vigosan / big_ml Goto Github PK

would be cool to support at least 1.8.x and 1.9.2 appart from 1.9.3

raise error in a more readable way when the status of the action is not correct

Need to update the version that is installed using gem install big_ml . The newest version includes in client.rb:

VALID_OPTIONS_KEYS     = [:username, :api_key, :dev_mode]

However, the version from gem install is:

VALID_OPTIONS_KEYS     = [:username, :api_key]

With installation information, how to run test suit and use case, also add more information from BIGML

Have the ability to configure the gem for use the development mode, f.e

BigML.configure do |c|
  c.username = 'foo'
  c.api_key = 'bar'
  c.dev = true
end

Make it go free, need to version correctly and update the gemspec

And you can't inspect right now the status of the request, make sure we provide a mechanism for improving the visibility over there.

Hi,

Documentation is a bit sparse, so I'm having trouble figuring out if I'm taking the right approach here. To predict a single data point, I have the following working:

  desc "Predict categories"
  task :predict => :environment do
    model = BigML::Model.find("...")
    product = Product.find(1)
    fields = {
      "000000" => product.name,
      "000001" => product.brand_name,
      "000002" => product.description
    }
    prediction = BigML::Prediction.create(model.resource, :input_data => fields)
    puts "#{prediction.attrs['output']} (#{prediction.attrs['confidence']})"
    prediction.destroy
  end

But if I want to predict en masse, this seems rather inefficient. What I'd like to do is to predict a multiple inputs at once. E.g.:

  desc "Predict categories"
  task :predict => :environment do
    model = BigML::Model.find("...")
    input_data = []
    Product.each do |product|
      input_data << {
        "000000" => product.name,
        "000001" => product.brand_name,
        "000002" => product.description
      }
    end
    prediction = BigML::Prediction.create(model.resource, :input_data => input_data)
    ...
    prediction.destroy
  end

But that doesn't work. Should I create a dataset in this case? And how would the code for that look?

No need to include the testing gems as general dependency for the gem.

I think we can put it out of the Gemfile in that way you only loaded them when you need to run the suit