Tutorial: Write and Run Your First App
Last updated
Was this helpful?
Last updated
Was this helpful?
An AutoMan app always has at least four parts.
As is typical in most programming languages, we have to tell Scala that we want to use AutoMan. Paste the following import
statement into the top of your source file:
Specifically, we're telling Scala that we want to use AutoMan to manage Amazon Mechanical Turk (mturk
) jobs using its MTurk domain-specific language (DSL).
If you have not yet created an account for Amazon Mechanical Turk, .
We also need to create the main class for our application. Scala makes this a little easier than Java. Paste the following into your editor below your import
statement.
The code above is a complete Scala hello-world app. As a little sanity check, before we dig into AutoMan, let's just try it out. Run the following in your terminal:
You should see output that looks a bit like this:
Notice, buried in all that output, that your program printed out Hello world!
. If you see Hello world!
in your output, move on to the next step. If not, read the sbt
output carefully to diagnose and fix the problem.
Paste the following into your MyFirstAutoManApp
class.
The code above defines a variable a
that stores an instance of mturk
. AutoMan needs what we call a platform adapter in order to know which service to connect to. In this case, we are connecting to Mechanical Turk.
Never embed your access_key_id
or your secret_access_key
in your source code! Doing so makes it easy to accidentally push your code to a public site like GitHub where they can be stolen and abused.
Be sure that you really want to run your program before setting sandbox_mode
to false
. It will run real jobs and it will spend real money!
Paste the following code below your platform adapter code.
The above defines a human function called which_one
that takes no arguments. It is important to note that which_one
is just an ordinary function in Scala, although it does behave in some special ways that we will describe in the next section.
This function creates a "radio button question" on Mechanical Turk by calling the radio
constructor. "Radio button questions" allow MTurk users to select one of n options.
The key elements in the question function above are:
budget
: This parameter specifies the maximum amount of money AutoMan will spend on this task. AutoMan always tries to spend less. If the cost of a task exceeds the budget you supply, AutoMan will shut down the task and return a "low-confidence answer."
text
: This parameter supplies the text of the question. You describe what you want workers to do here.
options
: This parameter supplies the valid options. Since this is a radio button question, each option will produce a radio button.
The choice
constructor takes three parameters:
A label
of type Symbol
. You can think of a Symbol
as a special string designed for easy comparison. This parameter is not visible to MTurk workers.
A name
, which is visible to MTurk workers.
Now that you have a human function defined, you can call it. First, paste the following AutoMan block below your function definition.
We are going to call our function inside that block. The purpose of an AutoMan block is to delineate when you are done using MTurk. AutoMan needs this information so that it knows you are ready to shut down your program.
Failing to tell AutoMan to shutdown will cause your program to hang.
Now we can call which_one()
inside our AutoMan block.
Back on the command-line, we can now run our app with:
After posted jobs are completed, somewhere in AutoMan's voluminous output, you should see something like:
If you want to see AutoMan's quality control algorithm in action, use the sandbox to supply worker responses that disagree. You should observe that after AutoMan obtains all of the responses for a given round, it will decide whether disagreement is strong enough to warrant asking the crowd for more responses.
If something goes wrong as your job runs, be sure to read AutoMan's (voluminous) output. AutoMan provides a great deal of detail to make diagnosing problems easy.
is a wonderful resource for Scala and SBT questions.
When you run this code, you will need to supply your MTurk access key and secret key on the command line. args
is an argument array, and args(0)
is the first element of that array (if you come from Java, note that arrays in Scala use ()
instead of []
). These are the same credentials you downloaded .
AutoMan has constructors for numerous question types. Refer to the for more information.
An optional image_url
, which is a link to an image hosted somewhere on the Internet, .
confidence
(not shown): This parameter stands for the and is a floating-point number between 0
and 1
(exclusive). A number approaching zero tells AutoMan that virtually any answer is fine. A number approaching one tells AutoMan that you want to be very certain that it is correct. Although the confidence
parameter is not shown above, it is set to the default of 0.95
, which is something of a standard threshold across empirical science.
Here is the complete source code for our first AutoMan program (with the hello-world bit removed). You can find a copy of this program in the of AutoMan's GitHub repository, along with many other examples.
Notice that the above program does not store any private access keys in the source text! Instead, they must be passed in using command line arguments. For additional examples with more sophisticated command line parsing, see the .
Assuming you have not changed the mturk
initializer, this program will post jobs to the MTurk sandbox. When you run this program in the sandbox, no work will get done, because only the live production site has active workers. You must simulate the job yourself. We describe the process of simulating a job in the section titled .
We provide additional example programs that work around these issues in the apps
directory of our . An efficient, error-handling program that defines essentially the same task .