def file_size(filename):
"""Returns size of FILENAME in bytes (or -1 if not found)"""
size = -1
if os.path.exists(filename):
size = os.path.getsize(filename)
tpo.debug_format("file_size({f}) => {s}", 5, f=filename, s=size)
return size
def read_lines(filename=None, make_unicode=False):
"""Returns list of lines from FILENAME without newlines (or other extra whitespace)
@notes: Uses stdin if filename is None. Optionally returned as unicode."""
# TODO: use enumerate(f); refine exception in except;
# TODO: force unicode if UTF8 encountered
lines = []
f = None
try:
# Open the file
if not filename:
tpo.debug_format("Reading from stdin", 4)
f = sys.stdin
else:
f = open(filename)
if not f:
raise IOError
# Read line by line
for line in f:
line = line.strip("\n")
if make_unicode:
line = tpo.ensure_unicode(line)
lines.append(line)
except IOError:
debug_print("Warning: Exception reading file %s: %s" % (filename, str(sys.exc_info())), 2)
finally:
if f:
f.close()
debug_print("read_lines(%s) => %s" % (filename, lines), 6)
return lines
def create_directory(path):
"""Wrapper around os.mkdir over PATH (with tracing)"""
if not os.path.exists(path):
os.mkdir(path)
debug_format("os.mkdir({p})", 6, p=path)
else:
assertion(os.path.isdir(path))
return
def getenv_filename(var, default="", description=None):
"""Returns text filename based on environment variable VAR (or string version of DEFAULT)
with optional DESCRIPTION. This includes a sanity check for file being non-empty."""
debug_format("getenv_filename({v}, {d}, {desc})", 6,
v=var, d=default, desc=description)
filename = tpo.getenv_text(var, default, description)
if filename and not non_empty_file(filename):
tpo.print_stderr("Error: filename %s empty or missing for environment option %s" % (filename, var))
return filename
def tokenize(text):
"""Tokenize TEXT according to regex word tokens (i.e., \W+), which defaults to [A-Za-z0-9_]+"""
# TODO: Allow for tokenization regex to be overwritten
token_regex = r"\W+" if not PRESERVE else r"\S+"
tokens = [t.strip() for t in re.split(token_regex, text) if t.strip()]
if DOWNCASE:
tokens = [t.lower() for t in tokens]
tpo.debug_format("tokenize({txt}) => t", 7, txt=text, t=tokens)
return tokens
def delete_file(filename):
"""Deletes FILENAME"""
debug_print("delete_file(%s)" % tpo.normalize_unicode(filename), 5)
assertion(os.path.exists(filename))
ok = False
try:
ok = os.remove(filename)
debug_format("remove{f} => {r}", 6, f=filename, r=ok)
except OSError:
debug_print("Exception during deletion of {filename}: " + str(sys.exc_info()), 5)
return ok
def get_directory_listing(dirname, make_unicode=False):
"""Returns files in DIRNAME"""
all_file_names = []
try:
all_file_names = os.listdir(dirname)
except OSError:
tpo.debug_format("Exception during get_directory_listing: {exc}", 4,
exc=str(sys.exc_info()))
if make_unicode:
all_file_names = [tpo.ensure_unicode(f) for f in all_file_names]
tpo.debug_format("get_directory_listing({dir}) => {files}", 5,
dir=dirname, files=all_file_names)
return all_file_names
def resolve_path(filename, base_dir=None):
"""Resolves path for FILENAME related to BASE_DIR if not in current directory. Note: this uses the script directory for the calling module if BASE_DIR not specified (i.e., as if os.path.dirname(__file__) passed)."""
path = filename
if not os.path.exists(path):
if not base_dir:
frame = None
try:
frame = inspect.currentframe().f_back
base_dir = os.path.dirname(frame.f_globals['__file__'])
except (AttributeError, KeyError):
base_dir = ""
debug_print("Exception during resolve_path: " + str(sys.exc_info()), 5)
finally:
if frame:
del frame
path = os.path.join(base_dir, path)
debug_format("resolve_path({f}) => {p}", 4, f=filename, p=path)
return path
def extract_matches(pattern, lines, fields=1):
"""Checks for PATTERN matches in LINES of text returning list of tuples with replacement groups"""
# ex: extract_matches(r"^(\S+) \S+", ["John D.", "Jane D.", "Plato"]) => ["John", "Jane"]
assert type(lines) == list
if pattern.find("(") == -1:
pattern = "(" + pattern + ")"
matches = []
for line in lines:
try:
match = re.search(pattern, line)
if match:
result = match.group(1) if (fields == 1) else [match.group(i + 1) for i in range(fields)]
matches.append(result)
except (re.error, IndexError):
debug_print("Warning: Exception in pattern matching: %s" % str(sys.exc_info()), 2)
debug_print("extract_matches(%s, _, [%s]) => %s" % (pattern, fields, matches), 7)
double_indent = INDENT + INDENT
debug_format("{ind}input lines: {{\n{res}\n{ind}}}", 8,
ind=INDENT, res=indent_lines("\n".join(lines), double_indent))
return matches
def __iter__(self):
"""Returns iterator producing one line at a time"""
# Derive the list of filenames to process
# TODO: support recursive directory descent
tpo.debug_print("in MySentences.__iter__()", 6)
file_names = None
if os.path.isdir(self.file_name):
dir_name = self.file_name
file_names = [
os.path.join(dir_name, f) for f in os.listdir(dir_name)
]
else:
file_names = [self.file_name]
# Feed each sentence individually from each file
# TODO: add preprocessing (e.g., tokenize, make lowercase, etc.)
for file_name in file_names:
if os.path.isdir(file_name):
tpo.debug_format("Warning: skipping subdirectory {f}",
tpo.WARNING,
f=file_name)
continue
tpo.debug_format("Processing file {f}", tpo.DETAILED, f=file_name)
for line in open(file_name):
## OLD: tokens = line.split()
tokens = tokenize(line)
tpo.debug_format("MySentences.__iter__: yielding {t}",
6,
t=tokens)
yield tokens
tpo.debug_print("out MySentences.__iter__()", 6)
return
def assertion(condition):
"""Issues warning if CONDITION doesn't hold"""
# EX: assertion(2 + 2 != 5)
# TODO: rename as soft_assertion???; add to tpo_common.py (along with run???)
if not condition:
# Try to get file and line number from stack frame
# note: not available during interactive use
filename = None
line_num = -1
frame = None
try:
frame = inspect.currentframe().f_back
tpo.debug_trace("frame=%s", frame, level=8)
tpo.trace_object(frame, 9, "frame")
filename = frame.f_globals.get("__file__")
if filename and filename.endswith(".pyc"):
filename = filename[:-1]
line_num = frame.f_lineno
finally:
if frame:
del frame
# Get text for line and extract the condition from invocation,
# ignoring comments and function name.
# TODO: define function for extracting line, so this can be put in tpo_common.py
line = "???"
if filename:
line = run("tail --lines=+{l} '{f}' | head -1",
subtrace_level=8, f=filename, l=line_num)
condition = re.sub(r"^\s*\S*assertion\((.*)\)\s*(\#.*)?$",
"\\1", line)
# Print the assertion warning
line_spec = "???"
if filename:
line_spec = "{f}:{l}".format(f=filename, l=line_num)
debug_format("*** Warning: assertion failed: ({c}) at {ls}",
tpo.WARNING, c=condition, ls=line_spec)
return
def get_matching_files(pattern):
"""Get list of files matching pattern via shell globbing"""
files = glob.glob(pattern)
tpo.debug_format("get_matching_files({p}) => {l}", 5,
p=pattern, l=files)
return files
def form_path(*filenames):
"""Wrapper around os.path.join over FILENAMEs (with tracing)"""
path = os.path.join(*filenames)
debug_format("form_path{f} => {p}", 6, f=tuple(filenames), p=path)
return path
def main():
"""Entry point for script"""
tpo.debug_print("main(): sys.argv=%s" % sys.argv, 4)
# Parse command-line arguments
env_options = tpo.formatted_environment_option_descriptions(indent=" ")
usage_description = tpo.format("""
Creates Google word2vec model (via gensim) of word distributions inferrred from
the occurrences in the input text file. Note: input should be a text file
(or directory) when creating from scratch or the basename of model file
if loading existing model.
Notes:
- The input file should have one document per line (multiple sentences allowed).
- The following environment options are available:
{env}
""",
env=env_options)
parser = argparse.ArgumentParser(
description=usage_description,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument("--save",
default=False,
action='store_true',
help="Save model to disk")
parser.add_argument("--load",
default=False,
action='store_true',
help="Load model from disk")
parser.add_argument("--print",
default=False,
action='store_true',
help="Print vectors on standard output")
parser.add_argument(
"filename",
default=None,
help=
"Input data filename (or basename when loading previously saved model); if a directory all files within are processed"
)
parser.add_argument(
"--output-basename",
default=None,
help=
"Basename to use for output (by default input file without .txt extension)"
)
parser.add_argument(
"--show-similarity",
default=False,
action='store_true',
help="Show similar terms for those from input (one per line)")
# TODO: parser.add_argument("--language-model", default=None, help="Language model to use for rating similar terms")
args = vars(parser.parse_args())
tpo.debug_print("args = %s" % args, 5)
filename = args['filename']
save = args['save']
load = args['load']
print_vectors = args['print']
show_similarity = args['show_similarity']
output_basename = args['output_basename']
# TODO: put version of glue_helper's assertion into tpo_common.py already!
gh.assertion(filename)
# Derive the basename if not given (checking one of .txt/.list/.prep extensions if training or .word2vec if loading)
# TODO: rework in terms of stripping whatever file extension is used (e.g., "it.fubar" => "it")
if not output_basename:
input_extensions = [".txt", ".list", ".prep"
] if (not load) else [WORD2VEC_MODEL_EXT]
output_basename = filename
for extension in input_extensions:
output_basename = gh.remove_extension(filename, extension)
if (output_basename != filename):
break
tpo.debug_print("output_basename=%s" % output_basename, 5)
# Enable logging if debugging
if (tpo.debugging_level()):
# TODO: use mapping from symbolic LEVEL user option (e.g., via getenv)
level = logging.INFO if (tpo.debug_level < 4) else logging.DEBUG
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',
level=level)
# Optionally set random seed
if RANDOM_SEED != -1:
tpo.debug_format("Setting random seed to {RANDOM_SEED}")
numpy.random.seed(RANDOM_SEED)
# Process the input file(s), either creating model from scratch or loading existing one
if load:
model = Word2Vec.load(filename)
else:
sentences = MySentences(filename)
if tpo.verbose_debugging():
# TODO: try to develop develop read-only function that makes copy of iterator
sentences = list(sentences)
gh.assertion(len(sentences) > 0)
tpo.debug_format("sentences={s}", 6, s=sentences)
# Notes: 1 is default for word2vec (todo, try None)
seed = 1 if (RANDOM_SEED == -1) else RANDOM_SEED
model = Word2Vec(sentences, workers=NUM_WORKERS, seed=seed)
# Optionally save model to disk
if (save):
model.save(output_basename + WORD2VEC_MODEL_EXT)
# Print the vector representations
# TODO: add option to print word similarity matrix
if print_vectors:
all_words = sorted(model.vocab.keys())
tpo.debug_format("model={m}", 6, m=model)
print("Vocaulary terms: %s" % all_words)
for word in all_words:
tpo.debug_format("model[%s]=%s" % (word, model[word]), 5)
print("%s\t%s" % (word, model[word]))
# Show similarity info for terms from input
# TODO: add better recovery for terms unknown
if show_similarity:
tpo.debug_print("Show similarity for terms from stdin", 4)
print("term(s): similarity info")
for line in sys.stdin:
## OLD: terms = [t.strip() for t in re.split(r"\W+", line.strip().lower())]
terms = tokenize(line)
try:
# TODO: shows language model score for terms replaced by related terms
if not terms:
pass
elif len(terms) > 1 or SKIP_INDIVIDUAL:
print(
"[%s]: %s" %
(", ".join(terms), format_related_terms(model, terms)))
else:
if not SKIP_INDIVIDUAL:
for term in terms:
print("[%s]: %s" %
(term, format_related_terms(model, [term])))
print("")
except KeyError:
tpo.print_stderr("Error: %s" % str(sys.exc_info()))
return
def __init__(self, file_name):
"""Class constructor: FILE_NAME is text file or directory"""
tpo.debug_format("MySentences.__init__({f})", 6, f=file_name)
self.file_name = file_name
return
