Explore
For Enterprise
Join for Free
Log In

艺术与人文
商务
计算机科学
数据科学
信息技术
健康
数学和逻辑
个人发展
物理科学与工程
社会科学
语言学习
學位
證書

瀏覽 Coursera 的全部課程

瀏覽
搜索
企業版
登錄
免費加入

Notebook for lesson 2

Loading...

Natural Language Processing in TensorFlow

deeplearning.ai

4.6（555 個評分） | 9.9K 名學生已註冊

課程 3（共 4 門，TensorFlow in Practice 專項課程）

If you are a software developer who wants to build scalable AI-powered algorithms, you need to understand how to use the tools to build them. This Specialization will teach you best practices for using TensorFlow, a popular open-source framework for machine learning. In Course 3 of the deeplearning.ai TensorFlow Specialization, you will build natural language processing systems using TensorFlow. You will learn to process text, including tokenizing and representing sentences as vectors, so that they can be input to a neural network. You’ll also learn to apply RNNs, GRUs, and LSTMs in TensorFlow. Finally, you’ll get to train an LSTM on existing text to create original poetry! The Machine Learning course and Deep Learning Specialization from Andrew Ng teach the most important and foundational principles of Machine Learning and Deep Learning. This new deeplearning.ai TensorFlow Specialization teaches you how to use TensorFlow to implement those principles so that you can start building and applying scalable models to real-world problems. To develop a deeper understanding of how neural networks work, we recommend that you take the Deep Learning Specialization.

查看授課大綱

您將學習的技能

Natural Language Processing, Tokenization, Machine Learning, Tensorflow, RNNs

審閱

4.6（555 個評分）

5 stars
395 ratings
4 stars
117 ratings
3 stars
29 ratings
2 stars
9 ratings
1 star
5 ratings

GI

Jun 22, 2019

Amazing course by Laurence Moroney. But only after finishing Sequence Models by Andrew NG, I was able to understand the concepts taught here.

SM

Aug 15, 2019

Course explains some of key concept very well, like sequence model to predict next word using existing corpus.

從本節課中

Sentiment in text

The first step in understanding sentiment in text, and in particular when training a neural network to do so is the tokenization of that text. This is the process of converting the text into numeric values, with a number representing a word or a character. This week you'll learn about the Tokenizer and pad_sequences APIs in TensorFlow and how they can be used to prepare and encode text and sentences to get them ready for training neural networks!

Introduction1:16

Word based encodings2:14

Notebook for lesson 12:32

Text to sequence3:05

Looking more at the Tokenizer1:53

Notebook for lesson 24:18

Sarcasm, really?2:52

Working with the Tokenizer1:42

Notebook for lesson 33:14

Week 1 Wrap up0:21

教學方

Laurence Moroney
AI Advocate

腳本

So let's take a look at this in

a slightly more sophisticated example.

So I'm going to take the tokenizer as we had before,

but I'm also going to introduce this Pad Sequences tool.

The idea behind the pad sequences tool is that it allows

you to use sentences of different lengths,

and use padding or truncation to

make all of the sentences the same length.

So in this case, I have the same sentences as before;

'I love my dog,' 'I love my cat,' You love my dog,

but I've added this new sentence;

'Do you think my dog is amazing?'

Which is a different length from these other sentences.

These all had four words, this one has more.

So my tokenizer, I'm going to create as before.

But I'm also going to use this parameter

called an OOV token.

The idea here is that I'm going to create a new token,

a special token that I'm going to

use for words that aren't recognized,

that aren't in the word index itself.

So I am going to just create this.

I'm going to create something unique here

that I wouldn't expect to see in the corpus.

Something like bracket OOV,

and I'm going to specify my OOV token is that.

So then I'm going to call tokenizer

fit and text sentences,

and I'm going to take a look at the word index for that.

Let's actually run this.

We'll see now, that in my word index,

OOV is now value one,

my is value two,

love is three, etc.

We have a total of 11 words,

11 unique words in this corpus.

It's actually 10 words plus the OOV token.

So on the tokenizer,

I can then convert the words in those sentences to

sequences of tokens by calling

the text to sequences method.

That's going to produce sequences.

That's what I'm printing out here.

So my sequences are five, three,

two, four, for the first sentence, which is,

'I love my dog,' five,

three, two, four, etc.

So these are the sequence is 5324,

5327, 6324, 8692, 41011.

Now, we can see these are all different length,

but we want to make them the same length.

So that's where Pad Sequences comes into it.

So I'm going to say here my pad is set is

pad sequences with the sequences.

Let's make it a maximum length of five words.

So this maximum length of

five words means that these four-letter,

or these four-words sentences end up

being pre-padded with a zero.

This six-word sentence ends up having

the first word cutoff

because we did say maximum length equals five.

I said maximum length equals eight,

for example, and then ran this.

We can see now that they're all pre-padded with zeros,

including this long sentence,

it's being prepared with a single zero.

There are methods on pad sequences that we saw,

and the lessons that will allow us to do

a post if we want to do so,

and then the zeros would appear afterwards.

So now, if I want to take a look at

words that the tokenizer wasn't fit to.

So for example, my test data is I really

love my dog and my dog loves my manatee,

if I now tokenized them and create sequences out of that,

we'll see 51324 for the first sentence.

Five is I, one is out of vocabulary,

because really wasn't actually there,

and three to four,

'I still love my dog.'

So this is how the outer vocabulary token comes into it.

When it sees a word that wasn't in the word index,

it will replace it, it will just use

the out of vocabulary token one for that.

Similarly, for 'my dog loves my manatee,' I get 24121,

the word 'loves' is not in it,

even though the word love is,

and of course, manatee isn't in it either.

So I end up with just 242,

other words that really have meaning in this,

and that's 'my dog,' 'my,' which is,

'my dog my loves manatee,' out of vocabulary tokens.

Of course here, you can see I'm also padding them.

So my 51324 gets padded,

and my 24121 also gets padded.

Because I'd said a max length of 10,

if I set that for example to two,

we'll see they end up getting truncated.

I am getting the last two words here,

and getting the last two words here.

So that's a basic introduction to how tokenizer works,

and how padding actually works,

to give you padding,

to be able to get your sentences all the same length.

I hope this was useful for you.

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera 致力於普及全世界最好的教育，它與全球一流大學和機構合作提供在線課程。

© 2019 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取

Coursera
關於
管理團隊
工作機會
目錄
證書
MasterTrack™ 證書
學位
企業版
政府版

社區
Learners
Partners
Developers
Beta Testers
Translators

連接
博客
Facebook
領英
Twitter
Instagram
YouTube
技術博客

更多
條款
隱私
幫助
內容訪問
媒體
聯繫我們
目錄
附屬公司