app-i18n/libkkc-data: resuming work on it when it becomes feasible
Signed-off-by: Shin'ya Minazuki <shinyoukai@laidback.moe>
This commit is contained in:
Shin'ya Minazuki
121
app-i18n/libkkc-data/files/genfilter.py
Normal file
121
app-i18n/libkkc-data/files/genfilter.py
Normal file
@@ -0,0 +1,121 @@
|
||||
#!/usr/bin/python3.13
|
||||
|
||||
# Copyright (C) 2011-2013 Daiki Ueno <ueno@gnu.org>
|
||||
# Copyright (C) 2011-2013 Red Hat, Inc.
|
||||
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU General Public License for more details.
|
||||
|
||||
# You should have received a copy of the GNU General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
import os
|
||||
import mmap
|
||||
import math
|
||||
import struct
|
||||
|
||||
ERROR_RATE = 0.25
|
||||
|
||||
def murmur_hash3_32(b0, b1, seed):
|
||||
h1 = seed
|
||||
|
||||
c1 = 0xcc9e2d51
|
||||
c2 = 0x1b873593
|
||||
|
||||
# body: b0
|
||||
b0 *= c1
|
||||
b0 &= 0xFFFFFFFF
|
||||
b0 = (b0 << 15) | (b0 >> (32 - 15))
|
||||
b0 &= 0xFFFFFFFF
|
||||
b0 *= c2
|
||||
b0 &= 0xFFFFFFFF
|
||||
|
||||
h1 ^= b0
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 = (h1 << 13) | (h1 >> (32 - 13))
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 = h1 * 5 + 0xe6546b64
|
||||
h1 &= 0xFFFFFFFF
|
||||
|
||||
# body: b1
|
||||
b1 *= c1
|
||||
b1 &= 0xFFFFFFFF
|
||||
b1 = (b1 << 15) | (b1 >> (32 - 15))
|
||||
b1 &= 0xFFFFFFFF
|
||||
b1 *= c2
|
||||
b1 &= 0xFFFFFFFF
|
||||
|
||||
h1 ^= b1
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 = (h1 << 13) | (h1 >> (32 - 13))
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 = h1 * 5 + 0xe6546b64
|
||||
h1 &= 0xFFFFFFFF
|
||||
|
||||
# No tail processing needed.
|
||||
|
||||
# fmix
|
||||
h1 ^= 8
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 ^= h1 >> 16
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 *= 0x85ebca6b
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 ^= h1 >> 13
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 *= 0xc2b2ae35
|
||||
h1 &= 0xFFFFFFFF
|
||||
h1 ^= h1 >> 16
|
||||
h1 &= 0xFFFFFFFF
|
||||
return h1
|
||||
|
||||
class FilterGenerator(object):
|
||||
def __init__(self, infile, outfile, record_size):
|
||||
self.infile = infile
|
||||
self.outfile = outfile
|
||||
self.record_size = record_size
|
||||
|
||||
def generate(self):
|
||||
size = os.fstat(self.infile.fileno()).st_size
|
||||
n = size / self.record_size
|
||||
m = int(math.ceil(-n*math.log10(ERROR_RATE) /
|
||||
math.pow(math.log10(2), 2)))
|
||||
m = (m/8 + 1)*8
|
||||
inmem = mmap.mmap(self.infile.fileno(),
|
||||
size,
|
||||
access=mmap.ACCESS_READ)
|
||||
outmem = bytearray(m/8)
|
||||
for i in range(0, n):
|
||||
offset = i*self.record_size
|
||||
b0, b1 = struct.unpack("=LL", inmem[offset:offset+8])
|
||||
for k in range(0, 4):
|
||||
h = murmur_hash3_32(b0, b1, k)
|
||||
h = int(h * (m / float(0xFFFFFFFF)))
|
||||
outmem[h/8] |= (1 << (h%8))
|
||||
inmem.close()
|
||||
# Convert bytearray to str, for Python 2.6 compatibility.
|
||||
self.outfile.write(str(outmem))
|
||||
|
||||
if __name__ == '__main__':
|
||||
import sys
|
||||
import argparse
|
||||
|
||||
parser = argparse.ArgumentParser(description='filter')
|
||||
parser.add_argument('infile', type=argparse.FileType('r'),
|
||||
help='input file')
|
||||
parser.add_argument('outfile', type=argparse.FileType('w'),
|
||||
help='output file')
|
||||
parser.add_argument('record_size', type=int,
|
||||
help='record size')
|
||||
args = parser.parse_args()
|
||||
generator = FilterGenerator(args.infile,
|
||||
args.outfile,
|
||||
args.record_size)
|
||||
generator.generate()
|
||||
188
app-i18n/libkkc-data/files/sortlm.py
Normal file
188
app-i18n/libkkc-data/files/sortlm.py
Normal file
@@ -0,0 +1,188 @@
|
||||
#!/usr/bin/python
|
||||
|
||||
# Copyright (C) 2011-2013 Daiki Ueno <ueno@gnu.org>
|
||||
# Copyright (C) 2011-2013 Red Hat, Inc.
|
||||
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU General Public License for more details.
|
||||
|
||||
# You should have received a copy of the GNU General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
import struct
|
||||
import marisa
|
||||
import re
|
||||
|
||||
NGRAM = 3
|
||||
NGRAM_LINE_REGEX = '^([-0-9.]+)[ \t]+([^\t]+?)(?:[ \t]+([-0-9.]+))?$'
|
||||
|
||||
class SortedGenerator(object):
|
||||
def __init__(self, infile, output_prefix):
|
||||
self.__infile = infile
|
||||
self.__output_prefix = output_prefix
|
||||
self.__ngram_line_regex = re.compile(NGRAM_LINE_REGEX)
|
||||
|
||||
self.__ngram_entries = [{} for x in range(0, NGRAM)]
|
||||
|
||||
self.__vocab_keyset = marisa.Keyset()
|
||||
self.__input_keyset = marisa.Keyset()
|
||||
|
||||
self.__vocab_trie = marisa.Trie()
|
||||
self.__input_trie = marisa.Trie()
|
||||
|
||||
self.__min_cost = 0.0
|
||||
|
||||
def read(self):
|
||||
print("reading N-grams")
|
||||
self.__read_tries()
|
||||
self.__read_ngrams()
|
||||
print("min cost = %lf" % self.__min_cost)
|
||||
|
||||
def __read_tries(self):
|
||||
while True:
|
||||
line = self.__infile.readline()
|
||||
if line == "":
|
||||
break
|
||||
if line.startswith("\\1-grams"):
|
||||
break
|
||||
|
||||
unigram_count = 0
|
||||
while True:
|
||||
line = self.__infile.readline()
|
||||
if line == "":
|
||||
break
|
||||
line = line.strip()
|
||||
if line == "":
|
||||
break
|
||||
match = self.__ngram_line_regex.match(line)
|
||||
if not match:
|
||||
continue
|
||||
strv = match.groups()
|
||||
self.__vocab_keyset.push_back(strv[1])
|
||||
if not strv[1] in ("<s>", "</s>", "<UNK>"):
|
||||
if "/" not in strv[1]:
|
||||
continue
|
||||
(input, output) = strv[1].split("/")
|
||||
self.__input_keyset.push_back(input)
|
||||
|
||||
self.__vocab_trie.build(self.__vocab_keyset)
|
||||
self.__input_trie.build(self.__input_keyset)
|
||||
|
||||
def __read_ngrams(self):
|
||||
self.__infile.seek(0)
|
||||
for n in range(1, NGRAM + 1):
|
||||
while True:
|
||||
line = self.__infile.readline()
|
||||
if line == "":
|
||||
break
|
||||
if line.startswith("\\%s-grams:" % n):
|
||||
break
|
||||
|
||||
while True:
|
||||
line = self.__infile.readline()
|
||||
if line == "":
|
||||
break
|
||||
line = line.strip()
|
||||
if line == "":
|
||||
break
|
||||
match = self.__ngram_line_regex.match(line)
|
||||
if not match:
|
||||
continue
|
||||
strv = match.groups()
|
||||
ngram = strv[1].split(" ")
|
||||
ids = []
|
||||
for word in ngram:
|
||||
agent = marisa.Agent()
|
||||
agent.set_query(word)
|
||||
if not self.__vocab_trie.lookup(agent):
|
||||
continue
|
||||
ids.append(agent.key_id())
|
||||
cost = float(strv[0])
|
||||
if cost != -99 and cost < self.__min_cost:
|
||||
self.__min_cost = cost
|
||||
backoff = 0.0
|
||||
if strv[2]:
|
||||
backoff = float(strv[2])
|
||||
self.__ngram_entries[n - 1][tuple(ids)] = (cost, backoff)
|
||||
|
||||
def write(self):
|
||||
self.__min_cost = -8.0
|
||||
self.__write_tries()
|
||||
self.__write_ngrams()
|
||||
|
||||
def __write_tries(self):
|
||||
self.__vocab_trie.save(self.__output_prefix + ".1gram.index")
|
||||
self.__input_trie.save(self.__output_prefix + ".input")
|
||||
|
||||
def __write_ngrams(self):
|
||||
def quantize(cost, min_cost):
|
||||
return max(0, min(65535, int(cost * 65535 / min_cost)))
|
||||
|
||||
def cmp_header(a, b):
|
||||
return cmp(a[0], b[0])
|
||||
|
||||
print("writing 1-gram file")
|
||||
unigram_offsets = {}
|
||||
unigram_file = open("%s.1gram" % self.__output_prefix, "wb")
|
||||
offset = 0
|
||||
for ids, value in sorted(self.__ngram_entries[0].items()):
|
||||
unigram_offsets[ids[0]] = offset
|
||||
s = struct.pack("=HHH",
|
||||
quantize(value[0], self.__min_cost),
|
||||
quantize(value[1], self.__min_cost),
|
||||
0 # reserved
|
||||
)
|
||||
unigram_file.write(s)
|
||||
offset += 1
|
||||
unigram_file.close()
|
||||
|
||||
print("writing 2-gram file")
|
||||
bigram_offsets = {}
|
||||
bigram_file = open("%s.2gram" % self.__output_prefix, "wb")
|
||||
keys = list(self.__ngram_entries[1].keys())
|
||||
items = [(struct.pack("=LL", ids[1], unigram_offsets[ids[0]]), ids) for ids in keys]
|
||||
offset = 0
|
||||
for header, ids in sorted(items, cmp=cmp_header):
|
||||
value = self.__ngram_entries[1][ids]
|
||||
bigram_offsets[ids] = offset
|
||||
s = struct.pack("=HH",
|
||||
quantize(value[0], self.__min_cost),
|
||||
quantize(value[1], self.__min_cost))
|
||||
bigram_file.write(header + s)
|
||||
offset += 1
|
||||
bigram_file.close()
|
||||
|
||||
if len(self.__ngram_entries[2]) > 0:
|
||||
print("writing 3-gram file")
|
||||
trigram_file = open("%s.3gram" % self.__output_prefix, "wb")
|
||||
keys = list(self.__ngram_entries[2].keys())
|
||||
items = [(struct.pack("=LL", ids[2], bigram_offsets[(ids[0], ids[1])]), ids) for ids in keys]
|
||||
for header, ids in sorted(items, cmp=cmp_header):
|
||||
value = self.__ngram_entries[2][ids]
|
||||
s = struct.pack("=H",
|
||||
quantize(value[0], self.__min_cost))
|
||||
trigram_file.write(header + s)
|
||||
trigram_file.close()
|
||||
|
||||
if __name__ == '__main__':
|
||||
import sys
|
||||
import argparse
|
||||
|
||||
parser = argparse.ArgumentParser(description='sortlm')
|
||||
parser.add_argument('infile', nargs='?', type=argparse.FileType('r'),
|
||||
default=sys.stdin,
|
||||
help='language model file')
|
||||
parser.add_argument('output_prefix', metavar='OUTPUT_PREFIX', type=str,
|
||||
help='output file prefix')
|
||||
args = parser.parse_args()
|
||||
|
||||
generator = SortedGenerator(args.infile, args.output_prefix)
|
||||
generator.read();
|
||||
generator.write();
|
||||
@@ -1,8 +1,8 @@
|
||||
# Copyright 2026 Shin'ya Minazuki
|
||||
EAPI=8
|
||||
|
||||
PYTHON_COMPAT=( python3_{11..12} )
|
||||
|
||||
PYTHON_COMPAT=( python2_7 )
|
||||
inherit autotools python-single-r1
|
||||
|
||||
LIBKKC_PV="0.3.5"
|
||||
@@ -11,17 +11,21 @@ DESCRIPTION="Language model data for app-i18n/libkkc"
|
||||
HOMEPAGE="https://github.com/ueno/libkkc"
|
||||
SRC_URI="https://github.com/ueno/libkkc/releases/download/v${LIBKKC_PV}/${P}.tar.xz"
|
||||
|
||||
LICENSE="GPL-2"
|
||||
LICENSE="GPL-3+"
|
||||
SLOT="0"
|
||||
KEYWORDS="~amd64"
|
||||
|
||||
REQUIRED_USE="${PYTHON_REQUIRED_USE}"
|
||||
|
||||
DEPEND="dev-libs/marisa"
|
||||
DEPEND="
|
||||
dev-libs/marisa[python]
|
||||
"
|
||||
RDEPEND="${DEPEND}"
|
||||
|
||||
src_prepare() {
|
||||
default
|
||||
cp -f "${FILESDIR}/genfilter.py" "${S}/tools/genfilter.py"
|
||||
cp -f "${FILESDIR}/sortlm.py" "${S}/tools/sortlm.py"
|
||||
python_fix_shebang tools/genfilter.py
|
||||
python_fix_shebang tools/sortlm.py
|
||||
eautoreconf
|
||||
@@ -30,3 +34,7 @@ src_prepare() {
|
||||
src_configure() {
|
||||
econf
|
||||
}
|
||||
|
||||
src_install() {
|
||||
emake DESTDIR="${D}" install
|
||||
}
|
||||
Reference in New Issue
Block a user