def block_widths(self):
"""Gets the widths of the blocks.
Note: This works with the property structure `_widths_cache` to avoid
having to recompute these values each time they are needed.
"""
if self._widths_cache is None:
if not isinstance(self._partitions_cache[0][0].width(), int):
try:
# The first column will have the correct lengths. We have an
# invariant that requires that all blocks be the same width in a
# column of blocks.
self._widths_cache = np.array(
ray.get([
obj.width().oid
for obj in self._partitions_cache[0]
]) if len(self._partitions_cache) > 0 else [])
except RayTaskError as e:
handle_ray_task_error(e)
except AttributeError:
self._widths_cache = np.array([
obj.width() if isinstance(obj.width(), int) else
ray.get(obj.width().oid)
for obj in self._partitions_cache[0]
])
else:
self._widths_cache = np.array([
obj.width() if isinstance(obj.width(), int) else ray.get(
obj.width().oid) for obj in self._partitions_cache[0]
])
return self._widths_cache
def get(self):
"""Gets the object out of the plasma store.
Returns:
The object from the plasma store.
"""
if len(self.call_queue):
self.drain_call_queue()
try:
return ray.get(self.oid)
except RayTaskError as e:
handle_ray_task_error(e)
def ip(self):
if self._ip_cache is None:
if len(self.call_queue):
self.drain_call_queue()
else:
self._ip_cache = self.apply(lambda df: df)._ip_cache
if isinstance(self._ip_cache, ObjectIDType):
try:
self._ip_cache = ray.get(self._ip_cache)
except RayTaskError as e:
handle_ray_task_error(e)
return self._ip_cache
def width(self):
if self._width_cache is None:
if len(self.call_queue):
self.drain_call_queue()
else:
self._length_cache, self._width_cache = get_index_and_columns.remote(
self.oid)
if isinstance(self._width_cache, ray.ObjectID):
try:
self._width_cache = ray.get(self._width_cache)
except RayTaskError as e:
handle_ray_task_error(e)
return self._width_cache
def block_lengths(self):
"""Gets the lengths of the blocks.
Note: This works with the property structure `_lengths_cache` to avoid
having to recompute these values each time they are needed.
"""
if self._lengths_cache is None:
try:
# The first column will have the correct lengths. We have an
# invariant that requires that all blocks be the same length in a
# row of blocks.
self._lengths_cache = np.array(
ray.get([obj.length().oid for obj in self._partitions_cache.T[0]])
if len(self._partitions_cache.T) > 0
else []
)
except RayTaskError as e:
handle_ray_task_error(e)
return self._lengths_cache
def wait(self):
self.drain_call_queue()
try:
ray.wait([self.oid])
except RayTaskError as e:
handle_ray_task_error(e)
