ptp/components/tasks/text_to_text/wikitext_language_modeling.py
# Copyright (C) tkornuta, IBM Corporation 2019
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
__author__ = "Tomasz Kornuta"
import os
from nltk.tokenize import WhitespaceTokenizer
import ptp.components.mixins.io as io
from ptp.configuration import ConfigurationError
from ptp.components.tasks.task import Task
from ptp.data_types.data_definition import DataDefinition
class WikiTextLanguageModeling(Task):
"""
Language modeling task using WikiText-2 (_dataset2) / WikiText-103 (_dataset103) datasets, featured at the Salesforce _website.
Task downloads the files, loads the file associated with a given subset (train/valid/test), concatenates all sencentes and tokenizes them using NLTK's WhitespaceTokenizer.
Resulting tokens are then passed to samples (source/target) as list of tokens of a given length (set by the user in configuration file).
Associated paper: Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer Sentinel Mixture Models (2016) (_arxiv)
.. _website: https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/
.. _dataset2: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-v1.zip
.. _dataset103: https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-103-v1.zip
.. _arxiv: https://arxiv.org/abs/1609.07843
"""
def __init__(self, name, config):
"""
The init method downloads the required files, loads the file associated with a given subset (train/valid/test),
concatenates all sencentes and tokenizes them using NLTK's WhitespaceTokenizer.
It also stores the intermediate results, so for example, it file with tokenized set is found, it simply loads it.
:param name: Name of the component.
:param class_type: Class type of the component.
:param config: Dictionary of parameters (read from configuration ``.yaml`` file).
"""
# Call constructor of parent classes.
Task.__init__(self, name, WikiTextLanguageModeling, config)
# Set streams key mappings.
self.key_sources = self.stream_keys["sources"]
self.key_targets = self.stream_keys["targets"]
# Get absolute path to data folder.
self.data_folder = os.path.expanduser(self.config['data_folder'])
# Get dataset.
if (self.config['dataset'] is None) or (self.config['dataset'] not in ["wikitext-2", "wikitext-103"]):
raise ConfigurationError("Task supports two 'dataset' options: 'wikitext-2', 'wikitext-103' ")
dataset = self.config['dataset']
# Get (sub)set: train/valid/test.
if (self.config['subset'] is None) or (self.config['subset'] not in ['train', 'valid', 'test']):
raise ConfigurationError("Task supports three 'subset' options: 'train', 'valid', 'test' ")
subset = self.config['subset']
# Check if file with tokenized words exists.
filename_tokenized_words = "wiki."+self.config['subset']+".tokenized_words"
if not io.check_files_existence(self.data_folder, filename_tokenized_words):
# If not, we must generate (and save it) using source files.
# Names of files used by this task.
filenames = ["wiki.train.tokens", "wiki.valid.tokens", "wiki.test.tokens"]
# Initialize dataset if files do not exist.
if not io.check_files_existence(self.data_folder, filenames):
# Set url and source filename depending on dataset.
if dataset == "wikitext-2":
url = "https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-2-v1.zip"
zipfile_name = "wikitext-2-v1.zip"
else:
url = "https://s3.amazonaws.com/research.metamind.io/wikitext/wikitext-103-v1.zip"
zipfile_name = "wikitext-103-v1.zip"
# Download and extract wikitext zip.
io.download_extract_zip_file(self.logger, self.data_folder, url, zipfile_name)
# Move extracted files to the right folder.
io.move_files_between_dirs(self.logger, os.path.join(self.data_folder, dataset) , self.data_folder, filenames)
else:
self.logger.info("Files {} found in folder '{}'".format(filenames, self.data_folder))
# Load the whole sentences.
sentences = io.load_string_list_from_txt_file(self.data_folder, "wiki."+subset+".tokens")
self.logger.info("Loaded {} sentences from the 'wiki.{}.tokens' subset".format(len(sentences), subset))
# Generate text full of tokens.
self.logger.info("Please wait, using NLTK to tokenize the loaded sentences...")
# Create a single text by replacing newlines with <eos> tokens.
text = " <eos> ".join(sentences)
# Tokenize.
tokenizer = WhitespaceTokenizer()
self.tokens = tokenizer.tokenize(text)
# Save fo file.
io.save_string_list_to_txt_file(self.data_folder, filename_tokenized_words, self.tokens)
self.logger.info("Created text consisting of {} tokens and saved it to '{}'".format(len(self.tokens), filename_tokenized_words))
else:
# Ok, file with tokens exists, load it.
self.tokens = io.load_string_list_from_txt_file(self.data_folder, filename_tokenized_words)
self.logger.info("Load text consisting of {} tokens from '{}'".format(len(self.tokens), filename_tokenized_words))
# Get the required sample length.
self.sentence_length = self.config['sentence_length']
# Calculate the size of dataset.
self.dataset_length = len(self.tokens) - self.sentence_length - 1 # as target is "shifted" by 1.
# Display exemplary sample.
self.logger.info("Exemplary sample:\n source: {}\n target: {}".format(self.tokens[0:self.sentence_length], self.tokens[1:self.sentence_length+1]))
def output_data_definitions(self):
"""
Function returns a dictionary with definitions of output data produced the component.
:return: dictionary containing output data definitions (each of type :py:class:`ptp.utils.DataDefinition`).
"""
return {
self.key_indices: DataDefinition([-1, 1], [list, int], "Batch of sample indices [BATCH_SIZE] x [1]"),
self.key_sources: DataDefinition([-1, self.sentence_length, 1], [list, list, str], "Batch of input sentences, each consisting of several words [BATCH_SIZE] x [SENTENCE_LENGTH] x [string]"),
self.key_targets: DataDefinition([-1, self.sentence_length, 1], [list, list, str], "Batch of target sentences, each consisting of several words [BATCH_SIZE] x [SENTENCE_LENGTH] x [string]")
}
def __len__(self):
"""
Returns the "size" of the "task" (total number of samples).
:return: The size of the task.
"""
return self.dataset_length
def __getitem__(self, index):
"""
Getter method to access the dataset and return a sample.
:param index: index of the sample to return.
:type index: int
:return: ``DataStreams({'indices', sources','targets'})``
"""
# Return data_streams.
data_streams = self.create_data_streams(index)
data_streams[self.key_sources] = self.tokens[index:index+self.sentence_length]
data_streams[self.key_targets] = self.tokens[index+1:index+self.sentence_length+1] # target is "shifted" by 1.
#print("task: index = {} source = {} target = {}".format(index, data_streams[self.key_sources], data_streams[self.key_targets]))
return data_streams
def collate_fn(self, batch):
"""
Generates a batch of samples from a list of individuals samples retrieved by :py:func:`__getitem__`.
:param batch: List of :py:class:`ptp.utils.DataStreams` retrieved by :py:func:`__getitem__`
:type batch: list
:return: DataStreams containing the created batch.
"""
# Collate indices.
data_streams = self.create_data_streams([sample[self.key_indices] for sample in batch])
# Collate sources.
data_streams[self.key_sources] = [sample[self.key_sources] for sample in batch]
data_streams[self.key_targets] = [sample[self.key_targets] for sample in batch]
return data_streams