Yevgnen Yevgnen

官方文档
其中PREFIX决定了所有缺省的路径

TMP_DIR      = <prefix>/tmp
STAMP_DIR    = <prefix>/src/<name>-stamp
DOWNLOAD_DIR = <prefix>/src
SOURCE_DIR   = <prefix>/src/<name>
BINARY_DIR   = <prefix>/src/<name>-build
INSTALL_DIR  = <prefix>

Addressing the Rare Word Problem in Neural Machine Translation

Introduction

NMT(Neural Machine Translation) systems perform poorly with respect to OOV(out-of-vocabulary) words or rare words.
The paper presents a word-alignment based technique for translating such rare words.
Link to the paper

	'''Test of stateful LSTM.

	This trains a LSTM to convert a frequency-modulated signal to a sine wave.
	The period of the signal is greater than the temporal dimension of the LSTM,
	so in theory the stateful version should have an advantage.
	'''
	from __future__ import print_function

	import os
	os.environ['KERAS_BACKEND'] = 'tensorflow'


	get_latest_release() {
	curl --silent "https://api.github.com/repos/$1/releases/latest" \| # Get latest release from GitHub api
	grep '"tag_name":' \| # Get tag line
	sed -E 's/."([^"]+)"./\1/' # Pluck JSON value
	}

	# Usage
	# $ get_latest_release "creationix/nvm"
	# v0.31.4

	import time
	from http.server import BaseHTTPRequestHandler, HTTPServer

	HOST_NAME = 'localhost'
	PORT_NUMBER = 9000


	class MyHandler(BaseHTTPRequestHandler):
	def do_HEAD(self):
	self.send_response(200)

	import logging
	import sys
	from urllib.parse import urljoin

	import asyncio
	import aiohttp
	from aiohttp import web


	TARGET_SERVER_BASE_URL = 'http://127.0.0.1:8888'

	"""
	When classifying upon a sequence usually we stack some LSTM returning sequences,
	then one LSTM returning a point, then Dense with softmax activation.

	Is it possible instead to give the last non-sequential LSTM a softmax activation?

	The answer is yes.

	In this example we have 3 sequential layers and one layer producing the final result.

	import multiprocessing
	import pandas as pd
	import numpy as np

	def _apply_df(args):
	df, func, num, kwargs = args
	return num, df.apply(func, **kwargs)

	def apply_by_multiprocessing(df,func,**kwargs):
	workers=kwargs.pop('workers')

	import torch.utils.data
	from torchvision import datasets, transforms

	class PartialDataset(torch.utils.data.Dataset):
	def __init__(self, parent_ds, offset, length):
	self.parent_ds = parent_ds
	self.offset = offset
	self.length = length
	assert len(parent_ds)>=offset+length, Exception("Parent Dataset not long enough")
	super(PartialDataset, self).__init__()

Yevgnen Yevgnen

Addressing the Rare Word Problem in Neural Machine Translation

Introduction

Technique