import logging
import os
from pathlib import Path
from typing import Any, List, Mapping, Union
import numpy as np
from bed_reader import get_num_threads, open_bed
from .bed_reader import ( # type: ignore
encode1_f32,
encode1_f64,
encode1_i8,
write_f32,
write_f64,
write_i8,
)
[docs]
class create_bed:
"""
Open a file to write values into PLINK .bed format.
Values may be given in a SNP-by-SNP (or individual-by-individual) manner.
For large datasets, `create_bed` requires less memory than :func:`to_bed`.
Parameters
----------
location: pathlib.Path or str, optional
local .bed file to create.
iid_count: int:
The number of individuals (samples).
sid_count: int:
The number of SNPs (variants).
properties: dict, optional
A dictionary of property names and values to write to the .fam and .bim files.
Any properties not mentioned will be filled in with default values.
The possible property names are:
"fid" (family id), "iid" (individual or sample id), "father" (father id),
"mother" (mother id), "sex", "pheno" (phenotype), "chromosome", "sid"
(SNP or variant id), "cm_position" (centimorgan position), "bp_position"
(base-pair position), "allele_1", "allele_2".
The values are lists or arrays. See example, below.
count_A1: bool, optional
True (default) to count the number of A1 alleles (the PLINK standard).
False to count the number of A2 alleles.
fam_location: pathlib.Path or str, optional
Path to the file containing information about each individual (sample).
Defaults to replacing the .bed file’s suffix with .fam.
bim_location: pathlib.Path or str, optional
Path to the file containing information about each SNP (variant).
Defaults to replacing the .bed file’s suffix with .bim.
major: str, optional
Use "SNP" (default) to write the file is usual SNP-major mode. This makes reading
the data SNP-by-SNP faster. Use "individual" to write the file in the uncommon
individual-major mode.
force_python_only
If False (default), uses the faster Rust helper functions; otherwise it uses the slower
pure Python code.
num_threads: None or int, optional
Not currently used.
create_bed Errors
-----------------
Raises an error if you write the wrong number of vectors or if any vector has the wrong length.
Also, all vector values must be 0, 1, 2, or missing. If floats, missing is ``np.nan``. If integers, missing is -127.
Behind the scenes, `create_bed` first creates a temporary bed file. At the end, if there are no errors,
it renames the temporary file to the final file name. This helps prevent creation
of corrupted files.
Examples
--------
In this example, all properties are given and we write
the data out SNP-by-SNP. The data is floats.
.. doctest::
>>> import numpy as np
>>> from bed_reader import create_bed, tmp_path
>>>
>>> output_file = tmp_path() / "small.bed"
>>> properties = {
... "fid": ["fid1", "fid1", "fid2"],
... "iid": ["iid1", "iid2", "iid3"],
... "father": ["iid23", "iid23", "iid22"],
... "mother": ["iid34", "iid34", "iid33"],
... "sex": [1, 2, 0],
... "pheno": ["red", "red", "blue"],
... "chromosome": ["1", "1", "5", "Y"],
... "sid": ["sid1", "sid2", "sid3", "sid4"],
... "cm_position": [100.4, 2000.5, 4000.7, 7000.9],
... "bp_position": [1, 100, 1000, 1004],
... "allele_1": ["A", "T", "A", "T"],
... "allele_2": ["A", "C", "C", "G"],
... }
>>> with create_bed(output_file, iid_count=3, sid_count=4, properties=properties) as bed_writer:
... bed_writer.write([1.0, 2.0, 0.0])
... bed_writer.write([0.0, 0.0, 1.0])
... bed_writer.write([np.nan, np.nan, 2.0])
... bed_writer.write([0.0, 2.0, 0.0])
In this next example, no properties are given, so default values are assigned.
Also, we write out ints. Finally, we write the same data out as above,
but individual-by-individual.
If we then read the new file and list the chromosome property,
it is an array of '0's, the default chromosome value.
.. doctest::
>>> output_file2 = tmp_path() / "small2.bed"
>>> with create_bed(output_file2, iid_count=3, sid_count=4, major="individual") as bed_writer:
... bed_writer.write([1, 0, -127, 0])
... bed_writer.write([2, 0, -127, 2])
... bed_writer.write([0, 1, 2, 0])
>>>
>>> from bed_reader import open_bed
>>> with open_bed(output_file2) as bed2:
... print(bed2.chromosome)
['0' '0' '0' '0']
"""
def __init__(
self,
location: Union[str, Path],
iid_count: int,
sid_count: int,
properties: Mapping[str, List[Any]] = {},
count_A1: bool = True,
fam_location: Union[str, Path] = None,
bim_location: Union[str, Path] = None,
major: str = "SNP",
force_python_only: bool = False,
num_threads=None,
):
self.location = Path(location)
self.iid_count = iid_count
self.sid_count = sid_count
self.count_A1 = count_A1
self.force_python_only = force_python_only
self.num_threads = num_threads or os.cpu_count()
if major == "SNP":
self.major_count = sid_count
self.minor_count = iid_count
elif major == "individual":
self.major_count = iid_count
self.minor_count = sid_count
else:
raise ValueError(f"major must be 'SNP' or 'individual', not '{major}'")
self.major = major
self.minor_count_div4 = (self.minor_count - 1) // 4 + 1
self.buffer = np.zeros(self.minor_count_div4, dtype=np.uint8)
if not count_A1:
self.zero_code = 0b00
self.two_code = 0b11
else:
self.zero_code = 0b11
self.two_code = 0b00
fam_location = (
Path(fam_location)
if fam_location is not None
else self._replace_extension(self.location, "fam")
)
bim_location = (
Path(bim_location)
if bim_location is not None
else self._replace_extension(self.location, "bim")
)
properties, _ = open_bed._fix_up_properties(
properties,
iid_count,
sid_count,
use_fill_sequence=True,
)
open_bed._write_fam_or_bim(self.location, properties, "fam", fam_location)
open_bed._write_fam_or_bim(self.location, properties, "bim", bim_location)
self.temp_filepath = self.location.with_suffix(".bed_temp")
self.file_pointer = open(self.temp_filepath, "wb")
# see http://zzz.bwh.harvard.edu/plink/binary.shtml
self.file_pointer.write(bytes(bytearray([0b01101100]))) # magic numbers
self.file_pointer.write(bytes(bytearray([0b00011011]))) # magic numbers
if self.major == "SNP":
self.file_pointer.write(bytes(bytearray([0b00000001]))) # snp major
else:
assert self.major == "individual" # real assert
self.file_pointer.write(bytes(bytearray([0b00000000])))
self.write_count = 0
if os.path.exists(self.location):
os.unlink(self.location)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_value, traceback):
self.close()
[docs]
def close(self):
"""
Close the bed_writer, writing the file to disk. If you use
:class:`create_bed` with the `with` statement,
you don't need to use this.
See :class:`create_bed` for more information.
"""
if self.file_pointer is not None:
try:
self.file_pointer.close()
self.file_pointer = None
if self.write_count != self.major_count:
raise ValueError(
f"Attempt to write fewer vectors ({self.write_count}) than expected ({self.major_count})"
)
os.rename(self.temp_filepath, self.location)
except Exception as e:
self._clean_up_error()
raise e
else:
self._clean_up_error()
def _clean_up_error(self):
try:
os.unlink(self.temp_filepath)
except Exception:
pass
@staticmethod
def _replace_extension(location, extension):
if open_bed._is_url(location):
return location.parent / (location.stem + "." + extension)
[docs]
def write(self, vector):
"""
Write a vector of values to the bed_writer.
See :class:`create_bed` for more information.
"""
if self.file_pointer is None:
raise RuntimeError("Attempt to write after file was closed for writing.")
try:
self.write_count += 1
if self.write_count > self.major_count:
raise ValueError(
f"Attempt to write more vectors ({self.write_count}) than expected ({self.major_count})"
)
vector = _fix_up_vector(vector)
if len(vector) != self.minor_count:
raise ValueError(
f"Expected vector to {self.minor_count} values, got {len(vector)} instead"
)
self._internal_write(vector)
except Exception as e:
self.file_pointer.close()
self.file_pointer = None # Prevent further writes
raise e # Re-raise the exception to handle it externally
def _internal_write(self, vector):
if not self.force_python_only:
if vector.dtype == np.int8:
encode1_i8(self.count_A1, vector, self.buffer, self.num_threads)
elif vector.dtype == np.float32:
encode1_f32(self.count_A1, vector, self.buffer, self.num_threads)
elif vector.dtype == np.float64:
encode1_f64(self.count_A1, vector, self.buffer, self.num_threads)
else:
raise ValueError(
f"dtype '{vector.dtype}' not known, only 'int8', 'float32', and 'float64' are allowed."
)
self.file_pointer.write(self.buffer)
else:
for minor_by_four in range(0, self.minor_count, 4):
vals_for_this_byte = vector[minor_by_four : minor_by_four + 4]
byte = 0b00000000
for val_index in range(len(vals_for_this_byte)):
val_for_byte = vals_for_this_byte[val_index]
if val_for_byte == 0:
code = self.zero_code
elif val_for_byte == 1:
code = 0b10 # backwards on purpose
elif val_for_byte == 2:
code = self.two_code
elif (vector.dtype == np.int8 and val_for_byte == -127) or np.isnan(
val_for_byte
):
code = 0b01 # backwards on purpose
else:
raise ValueError(
"Attempt to write illegal value to .bed file. "
+ "Only 0,1,2,missing allowed."
)
byte |= code << (val_index * 2)
# This is writing one byte at a time
self.file_pointer.write(bytes(bytearray([byte])))
def _fix_up_vector(input):
if not isinstance(input, np.ndarray):
return _fix_up_vector(np.array(input))
if np.issubdtype(input.dtype, np.integer) and input.dtype != np.int8:
return _fix_up_vector(np.array(input, dtype=np.int8))
elif np.issubdtype(input.dtype, np.floating) and input.dtype not in (
np.float32,
np.float64,
):
return _fix_up_vector(np.array(input, dtype=np.float32))
if len(input.shape) != 1:
raise ValueError("vector should be one dimensional")
return input
[docs]
def to_bed(
filepath: Union[str, Path],
val: np.ndarray,
properties: Mapping[str, List[Any]] = {},
count_A1: bool = True,
fam_filepath: Union[str, Path] = None,
bim_filepath: Union[str, Path] = None,
major: str = "SNP",
force_python_only: bool = False,
num_threads=None,
):
"""
Write values to a file in PLINK .bed format.
If your data is too large to fit in memory, use :class:`create_bed` instead.
Parameters
----------
filepath:
.bed file to write to.
val: array-like:
A two-dimension array (or array-like object) of values. The values should
be (or be convertible to) all floats or all integers.
The values should be 0, 1, 2, or missing.
If floats, missing is ``np.nan``. If integers, missing is -127.
properties: dict, optional
A dictionary of property names and values to write to the .fam and .bim files.
Any properties not mentioned will be filled in with default values.
The possible property names are:
"fid" (family id), "iid" (individual or sample id), "father" (father id),
"mother" (mother id), "sex", "pheno" (phenotype), "chromosome", "sid"
(SNP or variant id), "cm_position" (centimorgan position), "bp_position"
(base-pair position), "allele_1", "allele_2".
The values are lists or arrays. See example, below.
count_A1: bool, optional
True (default) to count the number of A1 alleles (the PLINK standard).
False to count the number of A2 alleles.
fam_filepath: pathlib.Path or str, optional
Path to the file containing information about each individual (sample).
Defaults to replacing the .bed file’s suffix with .fam.
bim_filepath: pathlib.Path or str, optional
Path to the file containing information about each SNP (variant).
Defaults to replacing the .bed file’s suffix with .bim.
major: str, optional
Use "SNP" (default) to write the file is usual SNP-major mode. This makes reading
the data SNP-by-SNP faster. Use "individual" to write the file in the uncommon
individual-major mode.
force_python_only
If False (default), uses the faster Rust helper functions; otherwise it uses the slower
pure Python code.
num_threads: None or int, optional
The number of threads with which to write data.
Defaults to all available processors.
Can also be set with these environment variables (listed in priority order):
'PST_NUM_THREADS', 'NUM_THREADS', 'MKL_NUM_THREADS'.
Examples
--------
In this example, all properties are given.
.. doctest::
>>> import numpy as np
>>> from bed_reader import to_bed, tmp_path
>>>
>>> output_file = tmp_path() / "small.bed"
>>> val = [[1.0, 0.0, np.nan, 0.0],
... [2.0, 0.0, np.nan, 2.0],
... [0.0, 1.0, 2.0, 0.0]]
>>> properties = {
... "fid": ["fid1", "fid1", "fid2"],
... "iid": ["iid1", "iid2", "iid3"],
... "father": ["iid23", "iid23", "iid22"],
... "mother": ["iid34", "iid34", "iid33"],
... "sex": [1, 2, 0],
... "pheno": ["red", "red", "blue"],
... "chromosome": ["1", "1", "5", "Y"],
... "sid": ["sid1", "sid2", "sid3", "sid4"],
... "cm_position": [100.4, 2000.5, 4000.7, 7000.9],
... "bp_position": [1, 100, 1000, 1004],
... "allele_1": ["A", "T", "A", "T"],
... "allele_2": ["A", "C", "C", "G"],
... }
>>> to_bed(output_file, val, properties=properties)
Here, no properties are given, so default values are assigned.
If we then read the new file and list the chromosome property,
it is an array of '0's, the default chromosome value.
.. doctest::
>>> output_file2 = tmp_path() / "small2.bed"
>>> val = [[1, 0, -127, 0], [2, 0, -127, 2], [0, 1, 2, 0]]
>>> to_bed(output_file2, val)
>>>
>>> from bed_reader import open_bed
>>> with open_bed(output_file2) as bed2:
... print(bed2.chromosome)
['0' '0' '0' '0']
"""
filepath = Path(filepath)
val = _fix_up_val(val)
iid_count = val.shape[0]
sid_count = val.shape[1]
if major == "SNP":
major_count = sid_count
minor_count = iid_count
elif major == "individual":
major_count = iid_count
minor_count = sid_count
else:
raise ValueError(f"major must be 'SNP' or 'individual', not '{major}'")
properties, _ = open_bed._fix_up_properties(
properties, iid_count=iid_count, sid_count=sid_count, use_fill_sequence=True
)
open_bed._write_fam_or_bim(filepath, properties, "fam", fam_filepath)
open_bed._write_fam_or_bim(filepath, properties, "bim", bim_filepath)
if not force_python_only:
if not val.flags["C_CONTIGUOUS"] and not val.flags["F_CONTIGUOUS"]:
raise ValueError("val must be contiguous.")
num_threads = get_num_threads(num_threads)
if major == "individual":
val = val.T
try:
if val.dtype == np.float64:
write_f64(
str(filepath),
is_a1_counted=count_A1,
val=val,
num_threads=num_threads,
)
elif val.dtype == np.float32:
write_f32(
str(filepath),
is_a1_counted=count_A1,
val=val,
num_threads=num_threads,
)
elif val.dtype == np.int8:
write_i8(
str(filepath),
is_a1_counted=count_A1,
val=val,
num_threads=num_threads,
)
else:
raise ValueError(
f"dtype '{val.dtype}' not known, only "
+ "'int8', 'float32', and 'float64' are allowed."
)
if major == "individual":
# change the 3rd byte in the file to 0b00000000
with open(filepath, "r+b") as bed_filepointer:
bed_filepointer.seek(2)
bed_filepointer.write(bytes(bytearray([0b00000000])))
except SystemError as system_error:
try:
os.unlink(filepath)
except Exception:
pass
raise system_error.__cause__
else:
if not count_A1:
zero_code = 0b00
two_code = 0b11
else:
zero_code = 0b11
two_code = 0b00
with open(filepath, "wb") as bed_filepointer:
# see http://zzz.bwh.harvard.edu/plink/binary.shtml
bed_filepointer.write(bytes(bytearray([0b01101100]))) # magic numbers
bed_filepointer.write(bytes(bytearray([0b00011011]))) # magic numbers
if major == "SNP":
bed_filepointer.write(bytes(bytearray([0b00000001]))) # snp major
else:
assert major == "individual"
bed_filepointer.write(
bytes(bytearray([0b00000000]))
) # individual major
for major_index in range(major_count):
if major_index % 1 == 0:
logging.info(
f"Writing {major} # {major_index} to file '{filepath}'"
)
if major == "SNP":
vector = val[:, major_index]
else:
assert major == "individual"
vector = val[major_index, :]
for minor_by_four in range(0, minor_count, 4):
vals_for_this_byte = vector[minor_by_four : minor_by_four + 4]
byte = 0b00000000
for val_index in range(len(vals_for_this_byte)):
val_for_byte = vals_for_this_byte[val_index]
if val_for_byte == 0:
code = zero_code
elif val_for_byte == 1:
code = 0b10 # backwards on purpose
elif val_for_byte == 2:
code = two_code
elif (
val.dtype == np.int8 and val_for_byte == -127
) or np.isnan(val_for_byte):
code = 0b01 # backwards on purpose
else:
raise ValueError(
"Attempt to write illegal value to .bed file. "
+ "Only 0,1,2,missing allowed."
)
byte |= code << (val_index * 2)
bed_filepointer.write(bytes(bytearray([byte])))
logging.info(f"Done writing {filepath}")
def _fix_up_val(input):
if not isinstance(input, np.ndarray):
return _fix_up_val(np.array(input))
if np.issubdtype(input.dtype, np.integer) and input.dtype != np.int8:
return _fix_up_val(np.array(input, dtype=np.int8))
elif np.issubdtype(input.dtype, np.floating) and input.dtype not in (
np.float32,
np.float64,
):
return _fix_up_val(np.array(input, dtype=np.float32))
if len(input.shape) != 2:
raise ValueError("val should be two dimensional")
return input
# if __name__ == "__main__":
# logging.basicConfig(level=logging.INFO)
# import pytest
# pytest.main(["--doctest-modules", __file__])