Pre-Game Talk: Tomorrow is Saturday Night, and another BOA shall commence 8PM MT on HNIC/CHED

[DrDrai]

How do the inputs predict success based on previous seasons? Can you take the first 40 games of a season to predict the next 40? How successful was it in predicting the top scorers/teams?

Without some training data to validate your model’s predictions, it is f***ing useless.

It’s giving a score based off of the current stats, it’s more of a running total at the moment.

Also nothing is based on previous seasons.

 

[yukoner88]

Okay, I missed the joke although still not all that funny but that's just an opinion.

If you dont find it funny, that's fine, Im not looking to offend anyway, and I respect that we all have different senses of humor. I dont mind if someone speaks up about a joke they dont like either, cause it helps me see where everyone's line is for what they find funny.

I just wanted to stress that I enjoyed what the OIlers did for the child, so none of the poke(s) were going in that direction. And if the Flames do something similar, I wouldn't joke about it, I never make fun of stuff like that, cause I truly respect/enjoy it when people go out of their way to make a child smile like that

 

[Nostradumbass]

It’s giving a score based off of the current stats, it’s more of a running total at the moment.

Also nothing is based on previous seasons.

You have no idea what you’re doing when it comes to analytics. 17 variables is too much, there’s too high a degree of correlation between them. How is each variable weighted? What “success” benchmark do you use to deem the weighting of each one? Do you use multivariate analysis to determine weighting? What’s the correlation to success in your MVA model?

Once you’ve developed a model, you need to use training data to validate it. If the model doesn’t work, you need to either tweak it or ditch it.

If you’re not doing any of that, you’re just telling us “this is who I think the best players are, and here are some made-up numbers to support it.”

I do a lot of work in analytics and unfortunately see too many people who use the latter approach.

 

[DrDrai]

You have no idea what you’re doing when it comes to analytics. 17 variables is too much, there’s too high a degree of correlation between them. How is each variable weighted? What “success” benchmark do you use to deem the weighting of each one? Do you use multivariate analysis to determine weighting? What’s the correlation to success in your MVA model?

Once you’ve developed a model, you need to use training data to validate it. If the model doesn’t work, you need to either tweak it or ditch it.

If you’re not doing any of that, you’re just telling us “this is who I think the best players are, and here are some made-up numbers to support it.”

I do a lot of work in analytics and unfortunately see too many people who use the latter approach.