Overhauled the prediction sample and updated to v1.4.

This commit is contained in:
Bob Aman 2011-10-11 23:55:53 +03:00
parent 2b746cb379
commit 1c300f091f
4 changed files with 633 additions and 4131 deletions

File diff suppressed because it is too large Load Diff

View File

@ -1,8 +1,8 @@
#!/usr/bin/ruby1.8
#!/usr/bin/ruby1.8
# -*- coding: utf-8 -*-
# Copyright:: Copyright 2011 Google Inc.
# License:: All Rights Reserved.
# License:: Apache 2.0
# Original Author:: Bob Aman, Winton Davies, Robert Kaplow
# Maintainer:: Robert Kaplow (mailto:rkaplow@google.com)
@ -12,7 +12,7 @@ require 'datamapper'
require 'google/api_client'
require 'yaml'
use Rack::Session::Pool, :expire_after => 86400 # 1 day
enable :sessions
# Set up our token store
DataMapper.setup(:default, 'sqlite::memory:')
@ -20,8 +20,8 @@ class TokenPair
include DataMapper::Resource
property :id, Serial
property :refresh_token, String
property :access_token, String
property :refresh_token, String, :length => 255
property :access_token, String, :length => 255
property :expires_in, Integer
property :issued_at, Integer
@ -43,10 +43,32 @@ class TokenPair
end
TokenPair.auto_migrate!
before do
def save_token_pair(session, client)
token_pair = if session[:token_id]
TokenPair.first_or_create(:id => session[:token_id])
else
TokenPair.new
end
token_pair.update_token!(client.authorization)
if token_pair.save
session[:token_id] = token_pair.id
else
token_pair.errors.each do |e|
raise e
end
end
end
# FILL IN THIS SECTION
# This is the name of the {bucket}/{object} you are using for the language
# file.
# ------------------------
DATA_OBJECT = "bucket/object"
# ------------------------
before do
# FILL IN THIS SECTION
# This will work if your yaml file is stored as ./google-api.yaml
# This will work if your yaml file is stored as .google-api.yaml
# ------------------------
oauth_yaml = YAML.load_file('.google-api.yaml')
@client = Google::APIClient.new
@ -59,20 +81,17 @@ before do
@client.authorization.redirect_uri = to('/oauth2callback')
# Workaround for now as expires_in may be nil, but when converted to int it becomes 0.
@client.authorization.expires_in = 1800 if @client.authorization.expires_in.to_i == 0
if session[:token_id]
# Load the access token here if it's available
token_pair = TokenPair.get(session[:token_id])
@client.authorization.update_token!(token_pair.to_hash)
@client.authorization.update_token!(token_pair.to_hash) if token_pair
end
if @client.authorization.refresh_token && @client.authorization.expired?
@client.authorization.fetch_access_token!
save_token_pair(session, @client)
end
@prediction = @client.discovered_api('prediction', 'v1.3')
@prediction = @client.discovered_api('prediction', 'v1.4')
unless @client.authorization.access_token || request.path_info =~ /^\/oauth2/
redirect to('/oauth2authorize')
end
@ -84,144 +103,80 @@ end
get '/oauth2callback' do
@client.authorization.fetch_access_token!
# Persist the token here
token_pair = if session[:token_id]
TokenPair.get(session[:token_id])
else
TokenPair.new
end
token_pair.update_token!(@client.authorization)
token_pair.save()
session[:token_id] = token_pair.id
save_token_pair(session, @client)
redirect to('/')
end
get '/' do
# FILL IN DATAFILE:
# ----------------------------------------
datafile = "BUCKET/OBJECT"
# ----------------------------------------
# Train a predictive model.
train(datafile)
# Check to make sure the training has completed.
if (is_done?(datafile))
# Do a prediction.
# FILL IN DESIRED INPUT:
# -------------------------------------------------------------------------------
# Note, the input features should match the features of the dataset.
prediction,score = get_prediction(datafile, ["Alice noticed with some surprise."])
# -------------------------------------------------------------------------------
# We currently just dump the results to output, but you can display them on the page if desired.
puts prediction
puts score
end
erb :index
end
##
# Trains a predictive model.
#
# @param [String] filename The name of the file in Google Storage. NOTE: this do *not*
# include the gs:// part. If the Google Storage path is gs://bucket/object,
# then the correct string is "bucket/object"
def train(datafile)
input = "{\"id\" : \"#{datafile}\"}"
puts "training input: #{input}"
result = @client.execute(:api_method => @prediction.training.insert,
:merged_body => input,
:headers => {'Content-Type' => 'application/json'}
)
status, headers, body = result.response
get '/train' do
training = @prediction.trainedmodels.insert.request_schema.new
training.id = 'language-sample'
training.storage_data_location = DATA_OBJECT
result = @client.execute(
:api_method => @prediction.trainedmodels.insert,
:headers => {'Content-Type' => 'application/json'},
:body_object => training
)
end
##
# Returns the current training status
#
# @param [String] filename The name of the file in Google Storage. NOTE: this do *not*
# include the gs:// part. If the Google Storage path is gs://bucket/object,
# then the correct string is "bucket/object"
# @return [Integer] status The HTTP status code of the training job.
def get_training_status(datafile)
result = @client.execute(:api_method => @prediction.training.get,
:parameters => {'data' => datafile})
status, headers, body = result.response
return status
end
get '/checkStatus' do
result = @client.execute(
:api_method => @prediction.trainedmodels.get,
:parameters => {'id' => 'language-sample'}
)
##
# Checks the training status until a model exists (will loop forever).
#
# @param [String] filename The name of the file in Google Storage. NOTE: this do *not*
# include the gs:// part. If the Google Storage path is gs://bucket/object,
# then the correct string is "bucket/object"
# @return [Bool] exists True if model exists and can be used for predictions.
def is_done?(datafile)
status = get_training_status(datafile)
# We use an exponential backoff approach here.
test_counter = 0
while test_counter < 10 do
puts "Attempting to check model #{datafile} - Status: #{status} "
return true if status == 200
sleep 5 * (test_counter + 1)
status = get_training_status(datafile)
test_counter += 1
end
return false
end
##
# Returns the prediction and most most likely class score if categorization.
#
# @param [String] filename The name of the file in Google Storage. NOTE: this do *not*
# include the gs:// part. If the Google Storage path is gs://bucket/object,
# then the correct string is "bucket/object"
# @param [List] input_features A list of input features.
#
# @return [String or Double] prediction The returned prediction, String if categorization,
# Double if regression
# @return [Double] trueclass_score The numeric score of the most likely label. (Categorical only).
def get_prediction(datafile,input_features)
# We take the input features and put it in the right input (json) format.
input="{\"input\" : { \"csvInstance\" : #{input_features}}}"
puts "Prediction Input: #{input}"
result = @client.execute(:api_method => @prediction.training.predict,
:parameters => {'data' => datafile},
:merged_body => input,
:headers => {'Content-Type' => 'application/json'})
status, headers, body = result.response
prediction_data = result.data
puts status
puts body
puts prediction_data
# Categorical
if prediction_data["outputLabel"] != nil
# Pull the most likely label.
prediction = prediction_data["outputLabel"]
# Pull the class probabilities.
probs = prediction_data["outputMulti"]
puts probs
# Verify we are getting a value result.
puts ["ERROR", input_features].join("\t") if probs.nil?
return "error", -1.0 if probs.nil?
# Extract the score for the most likely class.
trueclass_score = probs.select{|hash|
hash["label"] == prediction
}[0]["score"]
# Regression.
# Assemble some JSON our client-side code can work with.
json = {}
if result.status != 200
if result.data["error"]
message = result.data["error"]["errors"].first["message"]
json["message"] = "#{message} [#{result.status}]"
else
json["message"] = "Error. [#{result.status}]"
end
json["response"] = ::JSON.parse(result.body)
json["status"] = "error"
else
prediction = prediction_data["outputValue"]
# Class core unused.
trueclass_score = -1
json["response"] = ::JSON.parse(result.body)
json["status"] = "success"
end
puts [prediction,trueclass_score,input_features].join("\t")
return prediction,trueclass_score
return [
200,
[["Content-Type", "application/json"]],
::JSON.generate(json)
]
end
post '/predict' do
input = @prediction.trainedmodels.predict.request_schema.new
input.input = {}
input.input.csv_instance = [params["input"]]
result = @client.execute(
:api_method => @prediction.trainedmodels.predict,
:parameters => {'id' => 'language-sample'},
:headers => {'Content-Type' => 'application/json'},
:body_object => input
)
json = {}
if result.status != 200
if result.data["error"]
message = result.data["error"]["errors"].first["message"]
json["message"] = "#{message} [#{result.status}]"
else
json["message"] = "Error. [#{result.status}]"
end
json["response"] = ::JSON.parse(result.body)
json["status"] = "error"
else
json["response"] = ::JSON.parse(result.body)
json["status"] = "success"
end
return [
200,
[["Content-Type", "application/json"]],
::JSON.generate(json)
]
end

0
examples/prediction/setup.sh Normal file → Executable file
View File

View File

@ -0,0 +1,86 @@
<!DOCTYPE html>
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<title>Prediction API</title>
<style type="text/css">
body {
font-family: Arial, Helvetica, sans-serif;
}
#log {
font-family: monospace;
background-color: #eee;
padding: 1em;
}
#log p {
margin: 0;
}
#predict {
display: none;
}
#predict label, #predict textarea, #predict button {
margin: 1em 0;
font-size: 1em;
display: block;
width: 50%;
}
</style>
</head>
<body>
<h1>Prediction API: Language Sample</h1>
<div id="log">
</div>
<div id="predict">
<label for="input">Input</label>
<textarea id="input" placeholder="Généralement, les gens qui savant peu parlent beaucoup, et les gens qui savant beaucoup parlent peu."></textarea>
<button id="go">Submit</button>
</div>
<script src="//ajax.googleapis.com/ajax/libs/jquery/1.6.2/jquery.min.js"></script>
<script type="text/javascript">
function logMessage(message) {
$("#log").append("<p>" + message + "</p>");
}
$(document).ready(function(e) {
$.getJSON("/train", function (data) {
logMessage("Training started...");
var delay = 1000;
var checkStatus = function () {
logMessage("Checking training status...");
$.getJSON("/checkStatus", function(data) {
if (data && data.status == 'success') {
logMessage("Training complete.");
$("#predict").show();
$("#go").click(function () {
var input = $("#input").val();
$.ajax({
type: "POST",
url: "/predict",
data: {"input": input},
success: function(data) {
if (data && data.status == 'success') {
logMessage("Predicted label: " + data.response.outputLabel);
} else if (data && data.message) {
logMessage(data.message);
}
}
});
});
return;
} else if (data && data.message) {
logMessage(data.message);
}
delay = delay * 2;
if (delay > 30000) {
// Upper maximum delay.
delay = 30000;
}
logMessage("Checking again in " + (delay / 1000) + " seconds.");
setTimeout(checkStatus, delay);
});
};
setTimeout(checkStatus, delay);
});
})
</script>
</body>
</html>