| More progress on Full Data mode. |
|
More...
|
over 13 years ago
|
| Early changes to incorporate a "full data mode" to track exact results when you know the number of unique item-ids is tractable. |
|
More...
|
over 13 years ago
|
| Change settings for the example app. |
|
More...
|
over 13 years ago
|
| Added the output of the first big popservd test (didn't go so well, but there is decent debugging output). |
|
More...
|
over 13 years ago
|
| Added tracking of actual results to the example which uses the LyricWiki API (for comparing these to the results which popservd gives). |
|
More...
|
over 13 years ago
|
| Added a larger varnish log-file which should make the testing give better results. |
|
More...
|
over 13 years ago
|
| Added a very simplistic example which uses varnishlog output from a small amount of time to track the most popular LyricWiki requests. |
|
More...
|
over 13 years ago
|
| Made the top items be ranked by hits when actually getting the results (will still cause late-bloomers to be underreported). |
|
More...
|
over 13 years ago
|
| Changed PopularityService::DataStore to get ranked results by score for choosing whether a hit should become a tracked Top Result. Still sorting the final list by number of known hits, so the Late Bloomer unit test still fails (not sure it will ever pass). |
|
More...
|
over 13 years ago
|
| Added some debugging to help figure out why late-bloomers don't get added to the top results. |
|
More...
|
over 13 years ago
|
| Added "-port" command-line parameter to popserv_tests.pl as well as the "-debug" switch to put it into debug-mode (which will run a "dump" command after each command sent by a unit-test... this is useful to run after a test has failed to get really detailed output about what was happening in the server when something went wrong). |
|
More...
|
over 13 years ago
|
| Made the error message for no-connection more helpful (show the attempted host / port num ). |
|
More...
|
over 13 years ago
|
| Made the error-output more helpful for tests where the expected Top Results and actual Top Results differ. |
|
More...
|
over 13 years ago
|
| Added logging to the unit-testing suite (logs output of every test in one file, output of failed tests in another, and keeps a history of the results - but not output - of all runs in a third). |
|
More...
|
over 13 years ago
|
| Made the unit-testing script use a specific port (it will try to spawn a server on that port if running on unix - that code is not tested yet though). This will allow running of the unit-tests on a server which is also running a production server. |
|
More...
|
over 13 years ago
|
| Fixed the fact that displayList() was not limiting the number of results to list-size. The algorithm itself (perhaps just the number of buckets/hashes) still seems very broken according to the Test Late Bloomer test. |
|
More...
|
over 13 years ago
|
| Updated unit-test runner to handle comments and blank lines (for readability as the tests get more complex). |
|
More...
|
over 13 years ago
|
| Made unit-tester output the names of all failed tests. |
|
More...
|
over 13 years ago
|
| Fixed the bug which allowed a carriage-return to sneak into the last parameter (and therefore, often the config-id). |
|
More...
|
over 13 years ago
|
| More progress on the file-based unit-testing framework. Not tested yet. |
|
More...
|
over 13 years ago
|
| Ready to start testing the actual functionality. |
|
More...
|
over 13 years ago
|
| Added sorting of top-results (currently by number of known hits). |
|
More...
|
over 13 years ago
|
| Filled in a little more of the guts of PopularityService::DataStore. |
|
More...
|
over 13 years ago
|
| Added some code to record the top results. Until the algorithm is figured out for ranking Top Results against each other, new high-scoring items don't replace existing items. The first n results fill the top n results list and then the items never fall out of that list even if more popular items arize. This is bad. A ranking function needs to be determined. |
|
More...
|
over 13 years ago
|
| Pseudocode for tracking the top-results. Still some fogginess on what the right algorithm will be. |
|
More...
|
over 13 years ago
|
| Added the (currently undocumented) "dump" command which lets the client get a dump of the current state of the data for the configuration that it sends. This is helpful in debugging, but I'm not sure if it should remain enabled always (can't think of a reason it would be damaging though). |
|
More...
|
over 13 years ago
|
| Stored the results in the appropriate hash buckets. |
|
More...
|
over 13 years ago
|
| Recreating the repo from scratch (previous repo from last night was rolled back because the server crashed & restore was done from a backup). The current state of things is that there is a client (popserv_client.pl) and a server (popservd.pl) which can communicate. Most of the functionality is not implemented yet. Current place to work on is PopularityService::DataStore::recordHit(). For info about what this system does, refer to README.txt |
|
More...
|
over 13 years ago
|
