Before setting up Troia you have to configure environment on target computer so it can prepare and run this service. You have to download the following software:
- java - http://www.oracle.com/technetwork/java/javase/downloads/index.html
- tomcat - http://tomcat.apache.org/
If you want to compile Troia from source you will need:
- git - http://git-scm.com
- maven - http://maven.apache.org/
Note that if you have package manager, like Synaptic, you can simply download all this software with it. If you are going to use package manager remember to download tomcat6-admin package as well. As for servlet container you can use different one but this tutorial will be written with Tomcat in mind.
You can find detailed documentation on installing tomcat here.
In order to be able to use Tomcat's administrative functions, you must create an user with administrator privileges. You can do that by modifying tomcat-users.xml file.
- if you are using Linux, the file is probably located at /etc/tomcat6/tomcat-users.xml path
- on Windows, the file can be found at the location <CATALINA_HOME>/conf/tomcat-users.xml
To create an user with full required privileges you must add the following line into this file
<user username="administrator" password="drowssap" roles="admin,manager-gui" />
This will create user administrator with password drowssap and all privileges needed for this project.
To deploy the .war file, open Tomcat web application manager. If you are setting up Troia on the localhost and Tomcat runs on port 8080, web application manager will be up at the following address:
http://localhost:8080/manager/html
Find the part of the manager that allows you to upload and deploy .war files from disk. After that you should simply choose .war that was generated in previous step and deploy it. If the deployment succeeds, you will see Troia listed under Applications panel and its status will be 'Running'. If you'll the visit following url:
http://localhost:8080/troia/status
As a result you should get a JSON response saying that server status is OK and it's job storage is CachedWithRegularDumpMEMORY_FULLUsingExecutor.
That's all - now you can start playing with troia server. But it is worth mentioning that after restarting tomcat all data you've provided to troia will be lost (all data is stored in memory). To add persistance you also need to download:
- mysql - http://www.mysql.com/
and configure your server to use database.
Troia server should work out of the box (using in memory jobs storage). But there are some parameters you might find useful. To change them go to the following url:
http://localhost:8080/troia/config
Parameters meaning:
- CACHE_DUMP_TIME - for how long (in seconds) job should be stored in the cache
- CACHE_SIZE - how many jobs should be stored in the cache
- DOWNLOADS_PATH - the path for downloading the zip archives containing the computation results
- EXECUTOR_THREADS_NUM - number of executor threads
- RESPONSES_CACHE_SIZE - how many responses should be store in the cache
- RESPONSES_DUMP_TIME - for how long (in seconds) responses should be stored in the cache
- JOBS_STORAGE - you can select from: MEMORY_FULL, MEMORY_KV, DB_FULL, DB_KV_MEMCACHE_SIMPLE or DB_KV_SIMPLE
DB parameters (used only if JOBS_STORAGE is set to DB_something:
- DB_DRIVER_CLASS - using mysql is recommended but not required
- DB_NAME - the database name
- DB_PASSWORD - mysql user's password
- DB_URL - the url of mysql host machine
- DB_USER - mysql user who owns the database
Memcache parameters (used only if JOBS_STORAGE is set to DB_KV_MEMCACHE_something):
- MEMCACHE_EXPIRATION_TIME - for how long (in seconds) values should be stored in the memcache
- MEMCACHE_PORT - memcache service port
- MEMCACHE_URL - memcache service url
Freezing settings:
There is also checkbox settings freezed?. Once freezed settings can't be changed (in order to change them you need to redeploy the WAR). That's for security reasons to prevent others from changing our settings. If settings are freezed we also can't see confidental data like database user and password. There is one more parameter FREEZE_CONFIGURATION_AT_START which might be useful when you building your own war file. After setting it to true in troia.properties file and building war file, after each deploy troia settings will be freezed by default.
After filling the form, click Submit button. If you've changed job storage to use database you should now see recreate (drop and create) database button below Submit one. By clicking it you create all needed tables depending on selected data model (FULL/KV). To check if everything is working correctly you can punch the following into the browser's url bar:
http://localhost:8080/troia/status
As a result you should get a JSON response saying that everything is OK.
To build WAR file you should enter project main directory execute following command
mvn package
As a result target directory will appear in service directory. It should contain file service.war.
- Q: While loading assigned labels I get 500 error: Internal error: Packet for query is too large.
{
"timestamp": "2013-06-25T21:33:01.779-04:00",
"result": "Internal error: Packet for query is too large (1237720 > 1048576). You can change this value on the server by setting the max_allowed_packet' variable.",
"executionTime": 0.138,
"status": "ERROR"
}A: As exception message says: you need to change proper parameter in your MySQL server settings. It is described here: http://dev.mysql.com/doc/refman/5.5/en/packet-too-large.html . We would recommend to set it to 128MB. If you would like to handle much bigger jobs than probably you should use even higher value.
- Q: Workers quality are negative. What's wrong?
A: It works that way: quality is computed as 1 - (worker_cost / spammer_cost). Workers cost is expected costs that he will generate making an assign. Our baseline is spammer_cost which is using priors and costs to select label which minimizes cost. So in this case spammer is generating lower costs than workers...
Hint: More accurate priors should result in better worker quality estimation. Also for small jobs use BDS algorithm as IDS works well only with high volume of assigns.
- Q: Workers quality are NaN. What's wrong?
A: In most cases it means that algorithm doesn't have enough data to estimate worker quality. It might be the case when worker provided labels only for objects which weren't labeled by anyone else. In such situation we can't tell whether he is good or bad.