- Stats for 7.2 are out. I finally managed to fix the parallel processing error (it was multi-threading, not shared memory issue). Debugging parallel code is PIA.
- Stats for 7.1 are finally out. After adding the tank-pages the site build time has grown 10x and I have been looking for parallel processing to speed up the work. But I have ran into weird issues likely related to shared memory limits causing the parallel build script to fail.
- New tank-specific pages with lot of detailed information about each tank in comparison to the same tier/type tanks.
- Hit Rate and average kills charts added to Tanks, Tanks per tier, Tanks of Type and Tank
- Histograms are now pretty & colored and have mean/median values.
- Career stats are now calculated based on true career stats including players who did not play and tanks that were not played during the update.
- New graphs:
- New fancy parallel build system that cut down the built time 70%. It took insane amount of debugging.
- Reworking the code the create reasonable axis breaks and labels. This little “fix” took lot of time, enjoy :-D
- Updated home page layout.
- Images are now in SVG format instead of PNG. In case there are browser compatibility issues, please let me know.
- Stats added for Update 7.0.
- New pages for each tank type per tier have been added. These pages list all the tanks in the tier for each tank type.
- Stats added for update 6.10
- Also fixed a bug in the code (function params in wrong order..) that likely caused too many tank stats being filtered out.
Stats filtering unified to eliminate stock-grind tanks’ impact on tank performance stats (Averate WR, Relative WR, Average damage, etc.). Earlier new tanks got impacted by the stock-grind battles being taken into account. The number of required battle is a compromise based on average tank research XP divided by players’ average XP at the tier and divided by 1.5^2. This can be intepreted as playing with 50% WR and only the first 2x XP victories plus XP boosters. Making the filter too aggressive seriously limits the available data, therefore the compromise.
Required number of battles per tier
Relative WR formula revised to do the comparison to average of all the same tier tanks, including the tank itself. This seems to provide more sensible results than the earlier using other same tier tanks as the comparison point. Honestly speaking, I do not know why. The first algorithm seemed clever, but the detailed graphs showed it failed somehow.
The first Relative WR incarnation also had an error that while it filtered out stock-grind tanks, it included those in the calculation of the comparison WR (i.e. average WR of other same tier tanks). This caused RElative WR to be higher for all the tanks than it actually was.
Stats since Update 6.0 added. Graphs are also now based on plot templates to unify look’n’ feel.