def _save_block(self, file_num, tokens):
"""
Save an inverted block.
:param file_num: The file index.
:type file_num: int
:param tokens: List of tokens to save.
:type tokens: list
"""
for raw_token in tokens:
raw_token = raw_token.strip()
for token in tokenize(raw_token):
if len(token) > 0:
line = token + ' ' + str(file_num) + "\n"
self._lines.append(line)
if len(line) > self.block_size:
self._save_line_buffer()
def construct(self):
dictionary = dict()
postings = dict()
last_docs = dict()
files = dict()
token_id = 0
num_files = len(self._files)
# Apply lossless compression by only saving the gaps between document ids (not the document ids themselves)
for doc_id in range(0, num_files):
# Loop through all documents
file = self._files[doc_id]
files[doc_id] = file
with open(file, 'r') as handle:
# Open the file and tokenize
tokens = tokenize(handle.read())
for token in tokens:
# Update the dictionary and postings
if token not in dictionary:
dictionary[token] = token_id
postings[token_id] = deque([doc_id])
last_docs[token_id] = doc_id
token_id += 1
else:
found_token_id = dictionary[token]
gap = doc_id - last_docs[found_token_id]
postings[found_token_id].append(gap)
last_docs[found_token_id] = doc_id
if self.show_progress:
sys.stdout.write("\rIndexing: " + '{0} / {1}'.format(doc_id, num_files))
sys.stdout.flush()
# The index construction is done
if self.show_progress:
sys.stdout.write("\rIndex construction: done\n")
sys.stdout.flush()
# Save the index
index = (postings, dictionary, files)
with open(self.get_index_path(), 'wb') as handle:
pickle.dump(index, handle)
def construct(self):
"""
Construct the index.
"""
# Make a subdirectory to store the blocks
if os.path.exists(self._block_path):
shutil.rmtree(self._block_path)
os.mkdir(self._block_path)
self._lines = []
self._block_num = 0
file_num = 0
for file in self._files:
file_pointer = open(file)
buffered_token = ''
while True:
block = file_pointer.read(self.block_size + 1)
if not block:
break
tokens = tokenize(block)
# Only do something if there are found tokens
if len(tokens) > 0:
# Detect whether the block ends with a token
ends_with_token = (tokens[-1][-1] == block[-1])
# Detect whether the block start with a token
starts_with_token = (tokens[0][0] == block[0])
# Check whether the whole block is a token
block_is_token = (tokens[0] == block)
if block_is_token:
# If the full block is a token, then just enlarge the buffer
buffered_token += tokens[0]
else:
# Otherwise, there are at least two tokens in the block
block_tokens = []
if starts_with_token:
block_tokens.append(buffered_token + tokens[0])
buffered_token = ''
tokens = tokens[1:]
else:
block_tokens.append(buffered_token)
buffered_token = ''
if ends_with_token:
buffered_token = tokens[-1]
tokens = tokens[:-1]
for token in tokens:
block_tokens.append(token)
# Save the block tokens and increase the block number
if len(block_tokens) > 0:
self._save_block(file_num, block_tokens)
if len(buffered_token) > 0:
# If there is some left over token, save it as a block token and increase the block number
self._save_block(file_num, [buffered_token])
# Do not forget to close the file
file_pointer.close()
# Increase the file identifier
file_num += 1
# Display the progres
if self.show_progress:
sys.stdout.write("\rIndex construction: " + '{0} / {1}'.format(file_num, len(self._files)))
sys.stdout.flush()
# Save the line buffer if there is any left
self._save_line_buffer()
# The index construction is done
if self.show_progress:
sys.stdout.write("\rIndex construction: done\n")
sys.stdout.flush()
# Merge the blocks
self._merge_blocks()
# Clean up the temporary block folder
shutil.rmtree(self._block_path)
