Hello World - Machine Learning Recipes #1YouTube↗  IA↗(long shot)
Science & Technology  2.6M views 56.7K likes 579 dislikes   30 Mar 2016
Google for Developers 2.7M subscribers United States YouTube↗


Please select the problem
You must specify your email and describe the problem in detail
Describe your problem in detail if it's not obvious

Click to Play Video Click to Play Video
[MUSIC PLAYING]
｛機械学習｝
｛機械学習｝ レシピ
Six lines of code is all it takes
６行のコード それだけで
to write your first Machine Learning program.
機械学習プログラムが書けます
My name's Josh Gordon, and today I'll
ジョシュ・ゴードンです
walk you through writing Hello World for Machine learning.
今日は機械学習で 初歩のプログラムを書く説明をします
In the first few episodes of the series,
シリーズの初めの数回で
we'll teach you how to get started with Machine
一から機械学習を始める 方法を教えましょう
Learning from scratch. To do that, we'll work with two open source libraries,
その為に２つのオープンソース ライブラリで作業します
scikit-learn and TensorFlow.
scikit-learn と TensorFlow です
We'll see scikit in action in a minute.
すぐに scikit の動作を見ますが
But first, let's talk quickly about what Machine Learning is
まず機械学習とは何か なぜ必要かについて
and why it's important.
手早く話してみましょう
You can think of Machine Learning as a subfield
機械学習を人工知能の 下位領域と
of artificial intelligence.
考えればよいでしょう
Early AI programs typically excelled at just one thing.
初期の AI プログラムは通常 ただ１つに優れていました
For example, Deep Blue could play chess
例えば Deep Blue は選手権レベルで チェスができましたが
at a championship level, but that's all it could do.
できるのはそれだけでした
Today we want to write one program that
今日書き直す必要がなくて
can solve many problems without needing to be rewritten.
多くの問題が解ける１つの プログラムを書きたいのです
AlphaGo is a great example of that.
アルファ碁はその好例です
As we speak, it's competing in the World Go Championship.
今ちょうど世界碁選手権で 戦っています
But similar software can also learn to play Atari games.
しかし同様のソフトも アタリゲームを学べます
Machine Learning is what makes that possible.
機械学習はそれを 可能にしているものです
It's the study of algorithms that learn from examples and experience
それは例や経験から学ぶ アルゴリズムの研究です
instead of relying on hard-coded rules.
ハードコード化した規則に頼りません
So that's the state-of-the-art.
ですからこれは最新鋭です
But here's a much simpler example we'll start coding up today.
しかし 今日コーディングを始める ずっと簡単な例を示しましょう
I'll give you a problem that sounds easy but is
簡単そうだが
impossible to solve without Machine Learning.
機械学習がなければ 解けない問題を出しましょう
Can you write code to tell the difference
りんごとオレンジを区別する
between an apple and an orange?
コードが書けますか
Imagine I asked you to write a program that takes an image
入力として画像ファイルをとる プログラムを書いて
file as input, does some analysis,
分析をして果物の種類を 出力しなさいと言えば
and outputs the types of fruit.
どうこれを解けますか
How can you solve this? You'd have to start by writing lots of manual rules.
皆さんは手作業の規則を 沢山書くことから始めるでしょう
For example, you could write code
たとえばコードを書いて
to count how many orange pixels there are and compare that
オレンジピクセルはいくつあるか数え
to the number of green ones.
それを緑ピクセルの数と 比べたりするでしょう
The ratio should give you a hint about the type of fruit.
その比で果物の種類のヒントが 出てくる筈ですね
That works fine for simple images like these.
これらのような単純な画像では それでいいですが
But as you dive deeper into the problem,
問題をもっと深く掘り下げると
you'll find the real world is messy, and the rules you
現実世界は複雑で
write start to break.
皆さんの書く規則は破れ始めます
How would you write code to handle black-and-white photos
白黒写真を処理するコードは どう書くでしょう
or images with no apples or oranges in them at all?
または それらにりんごや オレンジが全くない画像は？
In fact, for just about any rule you write,
実は 皆さんが書く大体どんな規則でも
I can find an image where it won't work.
うまくいかない画像を 私は見つけられます
You'd need to write tons of rules,
膨大な規則を書く必要があり
and that's just to tell the difference between apples
しかもそれはただりんごとオレンジを 区別する為だけです
and oranges. If I gave you a new problem, you need to start all over again.
新しい問題を出せば また最初から繰り返さねばなりません
Clearly, we need something better.
明らかに もっと良いものが必要です
To solve this, we need an algorithm
これを解く為に
that can figure out the rules for us,
規則を把握してくれる アルゴリズムが必要です
so we don't have to write them by hand.
すると手作業で規則を 書かなくても済みますね
And for that, we're going to train a classifier.
その為に分類器を学習させます
For now you can think of a classifier as a function.
今は 分類器を関数だと 考えて結構です
It takes some data as input and assigns a label to it
入力としてデータを受け取り
as output.
出力としてそれにラベルを付けます
For example, I could have a picture
例えば絵があって
and want to classify it as an apple or an orange.
それをりんごかオレンジに 分類したいとか
Or I have an email, and I want to classify it
E メールがあってそれを スパムかスパムでないか
as spam or not spam.
分類したいとか
The technique to write the classifier
自動的に分類器を書く技術は
automatically is called supervised learning.
教師付き学習と呼ばれています
It begins with examples of the problem you want to solve.
それは解きたい問題の例で始まります
To code this up, we'll work with scikit-learn.
これをコード化する為に scikit-learn で作業します
Here, I'll download and install the library.
ここで そのライブラリをダウンロードし インストールします
There are a couple different ways to do that.
そうする為には異なる方法が いくつかありますが
But for me, the easiest has been to use Anaconda.
私には Anaconda を使うのが 最も簡単です
This makes it easy to get all the dependencies set up
これだと依存性の全ての 設定を簡単にしてくれ
and works well cross-platform.
クロスプラットフォームで うまく機能します
With the magic of video, I'll fast forward
ビデオの良さを利用して
through downloading and installing it.
ダウンロードとインストールを 早送りします
Once it's installed, you can test
インストールができたら
that everything is working properly
きちんと動くかテストできます
by starting a Python script and importing SK learn.
Python スクリプトを開始して SK learn をインポートします
Assuming that worked, that's line one of our program down,
うまく動いたとして これがプログラムの１行目です
five to go.
あと５行です
To use supervised learning, we'll
教師付き学習を使うために
follow a recipe with a few standard steps.
２，３の標準段階の レシピに従います
Step one is to collect training data.
第１段階は学習データを集めることです
These are examples of the problem we want to solve.
これらは解きたい問題の例です
For our problem, we're going to write a function
私達の問題では
to classify a piece of fruit.
１個の果物の種類を分類する 関数を書きます
For starters, it will take a description of the fruit
まず始めに それは入力として 果物の説明を受け取り
as input and predict whether it's
それがりんごかオレンジか
an apple or an orange as output, based on features like its weight and texture.
重さや質感のような特徴量に基づいて 出力として予測します
To collect our training data, imagine
学習データを集める為に
we head out to an orchard.
果樹園にでかけたとして
We'll look at different apples and oranges
種々のりんごやオレンジを見て
and write down measurements that describe them in a table.
それらを表す測定値を 表に書き留めます
In Machine Learning these measurements
機械学習ではこれらの測定値は
are called features.
特徴量と呼ばれます
To keep things simple, here we've used just two--
分かりやすくするために ここでは２つだけ使いました
how much each fruit weighs in grams and its texture, which can be bumpy or smooth.
各果物は何グラムで 質感はごつごつか滑らかか
A good feature makes it easy to discriminate
特徴量がうまいと
between different types of fruit.
異なる種類の果物を 判別しやすくなります
Each row in our training data is an example.
学習データの各行は例で
It describes one piece of fruit.
１個の果物を表しています
The last column is called the label.
最後の列はラベルと呼ばれます
It identifies what type of fruit is in each row,
各行にどの種の果物があるか 表します
and there are just two possibilities--
可能性は２つだけで
apples and oranges.
りんごかオレンジです
The whole table is our training data.
表全体が学習データです
Think of these as all the examples we want the classifier to learn from.
これらが分類器に学んでほしい 例の全部だと思ってください
The more training data you have, the better a classifier
学習データが多ければ多いほど
you can create.
良い分類器を作れます
Now let's write down our training data in code.
では 私達の学習データを コードで書きましょう
We'll use two variables-- features and labels.
２つの変数 特徴量とラベルを使います
Features contains the first two columns,
特徴量には最初の２列が入ります
and labels contains the last.
ラベルには最後の列が入ります
You can think of features as the input
特徴量を分類器への入力として考え
to the classifier and labels as the output we want.
ラベルを望んでいる出力と 思っていいでしょう
I'm going to change the variable types of all features
特徴量全部の変数の型を
to ints instead of strings, so I'll use 0 for bumpy and 1
文字列ではなく数値型に変えて
for smooth.
ごつごつには 0 滑らかには 1 を使います
I'll do the same for our labels, so I'll use 0 for apple
ラベルも同様にして りんごには 0
and 1 for orange.
オレンジには 1 を使います
These are lines two and three in our program.
これらが私達のプログラムの ２行目と３行目です
Step two in our recipes to use these examples to train
レシピの第２段階は これらの例を使って
a classifier.
分類器を学習させることです
The type of classifier we'll start with
私達が始める分類器の種類は
is called a decision tree.
決定木と呼ばれます
We'll dive into the details of how these work in a future episode.
以降の回でこれらの働き方を 詳細にやりますが
But for now, it's OK to think of a classifier as a box of rules.
今は分類器を規則の箱と 思って構いません
That's because there are many different types of classifier,
それは異なった種類の分類器が 沢山あるからです
but the input and output type is always the same.
しかし 入出力型は常に同じです
I'm going to import the tree.
木をインポートして
Then on line four of our script, we'll create the classifier.
スクリプトの４行目に その分類器を作ります
At this point, it's just an empty box of rules.
この時点では ただの規則の空箱です
It doesn't know anything about apples and oranges yet.
まだりんごやオレンジのことは 何も知りません
To train it, we'll need a learning algorithm.
それを学習させる為に 学習アルゴリズムが必要です
If a classifier is a box of rules,
分類器が規則の箱なら
then you can think of the learning algorithm
学習アルゴリズムを
as the procedure that creates them.
規則を作る手順だと 考えればいいでしょう
It does that by finding patterns in your training data.
学習データ中のパターンを 見つけることでそれをやります
For example, it might notice oranges tend to weigh more,
例えばオレンジの方が 重い傾向があると気づき
so it'll create a rule saying that the heavier fruit is,
重い方の果物という 規則を作ります
the more likely it is to be an orange.
オレンジとする確率が高くなります
In scikit, the training algorithm
scikit では学習アルゴリズムは
is included in the classifier object, and it's called Fit.
分類器オブジェクトに含まれていて それは Fit と呼ばれます
You can think of Fit as being a synonym for "find patterns
Fit を「データ中のパターンを見つける」と 同意語だと思えばいいでしょう
in data."
これが裏でどう起こるのか
We'll get into the details of how this happens under the hood in a future episode.
以降の回で詳細に勉強します
At this point, we have a trained classifier.
今の時点では 学習させた分類器があるので
So let's take it for a spin and use it to classify a new fruit.
早速試して 新しい果物を 分類させてみましょう
The input to the classifier is the features for a new example.
分類器への入力は 新しい例の特徴量です
Let's say the fruit we want to classify
例えば 分類したい果物は
is 150 grams and bumpy.
150 グラムでゴツゴツしているとします
The output will be 0 if it's an apple or 1 if it's an orange.
出力はそれが りんごなら 0 オレンジなら 1 です
Before we hit Enter and see what the classifier predicts,
Enter キーを押して分類器が 予測するのを見る前に
let's think for a sec.
ちょっと考えてみましょう
If you had to guess, what would you say the output should be?
推量するとしたら 出力はどうなると思いますか
To figure that out, compare this fruit to our training data.
それを把握する為にこの果物を 学習データと比べてください
It looks like it's similar to an orange
オレンジに似ているようですね
because it's heavy and bumpy.
重くてゴツゴツしていますからね
That's what I'd guess anyway, and if we hit Enter,
とにかく私はそう推量しますが Enter キーを押すと
it's what our classifier predicts as well.
分類器も同様に予測しています
If everything worked for you, then
皆さんの方で全てが機能したら
that's it for your first Machine Learning program.
それで皆さんの機械学習プログラムが できたということです
You can create a new classifier for a new problem
新しい問題の新しい分類器が作れます
just by changing the training data.
学習データを変えるだけでいいのです
That makes this approach far more reusable
このやり方の方が
than writing new rules for each problem.
問題毎に新しい規則を書くより はるかに再利用できます
Now, you might be wondering why we described our fruit
さて 皆さんはこう思うかもしれませんね
using a table of features instead of using pictures
学習データとしてなぜ特徴量の表を 使って果物を記述するのかと
of the fruit as training data.
果物の絵を使うのではなくて、ですね
Well, you can use pictures, and we'll
絵も使えます
get to that in a future episode.
以降の回でそれを扱います
But, as you'll see later on, the way we did it here
ですが 後で分かりますが ここでやった方法の方が
is more general.
もっと一般的です
The neat thing is that programming with Machine
冴えてることは 機械学習のプログラミングは
Learning isn't hard.
難しくないということです
But to get it right, you need to understand
しかし 正しくやるには
a few important concepts.
重要な概念を少し 理解する必要があります
I'll start walking you through those in the next few episodes.
それらを次回以降で 説明していきます
Thanks very much for watching, and I'll see you then. [MUSIC PLAYING]
ご視聴どうもありがとう またお会いしましょう
Similar videos:

What Makes a Good Feature? - Machine Learning Recipes #3
Google for Developers
507.3K views 6K likes 64 dislikes 27 Apr 201649% match

Train an Image Classifier with TensorFlow for Poets - Machine Learning Recipes #6
Google for Developers
458.7K views 5.7K likes 92 dislikes 30 Jun 201647% match

Visualizing a Decision Tree - Machine Learning Recipes #2
Google for Developers
942.4K views 11.9K likes 171 dislikes 13 Apr 201647% match

Writing Our First Classifier - Machine Learning Recipes #5
Google for Developers
346.5K views 4.3K likes 39 dislikes 08 Jun 201646% match

Let’s Write a Pipeline - Machine Learning Recipes #4
Google for Developers
472.3K views 6K likes 99 dislikes 11 May 201642% match

Classifying Handwritten Digits with TF.Learn - Machine Learning Recipes #7
Google for Developers
304K views 3K likes 47 dislikes 16 Aug 201641% match

Let’s Write a Decision Tree Classifier from Scratch - Machine Learning Recipes #8
Google for Developers
408.9K views 7.5K likes 96 dislikes 13 Sep 201740% match

Wide & Deep Learning with TensorFlow - Machine Learning
Google for Developers
58.5K views 572 likes 16 dislikes 29 Jun 201636% match

Intro to Feature Engineering with TensorFlow - Machine Learning Recipes #9
Google for Developers
108.3K views 1.7K likes 25 dislikes 30 Oct 201734% match

Magenta's AI Jam: Making Music with TensorFlow Models
Google for Developers
25.7K views 524 likes 28 dislikes 12 Jul 201734% match

Machine Learning over Coffee with Josh Gordon - Coffee with a Googler
Google for Developers
85.1K views 1.9K likes 20 dislikes 30 Mar 201631% match

TensorFlow Estimators: Managing Simplicity vs. Flexibility in High-Level ML Frameworks - KDD 2017
Google for Developers
17.8K views 264 likes 11 dislikes 30 Jun 201731% match

Machine Learning Foundations: Ep #1 - What is ML?
Google for Developers
99.6K views 3.4K likes 31 dislikes 06 May 202028% match

TensorFlow: Machine Learning for Everyone
Google for Developers
537.8K views 2.8K likes 59 dislikes 15 Feb 201727% match

Machine learning is not the future - Google I/O 2016
Google for Developers
175.5K views 928 likes 525 dislikes 06 Jun 201626% match

TensorFlow Debugger Screencast
Google for Developers
16.7K views 156 likes 5 dislikes 17 Feb 201725% match

Breakthroughs in Machine Learning - Google I/O 2016
Google for Developers
61.6K views 536 likes 12 dislikes 13 Jun 201625% match

Machine Learning Foundations: Ep #2 - First steps in computer vision
Google for Developers
27.8K views 747 likes 5 dislikes 09 May 202024% match

Machine Learning Foundations: Ep #3 - Convolutions and pooling
Google for Developers
16.5K views 463 likes 5 dislikes 13 May 202024% match

Machine Learning: Google's Vision - Google I/O 2016
Google for Developers
151.2K views 1K likes 36 dislikes 20 May 201624% match

AdWords API DevBytes Episode 2: Getting to “Hello, World!”
Google for Developers
41.2K views 111 likes 38 dislikes 29 Sep 201524% match

Machine learning & art - Google I/O 2016
Google for Developers
18.5K views 247 likes 2 dislikes 20 May 201622% match

Magenta: Music and Art Generation (TensorFlow Dev Summit 2017)
Google for Developers
22K views 333 likes 10 dislikes 15 Feb 201722% match

A.I. Experiments: The Infinite Drum Machine
Google for Developers
85.6K views 1.1K likes 13 dislikes 15 Nov 201621% match

Highlights from the 2017 TensorFlow Dev Summit
Google for Developers
29.1K views 199 likes 7 dislikes 24 Feb 201721% match

Join the TensorFlow Community
Google for Developers
19.7K views 324 likes 7 dislikes 24 Apr 201821% match

XLA: TensorFlow, Compiled! (TensorFlow Dev Summit 2017)
Google for Developers
33.5K views 241 likes 9 dislikes 15 Feb 201720% match

Wide & Deep Learning: Memorization + Generalization with TensorFlow (TensorFlow Dev Summit 2017)
Google for Developers
24.7K views 348 likes 11 dislikes 15 Feb 201720% match

Google Developers brings its Machine Learning Bootcamp to Indonesia
Google for Developers
26.4K views 526 likes 17 dislikes 20 Feb 201920% match


Settings Privacy About Contact Donate