def _compute_columns(self):
if isinstance(self._engine, LlvmEvaluationEngine):
fnptr = self._mapnode.get_result()
rowindex = self._engine.rowindex
if rowindex:
nrows = rowindex.nrows
else:
nrows = self.dt.nrows
return core.columns_from_mixed(self._elems, self.dt.internal,
nrows, fnptr)
else:
ee = self._engine
_dt = ee.dt.internal
_ri = ee.rowindex
ncols = len(self._elems)
if ee.groupby:
opfirst = reduce_opcodes["first"]
n_reduce_cols = 0
for elem in self._elems:
if isinstance(elem, int):
is_groupby_col = elem in ee.groupby_cols
n_reduce_cols += is_groupby_col
else:
n_reduce_cols += elem.is_reduce_expr(ee)
expand_dataset = (n_reduce_cols < ncols)
columns = ee.groupby_cols + self._elems
self._names = ([ee.dt.names[i]
for i in ee.groupby_cols] + self._names)
for i, elem in enumerate(columns):
if isinstance(elem, int):
col = core.expr_column(_dt, elem, _ri)
if not expand_dataset:
col = core.expr_reduceop(opfirst, col, ee.groupby)
else:
col = elem.evaluate_eager(ee)
if expand_dataset and elem.is_reduce_expr(ee):
col = col.ungroup(ee.groupby)
columns[i] = col
else:
columns = [
core.expr_column(_dt, e, _ri)
if isinstance(e, int) else e.evaluate_eager(ee)
for e in self._elems
]
return core.columns_from_columns(columns)
def _compute_columns(self):
if isinstance(self._engine, LlvmEvaluationEngine):
fnptr = self._mapnode.get_result()
rowindex = self._engine.rowindex
if rowindex:
nrows = rowindex.nrows
else:
nrows = self.dt.nrows
return core.columns_from_mixed(self._elems, self.dt.internal,
nrows, fnptr)
else:
ee = self._engine
_dt = ee.dt.internal
_ri = ee.rowindex
columns = [
core.expr_column(_dt, e, _ri)
if isinstance(e, int) else e.evaluate_eager(ee)
for e in self._elems
]
return core.columns_from_columns(columns)
def evaluate_eager(self, ee):
self.resolve()
dt = self._dtexpr.get_datatable()
ri = self._dtexpr.get_rowindex()
return core.expr_column(dt.internal, self._colid, ri)