions
+
         try:
             result = expressions.evaluate(op, str_rep, x, y,
                                           raise_on_error=True)

pandas/core/panel.py

@@ -15,7 +15,6 @@
     is_string_like, is_scalar)
 from pandas.core.dtypes.missing import notna
 
-import pandas.core.computation.expressions as expressions
 import pandas.core.common as com
 import pandas.core.ops as ops
 import pandas.core.missing as missing
@@ -1500,6 +1499,8 @@ def _add_aggregate_operations(cls, use_numexpr=True):
         def _panel_arith_method(op, name, str_rep=None, default_axis=None,
                                 fill_zeros=None, **eval_kwargs):
             def na_op(x, y):
+                import pandas.core.computation.expressions as expressions
+
                 try:
                     result = expressions.evaluate(op, str_rep, x, y,
                                                   raise_on_error=True,

pandas/tests/computation/test_compat.py

@@ -5,13 +5,13 @@
 
 from pandas.core.computation.engines import _engines
 import pandas.core.computation.expr as expr
-from pandas.core.computation import _MIN_NUMEXPR_VERSION
+from pandas.core.computation.check import _MIN_NUMEXPR_VERSION
 
 
 def test_compat():
     # test we have compat with our version of nu
 
-    from pandas.core.computation import _NUMEXPR_INSTALLED
+    from pandas.core.computation.check import _NUMEXPR_INSTALLED
     try:
         import numexpr as ne
         ver = ne.__version__

pandas/tests/frame/test_query_eval.py

@@ -17,7 +17,7 @@
                                  makeCustomDataframe as mkdf)
 
 import pandas.util.testing as tm
-from pandas.core.computation import _NUMEXPR_INSTALLED
+from pandas.core.computation.check import _NUMEXPR_INSTALLED
 
 from pandas.tests.frame.common import TestData
 

pandas/util/testing.py

@@ -41,8 +41,6 @@
     StringIO, PY3
 )
 
-from pandas.core.computation import expressions as expr
-
 from pandas import (bdate_range, CategoricalIndex, Categorical, IntervalIndex,
                     DatetimeIndex, TimedeltaIndex, PeriodIndex, RangeIndex,
                     Index, MultiIndex,
@@ -2660,7 +2658,11 @@ def __exit__(self, exc_type, exc_value, traceback):
 
 
 @contextmanager
-def use_numexpr(use, min_elements=expr._MIN_ELEMENTS):
+def use_numexpr(use, min_elements=None):
+    from pandas.core.computation import expressions as expr
+    if min_elements is None:
+        min_elements = expr._MIN_ELEMENTS
+
     olduse = expr._USE_NUMEXPR
     oldmin = expr._MIN_ELEMENTS
     expr.set_use_numexpr(use)