ak.layout.BitMaskedArray¶
Like ak.layout.ByteMaskedArray, BitMaskedArray implements an
ak.types.OptionType with two arrays, mask and content.
However, the boolean mask values are packed into a bitmap.
BitMaskedArray has an additional parameter, lsb_order; if True,
the position of each bit is in
Least-Significant Bit order
(LSB):
is_valid[j] = bool(mask[j // 8] & (1 << (j % 8))) == valid_when
If False, the position of each bit is in Most-Significant Bit order (MSB):
is_valid[j] = bool(mask[j // 8] & (128 >> (j % 8))) == valid_when
Note that NumPy’s
unpackbits
function before version 1.17 has MSB order, but now it is configurable
(bitorder="little" for LSB, bitorder="big" for MSB).
If the logical size of the array is not a multiple of 8, the mask
has to be padded. Thus, an explicit length is also part of the
class’s definition.
Below is a simplified implementation of a RecordArray class in pure Python
that exhaustively checks validity in its constructor (see
ak.is_valid) and can generate random valid arrays. The
random_number() function returns a random float and the
random_length(minlen) function returns a random int that is at least
minlen. The RawArray class represents simple, one-dimensional data.
class BitMaskedArray(Content):
def __init__(self, mask, content, valid_when, length, lsb_order):
assert isinstance(mask, np.ndarray)
assert isinstance(content, Content)
assert isinstance(valid_when, bool)
assert isinstance(length, int) and length >= 0
assert isinstance(lsb_order, bool)
assert len(mask) <= len(content)
self.mask = mask
self.content = content
self.valid_when = valid_when
self.length = length
self.lsb_order = lsb_order
@staticmethod
def random(minlen, choices):
mask = []
for i in range(random_length(minlen)):
mask.append(bool(random.randint(0, 1)))
lsb_order = bool(random.randint(0, 1))
bitmask = np.packbits(np.array(mask, dtype=np.uint8),
bitorder=("little" if lsb_order else "big"))
content = random.choice(choices).random(len(mask), choices)
return BitMaskedArray(bitmask, content, bool(random.randint(0, 1)),
len(mask), lsb_order)
def __len__(self):
return self.length
def __getitem__(self, where):
if isinstance(where, int):
assert 0 <= where < len(self)
if self.lsb_order:
bit = bool(self.mask[where // 8] & (1 << (where % 8)))
else:
bit = bool(self.mask[where // 8] & (128 >> (where % 8)))
if bit == self.valid_when:
return self.content[where]
else:
return None
elif isinstance(where, slice) and where.step is None:
# In general, slices must convert BitMaskedArray to ByteMaskedArray.
bytemask = np.unpackbits(self.mask,
bitorder=("little" if self.lsb_order else "big"))
return ByteMaskedArray(bytemask[where.start:where.stop],
self.content[where.start:where.stop])
elif isinstance(where, str):
return BitMaskedArray(self.mask, self.content[where])
else:
raise AssertionError(where)
def __repr__(self):
return ("BitMaskedArray(" + repr(self.mask) + ", " + repr(self.content)
+ ", " + repr(self.valid_when) + ", " + repr(self.length)
+ ", " + repr(self.lsb_order) + ")")
def xml(self, indent="", pre="", post=""):
out = indent + pre + "<BitMaskedArray>\n"
out += indent + " <valid_when>" + repr(self.valid_when) + "<valid_when>\n"
out += indent + " <length>" + repr(self.length) + "<length>\n"
out += indent + " <lsb_order>" + repr(self.lsb_order) + "<lsb_order>\n"
out += indent + " <mask>" + " ".join(str(x) for x in self.mask) + "</mask>\n"
out += self.content.xml(indent + " ", "<content>", "</content>\n")
out += indent + "</BitMaskedArray>\n"
return out
Here is an example:
BitMaskedArray(np.array([ 40, 173, 59, 104, 182, 116], dtype=np.uint8),
RawArray([5.5, 6.6, 1.5, 3.2, 9.8, 0.4, 5.7, 1.5, 0.2, 6.1, 5.4, 4.3, 5.9,
10.1, -2.3, 5.8, 3.4, 5.6, 6.2, 8.8, 3.1, 7.0, 1.2, 7.3, 5.8, 8.3,
9.7, 5.2, 3.4, 5.8, 1.7, 4.3, 5.8, 1.2, 1.7, 3.6, 4.4, 9.7, 5.0,
4.3, 7.8, 6.1, 3.3, 7.9, 7.1, 6.5, -0.6, 8.2, 3.7, 4.6, 3.9, 7.5]),
False,
46,
False)
<BitMaskedArray>
<valid_when>False<valid_when>
<length>46<length>
<lsb_order>False<lsb_order>
<mask>40 173 59 104 182 116</mask>
<content><RawArray>
<ptr>5.5 6.6 1.5 3.2 9.8 0.4 5.7 1.5 0.2 6.1 5.4 4.3 5.9 10.1 -2.3 5.8 3.4 5.6 6.2
8.8 3.1 7.0 1.2 7.3 5.8 8.3 9.7 5.2 3.4 5.8 1.7 4.3 5.8 1.2 1.7 3.6 4.4 9.7 5.0
4.3 7.8 6.1 3.3 7.9 7.1 6.5 -0.6 8.2 3.7 4.6 3.9 7.5</ptr>
</RawArray></content>
</BitMaskedArray>
which represents the following logical data.
[5.5, 6.6, None, 3.2, None, 0.4, 5.7, 1.5, None, 6.1, None, 4.3, None, None, -2.3, None, 3.4,
5.6, None, None, None, 7.0, None, None, 5.8, None, None, 5.2, None, 5.8, 1.7, 4.3, None,
1.2, None, None, 4.4, None, None, 4.3, 7.8, None, None, None, 7.1, None]
This is equivalent to all of Apache Arrow’s array types because they all
use bitmaps
to mask their data, with valid_when=True and lsb_order=True.
In addition to the properties and methods described in ak.layout.Content, a BitMaskedArray has the following.
ak.layout.BitMaskedArray.__init__¶
-
ak.layout.BitMaskedArray.__init__(mask, content, valid_when, length, lsb_order, identities=None, parameters=None)¶
ak.layout.BitMaskedArray.project¶
-
ak.layout.BitMaskedArray.project(mask=None)¶
Returns a non-ak.types.OptionType array containing only the valid elements.
If mask is a signed 8-bit ak.layout.Index in which 0 means valid
and 1 means missing, this mask is unioned with the BitMaskedArray’s
mask (after converting to 8-bit and to valid_when=False to match this mask).
ak.layout.BitMaskedArray.bytemask¶
-
ak.layout.BitMaskedArray.bytemask()¶
Returns an array of 8-bit values in which 0 means valid and 1 means missing.
ak.layout.BitMaskedArray.simplify¶
-
ak.layout.BitMaskedArray.simplify()¶
Combines this node with its content if the content also has
ak.types.OptionType; otherwise, this is a pass-through.
In all cases, the output has the same logical meaning as the input.
This method only operates one level deep.
ak.layout.BitMaskedArray.toByteMaskedArray¶
-
ak.layout.BitMaskedArray.toByteMaskedArray()¶
Converts to the equivalent ak.layout.ByteMaskedArray.
ak.layout.BitMaskedArray.toIndexedOptionArray¶
-
ak.layout.BitMaskedArray.toIndexedOptionArray()¶
Converts to the equivalent ak.layout.IndexedOptionArray.