This commit is contained in:
Robert Kaplow 2011-08-02 17:06:06 -04:00
parent 3affee9bc0
commit 320b049354
2 changed files with 71 additions and 20 deletions

View File

@ -1,14 +1,38 @@
APIs Console Project Setup:
------------
If you have not yet, you must set your APIs Console project to enable Prediction
API and Google Storage. Go to APIs Console https://code.google.com/apis/console/
and select the project you want to use. Next, go to Services, and enable both
Prediction API and Google Storage. You may also need to enable Billing (Billing)
in the left menu.
Data Setup: Data Setup:
---------- ----------
Before you can run the prediction sample prediction.rb, you must load some csv Before you can run the prediction sample prediction.rb, you must load some csv
formatted data into Google Storage. You can do this by running setup.sh with a formatted data into Google Storage.
bucket/object name of your choice. You must first create the bucket you want to
use. This can be done with the gsutil function or via the web UI (Storage 1 - You must first create the bucket you want to use. This can be done
Access) in the Google APIs Console. i.e.: with the gsutil function or via the web UI (Storage Access) in the Google
APIs Console. i.e.
# gsutil mb gs://BUCKET
OR
Go to APIs Console -> Storage Access (on left) and the Google Storage Manager,
and create your bucket there.
2 - We now load the data you want to use to Google Storage. We have supplied a
basic language identification dataset in the sample for testing.
# chmod 744 setup.sh # chmod 744 setup.sh
# ./setup.sh BUCKET/OBJECT # ./setup.sh BUCKET/OBJECT
Note you need gsutil in your path for this to work. Note you need gsutil in your path for this to work.
If you have your own dataset, you can do this manually as well.
gsutil cp your_dataset.csv gs://BUCKET/your_dataset.csv
In the script, you must then modify the datafile string. This must correspond with the In the script, you must then modify the datafile string. This must correspond with the
bucket/object of your dataset (if you are using your own dataset). We have bucket/object of your dataset (if you are using your own dataset). We have
provided a setup.sh which will upload some basic sample data. The section is provided a setup.sh which will upload some basic sample data. The section is
@ -28,7 +52,7 @@ API. You can also set it up so the user can grant access.
First, run the google-api script to generate access and refresh tokens. Ex. First, run the google-api script to generate access and refresh tokens. Ex.
# cd google-api-ruby-client # cd google-api-ruby-client
# ruby-1.9.2-p290 bin/google-api oauth-2-login --scope=https://www.googleapis.com/auth/prediction --client-id=NUMBER.apps.googleusercontent.com --client-secret=CLIENT_SECRET # ruby bin/google-api oauth-2-login --scope=https://www.googleapis.com/auth/prediction --client-id=NUMBER.apps.googleusercontent.com --client-secret=CLIENT_SECRET
Fill in your client-id and client-secret from the API Access page. You will Fill in your client-id and client-secret from the API Access page. You will
probably have to set a redirect URI in your client ID probably have to set a redirect URI in your client ID
@ -46,6 +70,25 @@ you are loading it as a yaml, ensure you rename/move the file, as the
move the .google-api.yaml file to the sample directory. move the .google-api.yaml file to the sample directory.
Usage :
-------
At this, point, you should have
- Enabled your APIs Console account
- Created a storage bucket, if required
- Uploaded some data to Google Storage
- Modified the script to point the 'datafile' variable to the BUCKET/OBJECT name
- Modified the script to put your credentials in, either in the code or by
loading the generated .yaml file
We can now run the service!
# ruby prediction.rb
This should start a service on http://localhost:4567. When you hit the service,
your ruby logs should show the Prediction API calls, and print the prediction
output in the debug.
This sample currently does not cover some newer features of Prediction API such This sample currently does not cover some newer features of Prediction API such
as streaming training, hosted models or class weights. If there are any as streaming training, hosted models or class weights. If there are any
questions or suggestions to improve the script please email us at questions or suggestions to improve the script please email us at

View File

@ -6,7 +6,6 @@
# Original Author:: Bob Aman, Winton Davies, Robert Kaplow # Original Author:: Bob Aman, Winton Davies, Robert Kaplow
# Maintainer:: Robert Kaplow (mailto:rkaplow@google.com) # Maintainer:: Robert Kaplow (mailto:rkaplow@google.com)
$LOAD_PATH:.unshift File.dirname('lib')
require 'rubygems' require 'rubygems'
require 'sinatra' require 'sinatra'
require 'datamapper' require 'datamapper'
@ -109,6 +108,7 @@ get '/' do
# Do a prediction. # Do a prediction.
# FILL IN DESIRED INPUT: # 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."]) prediction,score = get_prediction(datafile, ["Alice noticed with some surprise."])
# ------------------------------------------------------------------------------- # -------------------------------------------------------------------------------
@ -127,10 +127,11 @@ end
def train(datafile) def train(datafile)
input = "{\"id\" : \"#{datafile}\"}" input = "{\"id\" : \"#{datafile}\"}"
puts "training input: #{input}" puts "training input: #{input}"
status, headers, body = @client.execute(@prediction.training.insert, result = @client.execute(:api_method => @prediction.training.insert,
{}, :merged_body => input,
input, :headers => {'Content-Type' => 'application/json'}
{'Content-Type' => 'application/json'}) )
status, headers, body = result.response
end end
## ##
@ -141,8 +142,9 @@ end
# then the correct string is "bucket/object" # then the correct string is "bucket/object"
# @return [Integer] status The HTTP status code of the training job. # @return [Integer] status The HTTP status code of the training job.
def get_training_status(datafile) def get_training_status(datafile)
status, headers, body = @client.execute(@prediction.training.get, result = @client.execute(:api_method => @prediction.training.get,
{'data' => datafile}) :parameters => {'data' => datafile})
status, headers, body = result.response
return status return status
end end
@ -157,11 +159,14 @@ end
def is_done?(datafile) def is_done?(datafile)
status = get_training_status(datafile) status = get_training_status(datafile)
while true do # We use an exponential backoff approach here.
test_counter = 0
while test_counter < 10 do
puts "Attempting to check model #{datafile} - Status: #{status} " puts "Attempting to check model #{datafile} - Status: #{status} "
return true if status == 200 return true if status == 200
sleep 10 sleep 5 * (test_counter + 1)
status = get_training_status(datafile) status = get_training_status(datafile)
test_counter += 1
end end
return false return false
end end
@ -184,12 +189,15 @@ def get_prediction(datafile,input_features)
# We take the input features and put it in the right input (json) format. # We take the input features and put it in the right input (json) format.
input="{\"input\" : { \"csvInstance\" : #{input_features}}}" input="{\"input\" : { \"csvInstance\" : #{input_features}}}"
puts "Prediction Input: #{input}" puts "Prediction Input: #{input}"
status, headers, body = @client.execute(@prediction.training.predict, result = @client.execute(:api_method => @prediction.training.predict,
{'data' => datafile}, :parameters => {'data' => datafile},
input, :merged_body => input,
{'Content-Type' => 'application/json'}) :headers => {'Content-Type' => 'application/json'})
prediction_data = JSON.parse(body[0]) status, headers, body = result.response
prediction_data = result.data
puts status
puts body
puts prediction_data
# Categorical # Categorical
if prediction_data["outputLabel"] != nil if prediction_data["outputLabel"] != nil
# Pull the most likely label. # Pull the most likely label.