Class: GooglePrediction

Inherits:
Object
  • Object
show all
Defined in:
lib/google-prediction.rb

Overview

Provides an interface to the Google Prediction API.

Author

Sam King ([email protected])

Support

[email protected]

Copyright

InSTEDD

License

GPLv3

Usage

auth_token = GooglePrediction.get_auth_token('[email protected]', 'password') => long string of letters and numbers

predictor = GooglePrediction.new(auth_token, 'bucket', 'object')

predictor.#train => "data"=>{"data"=>"bucket/object"}

predictor.#check_training => "Training has not completed"

wait_some_time

predictor.#check_training => "no estimate available" or something between "0.0" and "1.0"

predictor.#predict "awesome company" => "Google" predictor.#predict "awesome nonprofit" => "InSTEDD" predictor.#predict 13 => "lucky" predictor.#predict [3, 5, 7, 11] => "prime"

Constant Summary collapse

PREDICTION_URL_PREFIX = 
'https://www.googleapis.com/prediction/v1/training'

Instance Attribute Summary collapse

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(auth_code, bucket, object) ⇒ GooglePrediction

auth_code: the login code generated from self.get_auth_token. If you get your auth_code from somewhere else, make sure that it doesn't include any extra characters like a newline at the end. That can cause bizarre errors.

bucket: the name of the bucket in Google Storage

object: the filename of the object to do prediction on



79
80
81
82
83
 
# File 'lib/google-prediction.rb', line 79

def initialize(auth_code, bucket, object)
  @auth_code=auth_code
  @bucket=bucket
  @object=object
end

Instance Attribute Details

#auth_code ⇒ Object (readonly)

Returns the value of attribute auth_code.



45
46
47
 
# File 'lib/google-prediction.rb', line 45

def auth_code
  @auth_code
end

#bucket ⇒ Object (readonly)

Returns the value of attribute bucket.



46
47
48
 
# File 'lib/google-prediction.rb', line 46

def bucket
  @bucket
end

#object ⇒ Object (readonly)

Returns the value of attribute object.



47
48
49
 
# File 'lib/google-prediction.rb', line 47

def object
  @object
end

Class Method Details

.check_training(auth_code, bucket, object) ⇒ Object

Wrapper. Creates a new object and runs #check_training on it.



167
168
169
170
 
# File 'lib/google-prediction.rb', line 167

def self.check_training(auth_code, bucket, object)
  predictor = GooglePrediction.new(auth_code, bucket, object)
  predictor.check_training
end

.get_auth_token(email, password, url = 'https://www.google.com/accounts/ClientLogin', args = {}) ⇒ Object

Gets the auth code from Google's ClientLogin using the provided email and password.

This will fail if Google requires a Captcha. If so, follow the instructions at
http://code.google.com/apis/predict/docs/getting-started.html and pass in the new URL and new arguments using the optional parameters.



56
57
58
59
60
61
62
63
64
65
66
67
68
69
 
# File 'lib/google-prediction.rb', line 56

def self.get_auth_token(email, password, url='https://www.google.com/accounts/ClientLogin', args={}) 
  curl = Curl::Easy.new(url)
  post_args = {
    "accountType" => "HOSTED_OR_GOOGLE",
    "Email" => email,
    "Passwd" => curl.escape(password),
    "source" => "companyName-applicationName-versionID",
    "service" => "xapi"
  }
  args.each {|key, val| post_args[key] = val }
  post_fields = post_args.map {|k,v| Curl::PostField.content(k, v) }
  curl.http_post(post_fields)
  curl.body_str.match('Auth.*')[0][5..-1]
end

.predict(auth_code, bucket, object, submission) ⇒ Object

Wrapper. Creates a new object and runs #predict on it.



173
174
175
176
 
# File 'lib/google-prediction.rb', line 173

def self.predict(auth_code, bucket, object, submission)
  predictor = GooglePrediction.new(auth_code, bucket, object)
  predictor.predict(submission)
end

.train(auth_code, bucket, object) ⇒ Object

Wrapper. Creates a new object and runs #train on it.



161
162
163
164
 
# File 'lib/google-prediction.rb', line 161

def self.train(auth_code, bucket, object)
  predictor = GooglePrediction.new(auth_code, bucket, object)
  predictor.train
end

Instance Method Details

#check_training ⇒ Object

Gets the training status of the specified object.

If the training is incomplete, returns "Training has not completed".
If the training did not have enough data to do cross-fold validation, returns "no estimate available". If the training went as desired, returns the accuracy of the training from 0 to 1.

Returns of your errors], "code"=>code, "message"=>message}} on error.



113
114
115
116
117
118
119
120
121
122
123
124
125
 
# File 'lib/google-prediction.rb', line 113

def check_training
  url = PREDICTION_URL_PREFIX + "/" + @bucket + "%2F" + @object
  curl = Curl::Easy.new(url)
  curl.headers = {"Authorization" => "GoogleLogin auth=#{@auth_code}"}
  curl.http_get

  # response will be 
  #   {"data"=>{"data"=>"bucket/object", "modelinfo"=>accuracy_prediction}}
  # on success
  response = JSON.parse(curl.body_str) 
  return response["data"]["modelinfo"] unless response["data"].nil?
  return response
end

#predict(submission) ⇒ Object

Submission must be either a string, a single number, or an array of strings or numbers

Gets the prediction for the label of the submission based on the training.

Returns the prediction on success and of your errors], "code"=>code, "message"=>message}} on error.



135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
 
# File 'lib/google-prediction.rb', line 135

def predict(submission)
  url = PREDICTION_URL_PREFIX + "/" + @bucket + "%2F" + @object + "/predict"
  curl = Curl::Easy.new(url)
  post_body = {:data => {:input => {}}}
  submission = [submission] unless submission.is_a? Array
  if submission[0].is_a? String
    post_body[:data][:input] = {:text => submission}
  elsif submission[0].is_a? Fixnum
    post_body[:data][:input] = {:numeric => submission}
  else
    raise Exception.new("submission must be String, Fixnum, or Array of Strings / Fixnums")
  end
  curl.post_body = JSON.generate(post_body)
  curl.headers = {"Content-Type" => "application/json", 
    "Authorization" => "GoogleLogin auth=#{@auth_code}"}
  curl.http("POST")

  # response will be
  #   {"data"=>{"output"=>{"output_label"=>label}}}
  # on success
  response = JSON.parse(curl.body_str)
  return response["data"]["output"]["output_label"] unless response["data"].nil?
  return response
end

#train ⇒ Object

Starts training on the specified object.

Returns "data"=>{"data"=>"bucket/object"} on success, and of your errors], "code"=>code, "message"=>message}} on error.



92
93
94
95
96
97
98
99
100
 
# File 'lib/google-prediction.rb', line 92

def train
  url = PREDICTION_URL_PREFIX + "?data=" + @bucket + "%2F" + @object
  curl = Curl::Easy.new(url)
  curl.post_body = JSON.generate({:data => {}})
  curl.headers = {"Content-Type" => "application/json", 
    "Authorization" => "GoogleLogin auth=#{@auth_code}"}
  curl.http("POST")
  JSON.parse(curl.body_str)
end