A practical guide to implementing "autocomplete"! It follows the sometimes misunderstood principles of conditional probability distributions and the generalized Markov Chain1.
pip install autocomplete
import autocomplete
# load pickled python Counter objects representing our predictive models
# I use Peter Norvigs big.txt (http://norvig.com/big.txt) to create the predictive models
autocomplete.load()
# imagine writing "the b"
autocomplete.predict('the','b')
[('blood', 204),
('battle', 185),
('bone', 175),
('best', 149),
('body', 149),
...]
# now you type an "o"
autocomplete.predict('the','bo')
[('bone', 175),
('body', 149),
('bones', 122),
('boy', 46),
('bottom', 32),
('box', 24),
...]
If you have your own language model in the form described in ELI5, then use the models submodule to call the training method:
from autocomplete import models
models.train_models('some giant string of text')
Want to run it as a server (bottlepy required)?
import autocomplete
autocomplete.run_server()
#output
Bottle v0.12.8 server starting up (using WSGIRefServer())...
Listening on http://localhost:8080/
Hit Ctrl-C to quit.
Now head over to http://localhost:8080/the/bo
http://localhost:8080/the/bo
#output
{"body": 149, "box": 24, "bottom": 32, "boy": 46, "borzois": 16, "bodies": 13, "bottle": 13, "bones": 122, "book": 14, "bone": 175}
http://localhost:8080/the/bos
#output
{"boscombe": 11, "boston": 7, "boss": 1, "bosom": 5, "bosses": 4}
###Obligatory tests
python setup.py test
##tl;dr
The following code excerpt is my interpretation of a series of lessons/concepts expressed in a number of different books.
The unifying concept can be said to be conditional probability:
P(A , B) = P(B | A) * P(A)
Which can read as saying:
The probability of A and B occuring is equal to the probability of B occuring, given that A has occured
# "preperation" step
# for every word in corpus, normalize ('The' -> 'the'), insert to list
WORDS = helpers.re_split(corpus)
# first model -> P(word)
# Counter constructor will take a list of elements and create a frequency distribution (histogram)
WORDS_MODEL = collections.Counter(WORDS)
# another preperation step
# [a,b,c,d] -> [[a,b], [c,d]]
WORD_TUPLES = list(helpers.chunks(WORDS, 2))
# second model -> P(next word | prev. word)
# I interpret "..| prev. word)" as saying "dictionary key
# leading to seperate and smaller (than WORDS_MODEL) freq. dist.
WORD_TUPLES_MODEL = {first:collections.Counter() for first, second in WORD_TUPLES}
for prev_word, next_word in WORD_TUPLES:
# this is called the "conditioning" step where we assert
# that the probability space of all possible "next_word"'s
# is "conditioned" under the event that "prev_word" has occurred
WORD_TUPLES_MODEL[prev_word].update([next_word])
Textbooks, and locations therein, where the concept-in-practice has been expressed:
I. Intro to Statistical Natural Language Processing - Manning, Schütze, 1999
a. frequency distribution showing the most common words and frequencies in *Tom Sawyer*, pg. 21
b. conditional probability definition expressed in page 42 - section 2.1.2
c. the intuition for *frequency* distributions found in pg. 153 (provided in the context of finding [*Collocations*](http://en.wikipedia.org/wiki/Collocation))
II. Probabilistic Graphical Models - Kohler, Friedman, 2009
a. conditional probability definition found on pg. 18 (hilariously and coincidentally found in section 2.1.2.1)
III. Artificial Intelligence - A Modern Approach - Russell, Norvig, 3rd. ed. 2010
a. conditional probability concept explained in pg. 485
b. the "language" (I take to mean "intuition" for asserting things in the probabilistic sense) pg. 486
c. the notion of "conditioning" found in pg. 492-494
##Motivation
Similar to the motivation behind eatiht, I found that it took far too long to find a palpable theory-to-application example of what amounts to more than a 500 pages of words across 3 books, each spanning a large index of, in certain cases, counter-intuitive nomenclature; read the light criticisms made by Michael I. Jordan on the matter (he was recently named #2 machine learning expert "we need to know" on dataconomy.com).
You can find similar thoughts being expressed in an article from 2008 (updated 2009) by Brennan O'Connor
This work is dedicated to my siblings.
Explain like I'm 5*
*Warning! This explanation is literally intended for young kids - I'm actually trying to see if these concepts can be explained to an audience unaware of the nomenclature used within the statistical nlp and other machine learning fields. For example, my 7, 9, 11, 14 y.o. siblings, and basically anyone else who's ever read a story to a child - they would be a part of the target audience.
If you've found this readable and informative, please consider putting on the goofiest face and reading this to your kids, if you have any :) If you do, please send me your thoughts on the experience.
I'm only interested in lowering the barrier to entry. I should have included this note since the beginning (sorry to those who undoubtedly left with a bad taste in their mouths).
You can contact me at [email protected]
Thanks for reading,
Rodrigo
No. I'm explaining this like you're 5. I know you're not 5 , you guys... Chris, stop jumping on your sister's back!
Ok, so I'm saying, imagine I'm 5!
Oh, that was easy now huh? Let's just forget the I'm 5 part.
Imagine a giant collection of books.
For example, all the Harry Potter and Hunger Games novels put together.
What if I asked you to go through all the pages and all the words in those pages?
Now I'm not asking you four to actually read the books. You know, just go through, beginning to end, and notice each word.
For every new word you see, write it down, and put a "1" next to it, and everytime you see a word again, add "1" more to the previous number.
So basically I'm asking y'all to keep count of how many times a word comes up.
Got it? If yes, cool! If not, find a sibling, friend, or adult near you and ask them to help you out :)
...
Say you start with Harry Potter and the Sorcerer's Stone:
Mr. and Mrs. Dursley of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much...
And imagine that you're on the 5th word. This or something close to this is what you're going for:
Mr. -> 1
and -> 1
Mrs. -> 1
Dursley -> 1
of -> 1
Or if you're a wannabe-Harry-Potter fan, ah I'm just kidding!
If you started with the-book-that-must-not-be-named - I know you guys won't get it, but persons my age will :)
Alright! So you started with The Hunger Games:
When I wake up, the other side of the bed is cold...
By the sixth word you have:
When -> 1
I -> 1
wake -> 1
up -> 1
the -> 1
You have a long day ahead of you...
...
1,105,285 words later
Now that you're done tallying up all those words, why not order all these words by the number of times you've seen them?
See you next week!
...
Back so soon? You should have gotten something like this:
psst*, remember, the format is:
word -> # of times the word appears
'the' -> 80030
'of' -> 40025
'and' -> 38313
'to' -> 28766
'in' -> 22050
'a' -> 21155
'that'-> 12512
'he' -> 12401
'was' -> 11410
'it' -> 10681
... there's a lot more words you've tallied up...
Those were the most common words.
Now on the less-frequent end, you'll find your words appearing not as often...
... 29137 words later.
'przazdziecka' -> 1
'disclosure' -> 1
'galvanism' -> 1
'repertoire' -> 1
'bravado' -> 1
'gal' -> 1
'ideological' -> 1
'guaiacol' -> 1
'expands' -> 1
'revolvers' -> 1
Yeah Chris? Oh, 'what does lez freekend' mean? Um, so it means something like: you probably won't hear or read that word very often.
Now what if I asked you to help me find this word I'm looking for? And I know this word starts with the letters: 'th'.
I'm pretty sure you guys can do this much faster!
...
5 minutes later!
...
Not bad! You only had to go through 29157 unique words after all!
'the' -> 80030
'that' -> 12512
'this' -> 4063
'they' -> 3938
'there'-> 2972
'their'-> 2955
'them' -> 2241
'then' -> 1558
'these'-> 1231
'than' -> 1206
... 229 words more...
239 words, still kind of lot though huh? And you know your big brother, he's too lazy to do this work by hand (cough program it up cough) ;)
So the word I'm looking for is on the tip of my tongue. I think the next letter is "i".
1 minute later
'this' -> 4063
'think' -> 557
'things' -> 321
'thing' -> 303
'third' -> 239
'thin' -> 166
'thinking' -> 137
'thirty' -> 123
'thick' -> 77
'thirds' -> 43
... 36 words more...
I scan through the first 10 words. Oh, I just remembered that the next letter is 'r'.
You start taking out even more words.
10 seconds later.
'third' -> 239
'thirty' -> 123
'thirds' -> 43
'thirteen' -> 32
'thirst' -> 13
'thirteenth' -> 11
'thirdly' -> 8
'thirsty' -> 5
'thirtieth' -> 3
'thirties' -> 2
Aha, 'thirdly' was the word I was looking for! What, you never heard of the word "thirdly" before?
Now you might be saying to yourself, "that's pretty cool!", and you're right!
And you know what's cooler? Making everyone's life a tiny bit easier is! :)
But how can you do that with just words?
Aren't words boring and dull?
It's like all we do is talk, write, and think with words. I mean, how lame, I can't even describe to you this autocomplete thing-slash-idea-thing without having to write it out with words!
Ugh! I hate words!
Whoah, wait a minute! That was not cool of me! Let's relax for a minute.
Let's try to give an imaginary hug to the word-factory in our brains. That part of our brain works so hard, even when we don't ask it to. How nice of our brain to do that. Not!
What I'm trying to is sometimes it's not so nice for our brains to distract us, especially when we have homework or other, real-world, problems like adult-homework.
So how about this: let's try to think about what the next sentence coming out of our own mouths will be*.
Now if you're thinking about what will be coming out of my mouth, or out of your mouth, or your mouth, or your mouth, or your mouth, you're doing it wrong! (to readers who aren't one of my 4 younger siblings, that's how many I have).
Try your best to think about what the next sentence coming out of your own mouth will be.
...
Did you decide on your sentence? Good!
Now what if I asked you to give me two reasons explaining why and how you chose the sentence you chose?
Wait, I can't even do that! Let's make it easier on ourselves and explain why and how we chose the first word.
Still pretty hard huh?
If you thought about it, and you thought it was pretty darn hard to give a good and honest reason as to why it is you chose the word you chose, let's bring out a word you guys might not understand: probability.
If you feel like you don't get what the word means, sure you do! Just use the word "probably" in one of your sentences, but but try to makes some sense.
What do I mean? Well, let's just consider the English language. Like most other things, the English language has rules.
The kind of rules that can be simplified down to:
-
"something action something".
-
Replace something's and action with words that make sense to you.
Fair enough, right?
Now, imagine you could put "pause" right after the first word that comes out of your mouth.
Let's just say that first word is "the".
Now in the case that you stuttered for reasons outside your conscientious control (for example: "thhh thhe the"). No big deal, you meant to say "the", so let's flatten it to just that!
With that word said, what words do you think you might have said after it?
You might tell me, "any word I want!
Of course you could have! I bet you spent a millisecond thinking about whether or not the next word you were going to say was going to be: guaiacol.
I know because I thought about using that word too!
I can remember the first time I heard (or read) guaiacol like it was yesterday. I read it in some funky article on the internet. I found the word in a list of words that don't appear too often in the English language.
After I read it, I was able to fit guaiacol nicely into that part of my brain where I... uhh.. was... able... uhh...
Oh, you know, that place in my brain where I get to choose whether I want to say the apple, the automobile, the austronaut, etc.
...
Ok, so clearly I'm no brainician, and that may or may not be the way our brain works - actually, it's probably super super unlikely.
But even though that idea is probably wrong, the idea itself sounds like a pretty darn good way of suggesting the next word or words somebody is trying to type.
What if you had a way to count the number of times you've heard "apple" said after the word "the"?
Ask yourself the same question, but now with the word "automobile" instead of "apple".
What if you had the time to think about every possible word that you've ever heard spoken after the word "the"? I'd say it might have looked something like this:
Words you might have heard following the word "the" and the number of times you might have heard it
'same' -> 996
'french' -> 688
'first' -> 652
'old' -> 591
'emperor' -> 581
'other' -> 528
'whole' -> 500
'united' -> 466
'room' -> 376
'most' -> 373
... 9331 more words...
Not impressed with your brain yet? Let's continue this little thought experiment further.
Imagine that you just said "the", and you could put pause after the first letter of the next word out of your mouth: "h".
Real quick, think of the shortest amount of time you can think of. Think of the shortest second you can think of. Now shorter than that too.
At this point, you can't even call that length of time a second. But in that length of time, your brain may have just done this:
Every word you've ever heard coming after the word "the":
'house' -> 284
'head' -> 117
'hands' -> 101
'hand' -> 97
'horses' -> 71
'hill' -> 64
'highest' -> 64
'high' -> 57
'history' -> 56
'heart' -> 55
And that brain you got did this realllllyyyyyy fast. Faster than Google, Bing, Yahoo and any other company can ever hope to beat. And your brain did this without even asking for your permission. I think our brains are trying to control us you guys, oh no!
##If you're not 5
The basic idea is this:
Assume you have a large collection of Enlish-understandable text merged into a single string.
Start by transforming that string into a list of words (AKA ngrams of word-legth), and also (but not required) normalize each word ('The' -> 'the').
Once you have a normalized list of words, you can start building a frequency distribution measuring the frequency of each word.
...
At this point you can start "predict" the "final state" of a word-in-progress. But consider the case where a user types in some query box:
"The th"
And he intends to write:
"The third"
With the above predictive model, you'll be suggesting something like:
[
('the', 80030),
('they', 3938),
('there', 2972),
...
]
This explains one specific type of predictive model, which can be written as P(word), and you've just seen the pitfalls of using just this model.
Now for the next word, ask yourself, what's the probability that I'm going to type the word "apple" given that I wrote "tasty"?
In machine learning and AI books, you'll be presented Conditional Probability with the following equation:
P(word A and word B) = P(word B | word A) * P(word A)
That equation addresses the problem that I mentioned.
We've handled P(wordA) already.
To handle P(word B | word A), which reads *probability of word A given word B *, I take a literall interpretation of the word "given", in that context, to mean the following:
"word A" is the key pointing to a probability distribution representing all the words that follow "word A"
Once we can represent this second model, we can also apply the filtering step - given that we know more letters in the second word, we can zone in on more precise suggestions.
###Afterword
notes: *I have to give a shout out to Sam Harris for being, AFAIK, the first person or one of the firsts, in wonderfully putting into words what I've borrowed and slightly adapted for this writing. I highly recommend his work
Another shoutout to Peter Norvig for inspiring me and probably many others with our own little "toy" programs. His Occam's Razor approach to problem solving will likely cause some confusion as it may appear that my work is an almost full on copy-paste of his How to Write a Spell Checker!
But I swear it's not! I actually I think I may have out-Norvig'ed Peter Norvig when it comes to describing conditional probability: P(wordA & wordB) = P(wordB | wordA)*P(wordA)
And another one to Rob Renaud's Gibberish Detector. I, out of pure chance, ran into his project some time after running into Norvig's article. I can't describe how much it helped to intuitively understand what the heavy hitters of "AI" consider to be introductory material; this was greatly needed b/c at the time, I felt overwhelmed by my own desire to really understand this area, and everything else going on.
I do have a second article about this exact thing, only expressed differently (audience is non-programming), and it may or may not be posted soon! Oh and the code too, that is if someone hasn't gotten to translating the above article to code before I can get to uploading the project :P I'm trying to get the kinks out of here and the code so it's simple, duh!
I dedicate this work to my sisters, Cat, Melissa and Christine, and my favorite brother, Christian :)
####note 1
To avoid confusion, I wrote this section in the form of a letter to my younger siblings
####note 2
*I'm borrowing, what I consider, one of the most beautiful thought experiments I've ever heard trying to describe one's self. I'm a big fan of Sam Harris's work. Highly recommend!