pdorg_site/talks/your-own-bayes/index.txt

# Write your own Bayesian Classifier!

John Melesky
(Open Source Bridge, June 2009)
---
# What's a Bayesian Classifier?
---
# What's a Bayesian Classifier?

Something which classifies based on:

1. Information about past categorizations
2. Bayesian statistics (Bayes' Theorem)
---
# What's Bayes' Theorem?

Let's check [Wikipedia](http://phaedrusdeinus.org/Bayes'_theorem.html).
---
# Derrr....
---
# An example: random drug testing

3% of the population are using Zopadrine.

We have a drug test with a 98% accuracy rate.
---
# An example: random drug testing

3% of the population are using Zopadrine.

We have a drug test with a 98% accuracy rate.

Bob is tested, and the result is positive. How likely is it that Bob uses Zopadrine?
---
# Break it down

Let's assume a population of 10000 people.
---
# Break it down

3% are users.

<table border=1>
<tr><td></td><td>Population</td></tr>
<tr><td>Clean</td><td>9700</td></tr>
<tr><td>Users</td><td>300</td></tr>
<tr><td>Total</td><td>10000</td></tr>
</table>
---
# Break it down

The test is 98% accurate.

<table border=1>
<tr><td></td><td>Population</td><td>Test negative</td><td>Test positive</td></tr>
<tr><td>Clean</td><td>9700</td><td>9506</td><td>194</td></tr>
<tr><td>Users</td><td>300</td><td>6</td><td>294</td></tr>
<tr><td>Total</td><td>10000</td><td>9512</td><td>488</td></tr>
</table>
---
# Break it down

Bob is tested, and the result is positive. How likely is it that Bob uses Zopadrine?

<table border=1>
<tr><td></td><td>Population</td><td>Test negative</td><td>Test positive</td></tr>
<tr><td>Clean</td><td>9700</td><td>9506</td><td>194</td></tr>
<tr><td>Users</td><td>300</td><td>6</td><td bgcolor="#ff6666">294</td></tr>
<tr><td>Total</td><td>10000</td><td>9512</td><td bgcolor="#ff6666">488</td></tr>
</table>
---
# Break it down

294 / 488 = 60.24%
---
# Back to Bayes' Theorem

![Bayes' Theorem](img/bayes.png)
---
# Back to Bayes' Theorem

<table>
<tr><td>P = probability</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>A = "is a user"</td></tr>
<tr><td>B = "tests positive"</td></tr>
<tr><td>x|y = x, given y</td></tr>
</table>
---
# Back to Bayes' Theorem

<table>
<tr><td>P(A) = probability of being a user</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>P(B|A) = probability of testing positive, given being a user</td></tr>
<tr><td>P(B) = probability of testing positive</td></tr>
<tr><td>P(A|B) = probability Bob's a user</td></tr>
</table>
---
# Back to Bayes' Theorem

<table>
<tr><td>P(A) = 3%</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>P(B|A) = probability of testing positive, given being a user</td></tr>
<tr><td>P(B) = probability of testing positive</td></tr>
<tr><td>P(A|B) = probability Bob's a user</td></tr>
</table>
---
# Back to Bayes' Theorem

<table>
<tr><td>P(A) = 3%</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>P(B|A) = 98%</td></tr>
<tr><td>P(B) = probability of testing positive</td></tr>
<tr><td>P(A|B) = probability Bob's a user</td></tr>
</table>
---
# Back to the numbers

<table border=1>
<tr><td></td><td>Population</td><td>Test negative</td><td>Test positive</td></tr>
<tr><td>Clean</td><td>9700</td><td>9506</td><td>194</td></tr>
<tr><td>Users</td><td>300</td><td>6</td><td>294</td></tr>
<tr><td>Total</td><td bgcolor="#ff6666">10000</td><td>9512</td><td bgcolor="#ff6666">488</td></tr>
</table>
---
# Back to Bayes' Theorem

<table>
<tr><td>P(A) = 3%</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>P(B|A) = 98%</td></tr>
<tr><td>P(B) = 4.88%</td></tr>
<tr><td>P(A|B) = probability Bob's a user</td></tr>
</table>
---
# Back to Bayes' Theorem

<table>
<tr><td>P(A) = 3%</td><td rowspan="4"><img src="img/bayes.png"></td></tr>
<tr><td>P(B|A) = 98%</td></tr>
<tr><td>P(B) = 4.88%</td></tr>
<tr><td>P(A|B) = (98% * 3%)/4.88% = 60.24%</td></tr>
</table>
---
# This works with population numbers, too

    P(A)   = 300
    P(B|A) = 9800
    P(B)   = 488
    P(A|B) = 6024

Which is useful for reasons we'll see later.
---
# Bayes' Theorem, in code

My examples are going to be in perl.

    sub bayes {
      my ($p_a, $p_b, $p_b_a) = @_;

      my $p_a_b = ($p_b_a * $p_a) / $p_b;

      return $p_a_b;
    }
---
# Bayes' Theorem, in code

But you could just as easily work in Python.

    def bayes(p_a, p_b, p_b_a):
      return (p_b_a * p_a) / p_b
---
# Bayes' Theorem, in code

Or Java

    public static Double bayes(Double p_a, Double p_b, Double p_b_a) {
      Double p_a_b = (p_b_a * p_a) / p_b;

      return p_a_b;
    }

---
# Bayes' Theorem, in code

Or SML

    let bayes(p_a, p_b, p_b_a) = (p_b_a * p_a) / p_b

---
# Bayes' Theorem, in code

Or Erlang

    bayes(p_a, p_b, p_b_a) ->
      (p_b_a * p_a) / p_b.

---
# Bayes' Theorem, in code

Or Haskell

    bayes p_a p_b p_b_a = (p_b_a * p_a) / p_b

---
# Bayes' Theorem, in code

Or Scheme

    (define (bayes p_a p_b p_b_a)
      (/ (* p_b_a p_a) p_b))
---
# Bayes' Theorem, in code

LOLCODE, anyone? Befunge? Unlambda?

If it supports floating point operations, you're set.
---
# How does that make a classifier?

    A = "is spam"
    B = "contains the string 'viagra'"

What's P(A|B)?
---
# What do we need for a classifier?

1. We need to tokenize our training set
2. Then build a model
3. Then test that model
4. Then apply that model to new data
---
# What do we need for a classifier?

1. **We need to tokenize our training set**
2. Then build a model
3. Then test that model
4. Then apply that model to new data

---
# Tokenizing your training set

*Fancy* perl

    sub tokenize {
      my $contents = shift;

      my %tokens = map { $_ => 1 } split(/\s+/, $contents);
      return %tokens;
    }

---
# Tokenizing your training set

    sub tokenize_file {
      my $filename = shift;
    
      my $contents = '';
      open(FILE, $filename);
      read(FILE, $contents, -s FILE);
      close(FILE);
    
      return tokenize($contents);
    }

---
# Tokenizing your training set

This is the "bag of words" model.

For each category (spam, not spam), we need to know how many documents in the training set contain a given word.
---
# Tokenizing your training set

    my %spam_tokens = ();
    my %notspam_tokens = ();
    
    foreach my $file (@spam_files) {
      my %tokens = tokenize_file($file);
      %spam_tokens = combine_hash(\%spam_tokens, \%tokens);
    }
    
    foreach my $file (@notspam_files) {
      my %tokens = tokenize_file($file);
      %notspam_tokens = combine_hash(\%notspam_tokens, \%tokens);
    }

---
# Tokenizing your training set

    sub combine_hash {
      my ($hash1, $hash2) = @_;
    
      my %resulthash = %{ $hash1 };
    
      foreach my $key (keys(%{ $hash2 })) {
        if ($resulthash{$key}) {
          $resulthash{$key} += $hash2->{$key};
        } else {
          $resulthash{$key} = $hash2->{$key};
        }
      }
    
      return %resulthash;
    }
---
# What do we need for a classifier?

1. We need to tokenize our training set
2. **Then build a model**
3. Then test that model
4. Then apply that model to new data
---
# Build a model

    my %total_tokens = combine_hash(\%spam_tokens, \%notspam_tokens);
    
    my $total_spam_files = scalar(@spam_files);
    my $total_notspam_files = scalar(@notspam_files);
    my $total_files = $total_spam_files + $total_notspam_files;
    my $probability_spam = $total_spam_files / $total_files;
    my $probability_notspam = $total_notspam_files / $total_files;
---
# Build a model

In this case, our model is just a bunch of numbers. 

---
# Build a model

In this case, our model is just a bunch of numbers. 

(a little secret: it's *all* a bunch of numbers)

---
# What do we need for a classifier?

1. We need to tokenize our training set
2. Then build a model
3. **Then test that model**
4. Then apply that model to new data
---
# \*cough\* \*cough\*

---
# What do we need for a classifier?

1. We need to tokenize our training set
2. Then build a model
3. Then test that model
4. **Then apply that model to new data**
---
# Apply that model to new data

    my %test_tokens = tokenize_file($test_file);

    foreach my $token (keys(%test_tokens)) {
      if (exists($total_tokens{$token})) {
        my $p_t_s = (($spam_tokens{$token} || 0) + 1) /
                    ($total_spam_files + $total_tokens);
        $spam_accumulator = $spam_accumulator * $p_t_s;

        my $p_t_ns = (($notspam_tokens{$token} || 0) + 1) /
                     ($total_notspam_files + $total_tokens);

        $notspam_accumulator = $notspam_accumulator * $p_t_ns;
      }
    }
---
# Apply that model to new data

    my $score_spam = bayes( $probability_spam,
                            $total_tokens,
                            $spam_accumulator );
    
    my $score_notspam = bayes( $probability_notspam,
                               $total_tokens,
                               $notspam_accumulator );
    
    my $likelihood_spam = $score_spam / ($score_spam + $score_notspam);
    my $likelihood_notspam = $score_notspam / ($score_spam + $score_notspam);
    
    printf("likelihood of spam email: %0.2f %%\n", ($likelihood_spam * 100));

---
# Boom
---
# What sucks?
---
# What sucks?

- Our tokenization
---
# What sucks?

- Our tokenization
- Our memory limitations
---
# What sucks?

- Our tokenization
- Our memory limitations
- Saving/loading models
---
# Improve memory use

### Limit the number of tokens

We want to use the tokens with the highest information values. That means tokens that are predominantly in one category but not the other.
---
# Improve memory use

### Limit the number of tokens

We want to use the tokens with the highest information values. That means tokens that are predominantly in one category but not the other.

There are a bunch of ways to calculate this, though the big one is Information Gain.

---
# Improve tokenization, simple stuff

- Weed out punctuation
- Weed out stopwords
- normaLize CASE
- Strip out markup
---
# Improve tokenization, advanced stuff

### Stemming

"wrestling", "wrestler", "wrestled", and "wrestle" are all the same word concept.

Pros: fewer tokens, related tokens match

Cons: some words are hard to stem correctly (e.g. "cactus")

---
# Improve tokenization, advanced stuff

### Include bigrams

Bigrams are token pairs. For example, "open source", "ron paul", "twitter addict".

Pros: we start distinguishing between Star Wars and astronomy wars

Cons: our memory use balloons

---
# Improve tokenization, advanced stuff

### Use numbers

Instead of binary (word x is in doc y), we store frequencies (word x appears z times in doc y).

Pros: damage from weak associations is reduced; easier to find the important words in a document

Cons: the math becomes more complex; in many cases, accuracy doesn't actually increase

---
# Improve tokenization, advanced stuff

### Use non-token features

Sometimes we want to use non-textual attributes of documents. For example, length of document, percent of capital letters.

---
# Improve tokenization, advanced stuff

### Use non-token features

Sometimes we want to use non-textual attributes of documents. For example, length of document, percent of capital letters.

We can also grab structural information, like the sender, or subject line, and treat them differently. Or whether the word appears early or late in the document.

---
# Improve tokenization, advanced stuff

### Use non-token features

Sometimes we want to use non-textual attributes of documents. For example, length of document, percent of capital letters.

We can also grab structural information, like the sender, or subject line, and treat them differently. Or whether the word appears early or late in the document.

Pros: a little can go a long way

Cons: selecting these can be a dark art. or an incredible memory burden.

---
# Which leads us to

---
# Which leads us to

Tokenization == Vectorization
---
# In other words

Our documents are all just vectors of numbers.
---
# Or even

Our documents are all just points in a high-dimensional Cartesian space.
---
# Vectors of numbers

This concept opens up a whole world of statistical methods for categorization, including decision trees, linear separations, and support vector machines.

---
# Points in space

And this opens up a whole different world of geometric methods for categorization and information manipulation, including k-nearest-neighbor classification and various clustering algorithms.
---
# Alright

It's been a long trip. Any questions?
---
# Thanks

Thanks for coming. Thanks to OS Bridge for having me.