Jump to content


PC Member
  • Posts

  • Joined

  • Last visited



Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. Alternatively, just use the Xoris and you can do this on any frame I think I did it on mostly Kong, after realizing that he was mission-overall just marginally slower to a spin-to-win Revenant, but more engaging than just pressing 4 and waiting for like 30 seconds.
  2. She is in Railjack - like Valkyr. Permanentely unvaulted. You can find the relics in caches 래일잭 에서 닉스는 나올수있어요 - 개이치 속에
  3. Wu Kong Umbra. 2 clones ftw. One of them is live already, so it‘s only half the work (pun intended)
  4. I’m doing Zephyr for a few defenses and rest Kong. Kong with an efficiency loadout and autohack imbued for rescues/spys, and one with mostly corrosive weapons and smite infusion replacing his 4. Clone uniquely on Proboscis Cernos and just get the job done fast
  5. I‘m still thinking back of Tabula Rasa these days. It was a shabby and way too early death though
  6. Indeed. I actually expected to be more like patching to 29.10.12+ today. Totally on your page. For the release cycles, I don‘t think they‘ll disappear. However, they trigger feedback posts (especially on the update forums), so I could imagine that with categorized ‚comments on the warframe board‘-data, one could find a relation from measurable posts to players like: Many posts -> many additional players active. Many negative posts -> less players in a day, week or the like. However, I‘m not willing crawling the entire boards into a database and thinking about good regexp functions categorizing them, that‘d be too much work for me - I am here to find distraction from such XD I actually quite enjoyed the discussion with you, Gravaarg, and iPhatos. It was way better than spamming Xoris on SP :) As I am busy the coming week and tomorrow, so I‘m soon getting ingame. I wish you guys good findings and May your skanas always be sharp ;)
  7. Pretty cool on that. You know, biggest problem on boards is that we use avatars and I and many others tend to stomp in any discussion kick-in-the-door style XD And then I, and I suppose others too, are way down to our daily practics. I’m pretty sure I’m not using the terms as I did when I learnt their proper use, as in daily life I just use them to get things done. And I alteady learnt them differently in different countries. E.g. where I originally come from nobody used the word ‚model‘; in my university everything on two legs (pun intended) was a model and I am still mostly sticking to that usage Big issue on the original topic seems to me that we did not yet look outside for playerbase declining / increasing reasons. I could imagine that employment situation has a huge impact at the very moment and impacts the steam numbers negatively. I‘d rather not play in many situations (like yesterday e.g.) I also have to admit that in the beginning I was thinking to ‚stocky‘ as I was treating the steam average player numbers like a performance chart, just because that‘s what I look mostly at. And then even ‚non-growth‘, or slight player decline in month X year-to-year (I was looking mostly at the monthly figures) implyed to me something different than just players that take a break. So, I found the lookouts of average steam players not looking well. It looks healthy enough to support the game many years - and we are in a special situation with Corona around anyway - and we cannot model for that as most players have different situations around the globe. Generally speaking, it‘d be cool getting daily player data, along their country of origin or IP based regionality, and average login time. With that one could nicely play :)
  8. The trendline is a model in itself trying to explain the playerbase over time. First of all, I'd say you have the motivation to become excellent at this, so let me give you some input going further with your data model: If you have a model explaining part of the data, here 69%, you can continue on that model by refining it until you reach something as close as possible to 100%. I learnt that the old way without computers, because a long time ago my thesis father was a calculator and pen nerdo, so I personally still do modelling often by making graphs and looking for similarities in other data I find relevant (and finding new hypothesis). It is actually quite easy in principle: Assume you have data that goes 1,2,3,4,5 and found a time trend of x - If you divide your data by your findings you'll get new data of 1, 1, 1, 1, 1 with no further variance. The model (time is x) perfectly describes your data, as we can no longer find any variables to better describe the data. Welcome to the visual world of modelling in 2D. Time alone can make for strong models, I see this often with historical data over stocks, where sometimes inflation improves the correlation and voila done is a two factor model for predicting the future of a company's stock growth. It won't help in day trading though and it won't even persuade myself buying only on these two factors. With less in line data, that accounts in your case for 31% of your made data, you now get a new graph that you could refine another line netting a quadratic time trend (to as many power to your x as you may find). Once there is no more explanation possible with time (r2 is not getting better the more lines or whatever you throw at it), you'll need to throw in other data that resembles what you have left over and has some kind of link to it that is still unknown, like for example amplified or reduced marketing spending lagged by a week/a month or whatever (following the hypothesis that marketing spending increase sales e.g. player base here). The problem is finding and identifying relevant data. I find graphs very useful for that, as did my above mentioned nerdo, as one can quickly identify similarities, then test against it and see whether it has good correlation or not. Then you get the relation out as a function and now end up a graph having time and marketing spending as variables. Original data divided by these two are then hopefully all parallel to x-axis (or very rarely y-axis), else the visual hunt continues looking for more relevant data. After all that one could have a model that explains that Warframe will grow by 600 players just by one month passing (people talk about it or whatever) + 3 players for each USD in marketing spending of prior month - 2 players for each negative forum post of prior week and a relations to substitute games that are just popular at the very moment. Do this a few hundred times and you will see trends in data graphs like I do. That's experience that some people deem magic, but at times experience is wrong, so I usually test against it, but not always, as here :) Off topic: My last real modelling work used 2 DVDs of original data, found an relation we didn't expect at the time starting (because they looked too damn close and we could argue a logical dependency) and it saved some money on hedging exchange currencies on Korean won to USD :)
  9. That was some longer reading :) I think it only missed that r2 in % is a human understandable indicator on how much your model can explain the data and I could almost use your comments as a statistics tutorial XD The model of player data over time has another big fault; it litterally says: „Warframe is growing since age 0 because it gets older“. So the hypothesis to that graph would habe been „was Warframe growing after launch?“. The linear model is a bad idea as is starting from „zero“, because any successful game will have more players than at their start in any charts. Ask a bad question - get a bad model ;) I think that dataset of average players/day, patch dates, advertisment budget, and feedback posts could lead to a meaningful partly recursive function. And I am certain it would proof your the point of ‚recently declining steam player base‘. But that is (should be) like apparent from the steam data on sight
  10. It‘s almost like I learnt in my childhood: An hotfix a day keeps the bugs at bay
  • Create New...