def evaluate(self):
"""
Evaluates this `GrizzlySeries` and returns itself.
Evaluation reduces the currently stored computation to a physical value
by compiling and running a Weld program. If this `GrizzlySeries` refers
to a physical value and no computation, no program is compiled, and this
method returns `self` unmodified.
"""
if not self.is_value:
result = self.weld_value_.evaluate()
# TODO(shoumik): it's unfortunate that this copy is needed, but
# things are breaking without it (even if we hold a reference to
# the WeldContext). DEBUG ME!
if isinstance(result[0], wenp.weldbasearray):
super(GrizzlySeries, self).__init__(result[0].copy2numpy())
else:
super(GrizzlySeries, self).__init__(result[0])
setattr(self, "evaluating_", 0)
self.weld_value_ = identity(PhysicalValue(self.values,\
self.output_type, GrizzlySeries._encoder),
GrizzlySeries._decoder)
delattr(self, "evaluating_")
return self
def __new__(cls, data, dtype=None, index=None, **kwargs):
"""
Internal initialization. Tests below are for internal visibility only.
>>> x = GrizzlySeries([1,2,3])
>>> x
0 1
1 2
2 3
dtype: int64
>>> x.__class__
<class 'weld.grizzly.series.GrizzlySeries'>
>>> x = GrizzlySeries(np.ones(5))
>>> x.__class__
<class 'weld.grizzly.series.GrizzlySeries'>
>>> y = GrizzlySeries(['hi', 'bye']) # Unsupported
>>> y.__class__
<class 'pandas.core.series.Series'>
>>> y = GrizzlySeries([1, 2, 3], index=[1, 0, 2]) # Unsupported
>>> y.__class__
<class 'pandas.core.series.Series'>
"""
s = None
if isinstance(data, WeldLazy):
self = super(GrizzlySeries, cls).__new__(cls)
super(GrizzlySeries, self).__init__(None, dtype=dtype, **kwargs)
self.weld_value_ = data
return self
elif index is not None and not isinstance(index, pd.RangeIndex):
# TODO(shoumik): This is probably incomplete, since we could have a
# RangeIndex that does not capture the full span of the data, has a
# non-zero step, etc.
return pd.Series(data, dtype=dtype, index=index, **kwargs)
elif len(kwargs) != 0:
return pd.Series(data, dtype=dtype, index=index, **kwargs)
elif not isinstance(data, np.ndarray):
# First, convert the input into a Series backed by an ndarray.
s = pd.Series(data, dtype=dtype, index=index, **kwargs)
data = s.values
# Try to create a Weld type for the input.
weld_type = GrizzlySeries._supports_grizzly(data)
if weld_type is not None:
self = super(GrizzlySeries, cls).__new__(cls)
super(GrizzlySeries, self).__init__(data, dtype=dtype, **kwargs)
self.weld_value_ = identity(
PhysicalValue(data, weld_type, GrizzlySeries._encoder),
GrizzlySeries._decoder)
return self
# Don't re-convert values if we did it once already -- it's expensive.
return s if s is not None else pd.Series(
data, dtype=dtype, index=index, **kwargs)
def __init__(self, data, dtype=None, name=None):
"""
Initialize a new GrizzlySeries.
>>> x = GrizzlySeries([1,2,3])
>>> x
0 1
1 2
2 3
dtype: int64
"""
s = None
if isinstance(data, WeldLazy):
self.name = name
self.values_ = None
self.weld_value = data
return
self.name = name
if isinstance(data, list) and len(data) > 0 and isinstance(
data[0], str):
# Try to convert a list of strings into a supported Numpy array.
self.values_ = np.array(data, dtype='S')
elif isinstance(data, pd.Series):
if self.name is None:
self.name = data.name
if data.values.dtype == 'object' and len(data) > 0 and isinstance(
data[0], str):
self.values_ = np.array(data, dtype='S')
else:
self.values_ = data.values
elif not isinstance(data, np.ndarray):
# First, convert the input into a Numpy array.
self.values_ = np.array(data, dtype=dtype)
else:
self.values_ = data
# Try to create a Weld type for the input.
weld_type = GrizzlySeries.supported(self.values_)
if weld_type:
self.weld_value = identity(
PhysicalValue(self.values_, weld_type, GrizzlySeries._encoder),
GrizzlySeries._decoder)
else:
raise GrizzlyError("unsupported data type '{}'".format(
self.values_.dtype))