v1.2.5

initialize triplets from random state

v1.2.4

MAINT: comment out user ID to help with MTurk (#220)

v1.2.3: Switch away from deprecated Jessie backports repositories

Debian recently removed the jessie-backports repo; as a result, our Docker containers wouldn't build. This release switches to stretch-backports.

v1.2.2: BUG: install mongodb and launch next

This prevented NEXT from launching.

1.2.1: Triplet validation filenames (#216)

* ENH: accept filenames for validation sampling

* ENH: Validation sampling takes string as input

* MAINT: better error handling

Boston Cream

Features

• User ID is now shown for all queries (#214)
• Database layer entirely rewritten, removing unused caching system (#182)
• Experiments can be retired for organizational purposes (#199)
• Added pythonic pop method to collections and database api (#174)
• Redis Memory is now namespaced (e.g, butler.algorithms.memory instead of butler.memory) (#176)
• Targets can be uploaded as zipped JSON (c5899f1...67bad37)
• Dashboard styling and markup have been cleaned up (#203)
• Added targets endpoint (#210)
• Dueling bandits: integrated KLUCB algorithm (#209)
• 404 page added for experiment dashboards which couldn't be found (#203)
• Base docker image off of python:2, install less stuff during container builds (#195, f82eaf1)
• Arbitrary keys can be added to all targets when using /assisstant/init/experiment (38173c8)
• Dueling bandits can now effectively display video controls (4c3b7db)

Fixes

• The worker hostname cache will no longer go stale, fixing issues with Celery workers not responding (#198)
• Fix faulty redirect after db restore if a SITE_KEY is not set (2142bb3)
• Allow HTML in debrief/instructions (e76b8f8)
• Fix issue with butler.experiment mutation in initExp() causing mysterious crashes (#183)
• Fix issue with butler.algorithms.memory namespacing (#197)
• Pin matplotlib==1.5.3 to fix a mpld3 rendering glitch (1bf15e8)
• Targets are now orded (9e82f9b)

Maintenance

• Properly sort the experiment list on the dashboard by descending launch time (#194)
• Participant data .zip files are now compressed (#202)
• Pin dependencies in requirements.txt (#205)
• Set GIT_HASH when running locally (#190)
• Cleaned up local/launch.py (#171)
• Add Tests app as a base for integrated tests (#183)
• Fix unicode issues with HTML meta tags in dashboard/templates/outline.html (9d100ac)
• butler.job(..., ignore_result=False) will now return the result (2dc302f)
• Some of the annoying debug noise when launching experiments has been removed (80fc195)
• Add an .editorconfig file to enforce indentation and encoding in supported editors (d956b4a)
• Link the examples folder on GitHub on the NEXT homepage (d7a727e)
• Improve target unpacker to discard empty lines in the CSV (f6cbea9)

Test

• Make sure can restore experiments. Download from NEXT 1.1, upload to NEXT 1.2.
• Speed test NEXT 1.1 and 1.2, make sure similar results are obtained (timing results below)

Timings

We have verified that this release has improvements over the last release. I ran the same tests and only changed NEXT_BACKEND_GLOBAL_HOST when running the tests between machines running 1.1.1 and 1.2. For triplets, I had 5 clients and 20 arms. For cardinal, I had 50 clients and 100 arms.

Name	1.1.1	1.2
Cardinal algorithm
Cardinal client
Triplets alg
Triplets client

(the images link to larger images; click to view)

Highlights:

• cyclic getQuery times for triplets on 1.1.1. These were a triangular wave. The database refactor helped with this.
• both 1.1.1 and 1.2 have the same getQuery time (0.6) seconds on triplets after ~200 API calls.
• the client-side timings show the network delay (or time waiting for the server) is much reduced on 1.2.
• The daemonProcess for triplets on 1.2 takes longer than 1.1.1. I'm not sure how the response delays play into this.

GUI launch bug fix (text targets)

MAINT: have main link to /home

not /dasboard/experiment_list

Glazed

Features

In this release we provide

• 752d198: a collection of useful NEXT links at /home
• #154: a new API endpoint to initialize experiments and an accompanying interface
• #153: fast in-memory RAM database through Redis (implemented in butler.memory)
• 7c6561b: custom alg selection
• #155: uniform names (myApp) for all the apps (meaning the app name only appears in one place)
• 30f4af0: allowing custom participant IDs on the query page

...and a collection of bug fixes, maintenance and cleaning

Checks

• py.test completes successfully
• checked downloading CSV does not succeed when myApp.format_responses does not exist
• CSV downloading tested with unicode, missing responses, error handling

1.1-alpha.3

modified dashboard to relative links

v1.1-alpha.2

open in new tab