-
For more information on how to avoid pop-up ads and still support SkiTalk click HERE.
The Snowcookie...no it's not edible
So, @Philpug, have you tried this out? Impressions?
Yes, exactly. The more data we have the better we are
We continuously improve scoring models with new runs and users, validate them and release as soon they are more accurate. If you are interested about any part of scoring, please do not hesitate to ask All best,
Mati
Mati, I'm very interested how you use the data generated by app users to improve the models. What are you optimizing and how do you score based on the user data?
Would it be a possible feature to offer different progressions, based on disciplines ( GS, slalom, bumps, powder) or even star profiles ( Shiffrin, Hirsher, Neurather,..)?
More concretely: when will you support Android and is there a current or planned feature to tracking the realized turn radius?
Would it be a possible feature to offer different progressions, based on disciplines ( GS, slalom, bumps, powder) or even star profiles ( Shiffrin, Hirsher, Neurather,..)?
More concretely: when will you support Android and is there a current or planned feature to tracking the realized turn radius?
I have a couple of days out with the Snowcookie. For some reason, it is not reading the chest sensor. I plan on taking it out again real soon.
I will keep playing with it. It does make me want to check my alignment.
I will keep playing with it. It does make me want to check my alignment.
What is the meaning of the axis in the radar graph, such as stance, stamina, etc.? What are they actually measuring?
I assume the alignment check concern is due to the symmetry score?
Hi @maxwerks,
Thanks for asking about improving models, as this is very complex and important procedure. We optimize models which describe how we score all values you can see in our app - from those low level to those high level ones. In the simplest form this procedure looks as follows: we have a set of skiers scored/rated by supporting coaches. This is our core reference and a base for our levels, which (as already mentioned) are inspired with PSIA levels. Obviously, each skier and run are completely unique, therefore firstly we look where is your rough approximation fitting best according to just a subset of measures we have (i.e. speed, track difficulty/rating, forces in turns, amount of turns, turn types). When we got that, we compare all measurements for that level and look, if some are outliers to what we would expect to see. If there are significant differences, we validate a particular model to introduce new rules describing behavior that caused them with help of coaches, who rate, if that is a desired/good behavior or not. Otherwise, we just refit/retrain the current model introducing your data to other reference data. This way we collect new reference data and improve models. Hope that have cleared this out a little. If you are interested more in this topic feel free to ask
As long we do not touch our secret sauce I will answer for sure As for the second question - short answer is: obviously it is possible. Long answer is: we are a startup and our main target right now are medium/advanced groomed slopes skiers. When we reach a point where this group is satisfied with what we deliver, we will be able to introduce various other profiles/disciplines in a very short time. I would expect some of them to appear for next season, maybe not in a full option, but some basic, limited functionality, and we will complete theme for the season 2021. And star profiles? Why not - they just need to ride with Snowcookie so we can capture their technique and compare yours with that. This is relatively easy - when we sell approx 500k+ sets they would probably come to us on their own to cooperate and get some insights to their skiing
But being serious - this is mainly a matter of how strong our brand would be and probably money, but still I think we might be able to surprise you shortly in that field Android - it is already happening and we plan to do a first release in February/March 2020, so relatively soon. We are building a new app version in a cross-platform environment. It should be completed for sure for iOS users, as we have done most of the work already, and for Android it would have most of the options, but some might still be missing at that time. A full synchronization for iOS and Android should happen before next season, where new features would be introduced in parallel to both platforms.
Turn radius - to be honest we already did some experiments, but the first approach was not satisfying us and it was abandoned for almost 2 years. Some partners are still pushing this feature strongly and we might return to it this season. If that happens, you could expect some first results in late April, so it would be rather end of this season just to test it. You can subscribe to our newsletter to be informed about upcoming features and releases
Hi @Philpug, hope you have enjoyed skiing with Snowcookies. From what I see you did around 6 runs in total. You need to be aware that our system provides best accuracy after 2/3 days of skiing so it can capture your style. Beside that your turns look really nice, I am only concerned about your Stance score - without it you would be at level 6, having strong base for level 7 (Style will keep you there with 57%). I don`t think there is any hardware problem with chest sensor, it rather looks like a calibration problem. Can I ask quick questions to clarify this? Did you calibrate sensors during on-boarding on a flat surface at home or on the slope? Secondly, when calibrating system before skiing, what was your posture - you stood in a skiers position, leaning a bit to the front, or you felt rather uncomfortable standing straight with chest being aligned with vertical, just like gravity force line? Also it would be nice to see your other stats too
Cheers and more sharp turns with cookies! Waiting for a further reviews
I calabrated it standing in a "skiing stance" I was skeletally stacked on a flat surface. I didn't realize that it would take a couple of days on the snow to accrue the information, no worries I will use it more. (my only concern is I have multiple pairs of skis I need to rotate through and the cookies are on one specific pair).Hi @Philpug, hope you have enjoyed skiing with Snowcookies. From what I see you did around 6 runs in total. You need to be aware that our system provides best accuracy after 2/3 days of skiing so it can capture your style. Beside that your turns look really nice, I am only concerned about your Stance score - without it you would be at level 6, having strong base for level 7 (Style will keep you there with 57%). I don`t think there is any hardware problem with chest sensor, it rather looks like a calibration problem. Can I ask quick questions to clarify this? Did you calibrate sensors during on-boarding on a flat surface at home or on the slope? Secondly, when calibrating system before skiing, what was your posture - you stood in a skiers position, leaning a bit to the front, or you felt rather uncomfortable standing straight with chest being aligned with vertical, just like gravity force line? Also it would be nice to see your other stats too
Cheers and more sharp turns with cookies! Waiting for a further reviews
Looks interesting and would be really cool to see the data.
Hi @Mike King, nice to have you here and you are interested in Snowcookie. We call this chart "Spiderchart" and each axis represents one of 6 main parameters describing your skiing from Snowcookie perspective. Within the graph you can see a level which is PSIA inspired. I will roughly describe all six axis right now and if you are still hungry for more please let me know
This will be probably interesting for our competitors but what the hell - may the best win!Speed describes how fast you can go keeping good control over your skis. Moreover, it is important to be able to keep your speed stable along the track as long as possible, not just continuously speeding up and slowing down.
Engagement tells you if the track was difficult for you or it was relatively easy according to your level. It also describes if you have tried taking turns or not.
Turn IQ measures quality of your turning considering different type of turns (plows, skidding or carving), long/short turn types, initiation timing (inner/outer delay), chest/skis initiation, edge control, edge angles, body balance, skis symmetry, hip angulation, speed
Style describes your turn streaks, how fluent they were, how long they were (in terms of time and number of turns and they type mixture). Additionally, it considers "correct" turns and non-plow turns quality in a run.
Stance measures your body position according to skies with a speed overlay on it. It means, the faster you go the more important your body position is. This lets you have some break at the beginning and at the end of your run, where you gain or loose speed and body position is not so important any more.
Stamina defines how actively you spent your time on the slopes. Did you take a 4h break after a 5min run or just opposite? We know most of you probably will take some break for a meal and a talk and we consider that measuring stamina. Also taking to many breaks on the slope, even they might be very short in time, is not a best idea if you are thinking about crafting this measure.
As I said - this is just a general description of those measures. If you want to know something more we are here - just ask
All best
Mati
@Philpug by "skiing stance" you meant such position, as presented in app while calibrating?
You should not lean forward, because we need to capture gravity reference and align virtually chest sensor to the gravity
If you were leaning forward you have cheated sensor, so when you are leaning exactly like this over the skis during run, it thinks you stand straight and penalize you for that, what probably we could observe on the provided screens. Next time, when calibrating chest please try to stand straight like on the picture, trying to even stand with a chest a bit backward. Then you would cheat sensor in a right way - your straight position would be taken as already correct skiing position with leaning slightly forward. I know it might be inconvenient to do it like this but this is critical for correct body position analysis.Waiting for new results after this hints
Hope it will help your Stance score
Hi @Ron, what kind of data you mean? The raw ones collected by sensors or post processed by our algorithms? If you wanna talk science contact me directly by PM
that was simply a general comment about using the output information
Thanks, but I'm looking for something a bit more specific. See below.
So is this simply a measure of the variance of speed, or is there some other measure that is incorporated?
This doesn't really describe what the measure is. Is this an estimate of carving or edge angle or what? It might be interesting if there was either, e.g. a measure of "slippage" of the ski across it base (edge engagement) or a measure of edge angle and the rate of change of it. But right now, I don't understand the specifics of what your measure is or how I'd use it.
As I understand it, you have three sensors; one on each ski and one on the chest. How do you determine hip angulation versus knee angulation? I presume body balance is fore/aft pressure control, but how do you determine that given the variation in anatomy of weight distribution/femur and tibia length? And is the raw data, or sub-aggregations of your overall metric available to use in coaching? I'm not sure I'd find a "Turn IQ" aggregate metric to be useful, but might want to look at fore/aft or foot to foot pressure by phase of the turn.
How do you measure "fluency" of a turn? Do you have a way to measure pressure? The location of maximum pressure in the turn might be pretty useful as well as measuring foot to foot pressure. And I presume "correct" is from training a model to some metrics collected by some group of skiers. What parameters are you training the model to?
There are a lot of issues in measuring stance. Dorsiflexion, knee flexion, and hip flexion being one set, along with rotational alignment between the shoulders, pelvis, knees, and feet. What precisely is being measured?
To translate time moving (presumably on snow) versus time not moving plus lift rides into an index still requires some sort of scoring. This metric is less interesting to me, but it still warrants some sort of description of how it is constructed.
Mike
Hi @Mike King, nice to have you here and you are interested in Snowcookie. We call this chart "Spiderchart" and each axis represents one of 6 main parameters describing your skiing from Snowcookie perspective. Within the graph you can see a level which is PSIA inspired. I will roughly describe all six axis right now and if you are still hungry for more please let me know
Speed describes how fast you can go keeping good control over your skis. Moreover, it is important to be able to keep your speed stable along the track as long as possible, not just continuously speeding up and slowing down.
So is this simply a measure of the variance of speed, or is there some other measure that is incorporated?
This doesn't really describe what the measure is. Is this an estimate of carving or edge angle or what? It might be interesting if there was either, e.g. a measure of "slippage" of the ski across it base (edge engagement) or a measure of edge angle and the rate of change of it. But right now, I don't understand the specifics of what your measure is or how I'd use it.
Turn IQ measures quality of your turning considering different type of turns (plows, skidding or carving), long/short turn types, initiation timing (inner/outer delay), chest/skis initiation, edge control, edge angles, body balance, skis symmetry, hip angulation, speed
As I understand it, you have three sensors; one on each ski and one on the chest. How do you determine hip angulation versus knee angulation? I presume body balance is fore/aft pressure control, but how do you determine that given the variation in anatomy of weight distribution/femur and tibia length? And is the raw data, or sub-aggregations of your overall metric available to use in coaching? I'm not sure I'd find a "Turn IQ" aggregate metric to be useful, but might want to look at fore/aft or foot to foot pressure by phase of the turn.
How do you measure "fluency" of a turn? Do you have a way to measure pressure? The location of maximum pressure in the turn might be pretty useful as well as measuring foot to foot pressure. And I presume "correct" is from training a model to some metrics collected by some group of skiers. What parameters are you training the model to?
There are a lot of issues in measuring stance. Dorsiflexion, knee flexion, and hip flexion being one set, along with rotational alignment between the shoulders, pelvis, knees, and feet. What precisely is being measured?
To translate time moving (presumably on snow) versus time not moving plus lift rides into an index still requires some sort of scoring. This metric is less interesting to me, but it still warrants some sort of description of how it is constructed.
Mike
In the first place sorry for letting you without a high quality answer - my bad, thought that provided level of details should be enough in that case. I will try to elaborate more details below
I appreciate you are asking for those details - we are quite similar in that field, I must also understand all math and algorithms idea behind before I apply it This way we might even engage our competitors to introduce similar measures or present and explain their approach to the same problems Speed:
It is not that simple. We correlate speed with skis vibrations to asses if you have still good control of them. Is it enough right now?
If not I will just mention that to analyze vibrations you can apply various signal analysis tools, i.e. frequency analysis with Fourier, Wavelets or DTW to mention in the first place. This is where we enter our secret sauce, where we have developed custom mixture of those and some other ones Engagement:
You are right - after a second read a person who has never used the system could not get what it measures. We have a distribution of slope ratings according to different skiing levels. Knowing at which level you are we can say if that slope was hard or not. Additionally, knowing your style (long/short turns) we have a similar model for turns pace. A mixture of them allows us to asses your overall engagement - that is if this run was a piece of cake for you (far below your true capabilities) or it was challenging for you, pushing your skills to an edge. It has nothing to do here with quality or dynamics of your turns, it`s just a measure of your altitude to that run - you wanted to ski harder or just relax.
Turn IQ:
For now we have two solutions - simple and complex one. Simple, seen in application right now, assumes that hip angulation is based on angle between chest and skis edge angle. Second one, at experimental stage, requires more high quality reference data as we build up an kinematic model of skier driven with data from our sensors. Unfortunately, amount of work to do with other features, which seems to be more interesting for users, forces us to move slower with that than we would like to :/ But this is definitely the direction we want to go with this task in the nearest future.
As for the body balance question: currently we present only an average statistics of a complete turn for body balance. Having said that, 48L/52R means that on average your body balance was 48% on the left ski and 52% on the right ski. We neglect the specificity in construction of individual bodies and their asymmetries in mass distribution. This is possible to take into account, but would require lots of advanced measurements before you would be able to start using the system. For some users this might be to difficult to perform or even impossible, as this is really pro approach requiring high precision and sophisticated knowledge. 3D scanner might be a solution here, but still - how many of us have access to such devices and would be able to provide required processed data as input to a system similar to Snowcookie? This is definitely a professional, super high performance approach which coaches have even not dreamed about I think.
But getting back to the question. It is possible to provide body balance in time (turn phases) instead of an aggregate for the whole turn and if enough users would ask for this we would have to think how and were present it in app. We already have this data, we are just not presenting it directly.
I did not get what you meant if our raw data or sub-aggregations are available to use in coaching. Could you explain this to me?
In general Turn IQ is a complete score of your entire turning. Looking at it you should know if turning has improved or not. Getting deeper you can observe it`s "ingredients" - edge angles, edge control, body balance and many others. This is just an aggregate and simplification. Is it useful for you? I would still say yes, but I have to agree that if you are working on a particular parameter of your turning it forces you right now to go through an additional level of abstraction that might cover what you were really looking for. Solution for that might be dedicated training program where you could define what you want to improve and there observe only those parameters. Something worth considering from our perspective for advanced and aware users.
Style:
The fluency is calculated only for consecutive turns aggregated together as a streaks. For each streak we consider turn types, their quality, speed fluctuations between turns and transition times in between. The better the turns, the smaller speed changes, the smoother transitions and motion dynamics, the stable edge angles the better fluency you have. Hope this is enough as I am afraid I can not continue to much further as we will be touching some critical details, which we do not want to expose right now. But you can always ask, maybe I could explain it in a different way though?
As for the question about pressure - we do not measure pressure directly as Carv does, but we can do it indirectly, based on ski vibration analysis, as mentioned in Speed calculations and used for body balance in turns description too.
About turn correctness - you are right. There are several factors to assume turn to be correct. I will focus on carving turns only, as this is what we all want to perform mostly. We use turn initiation timings data (chest/skis, inner/outer ski delays), body balance, and some other turn quality parameters. Some aspects like transition time or turn initiation were already discussed here in this thread with other users.
Stance:
The answer here is exactly the same as for hip angulation as this is almost the same model, but focused on different angles. We care here mainly for leaning forward/backward but as meantioned already we put additional speed layer on top of that as a weighting factor.
Stamina:
To be honest I don`t know how to explain it even simpler. For every run you have some sort of time buffer for pauses/breaks. When you use all of it we start to penalize you for additional pause time. If some time has left you will get 100%. Is it clear right now? Basically it is all about spending time on skiing actively as much as possible.
Hope I have now fulfilled your demands about different system details
If not, I would be happy to continue where it is required and possible.Mati
Thanks Mati. This is helpful. I'm wondering about the potential of these devices to serve as a coaching tool, and at what level, and for what tasks, they might be useful. I do hope @Philpug conducts a bunch more trials of it. Who knows, I might break down and buy one for myself to see what it can do.
@Mike King I think it might be interesting for coaches as well, i.e. to present their learners how did they progress during the lessons. Some of UK coaches are already testing it for that reason to additionally prove their authority and influence of their lessons on skiers skills. They simply do some test runs at the beginning of lesson(s) and then at the end and present how skiers have improved according to Snowcookie measures. Just a pilot, not a big deal, but still - might be interesting for other coaches too
This is also how we get reference data and how we cooperate with coaches 