First, I really enjoy MATLAB. The help pages are professionally maintained (as opposed to many open source contributors). They all follow the same format and are well written making learning easier. In addition, the error messages are really instructive. Unfortunately, the license is super expensive so it seems that very few people in the "real world" use it. I would suggest that if you are attempting to make yourself more marketable to choose another option. In addition, it is also a little harder to make good visualizations but the more recent versions have made real improvements and the parallel toolbox is so much better than doing stuff in openMPI and openMP.
R has lots of nice statistical packages and good visualization packages. R Studio is a really good IDE that makes package installation and loading/reading files really easy. The help pages vary depending on the author of the packages and can be difficult. The error messages suck, sometimes it seems like the error messages are purposely obtuse just to mess with you. It is slow, but it can be linked to C, C++, or FORTRAN for computationally intensive tasks. I know people who did this and said it was not too bad to implement.
Python seems to be all the rage (most marketable) but I have the least experience in it. Following that, has anyone tried Juypter Labs? It is supposed to give a Juypter Notebook an experience more like an R studio or spyder IDE but still in the form of a notebook. I have played with it a little but it seems a little buggy and a memory hog. Has anyone else had the same experience?
I agree with everything you said in the second paragraph, it is spot on. Failing to control for initial conditions like formations and coaches known tendencies is problematic. You are also correct that this dataset is not large enough to integrate those things into your model even if you wanted to, and one should be skeptical of this data. It is very likely that running breaks tendency and as soon as coaches start "running enough" the success rates of running will drop.
That does not make the data useless, it just requires responsible consumption of the data. I do not think it would be a mistake to start running more on 2 point conversions until the defenses do more to stop it.
Finally, I do disagree with your issue on the lack of controlled randomized experiments. Sure, it would be nice to have that type of data but that is not realistic. Datasets collected from real world outcomes are employed frequently and can have great outcomes. It just requires more careful thought put into the model and conclusions from the model.
I am going to suggest something contrary to most other suggestions. Don't try to incorporate everything into your initial model. You are going to miss things and there are going to be some counterintuitive forces acting in the model. Start with a simple model that is easy to build on and let the data "tell you" where to go next. I also suggest monte-carlo simulations as a more intuitive way to test different ideas on small data sets.
(Full disclosure-most of the modeling I do is Bayesian so I am bias towards monte-carlo)
Although neighborhood can be a good way to estimate the make-up of a classroom, it is not always a good indicator. I have taught an ELD (English Language Development) math class in a very wealthy district where the majority of my class was on free or reduced lunch. Furthermore, since people self-select for charters, a charter will often become known for working well for a particular population. For example, if a charter is known for having a good ELD program, they may attract more immigrant families from out of the area.
So I think I have seen this posted before, but graduate students get paid for their services. Right now I get fee remission and an additional salary as a graduate TA. We have a union (we are represented by the UAW). We can also tutor/intern/whatever on the side and accept money for it. Between teaching and research, we probably bring in more than our "salary" (which on its own is laughable, I would probably be better off in the fast food industry), but it is understood we are also receiving something of great value. Furthermore, different schools offer different amounts of money. And the "haves" for the most part get the best students. Finally, pay is loosely attached with performance. For example, a master’s student is less likely to get a TAship. Although they continue their fee remission, they don't get a salary for that quarter. We also went on strike this past year. Some of the biggest issues were class sizes and work load. I think people are thinking too much about how "Charles Woodson" would change the landscape of negotiations between the "union" and the NCAA. Most of the members would be people who are not going to get drafted and I think a union could really benefit them. There are a lot of imperfect things in this model, but I don't think it would turn this into a "semi-pro" system. (And knowing what it takes to survive on my "salary", I would still be able to view the athletes as college students.)
So after watching the game, I have a quick question that maybe someone more knowledgable about the game can answer (or I guess someone less knowledgable about the game could answer but I am not sure that would be helpful). Magnuson had a decent game at tackle. If we need to run that formation with Lewan and Schofield on the same side, why not put Magnuson on the other side? Since the bubble is not a possibility because it is well known by some coaches to be the dumbest play ever, at least with mags you could throw a quick slant or something without getting DG destroyed.
So I know this is not the correct string to post this, but I saw some info on Turner on this site. http://gobluemichiganwolverine.blogspot.com/
I really want some feed back, does this mean he is good to go. Has anybody verified this.
I am going to be in New York City for the weekend and need to find a good bar to watch the game. It would be preferable to watch close to where I am staying (Midtown- 28th and Madison), but I would also like to avoid watching the game with a bunch of investment banking asses.
Recent Comments
First, I really enjoy MATLAB. The help pages are professionally maintained (as opposed to many open source contributors). They all follow the same format and are well written making learning easier. In addition, the error messages are really instructive. Unfortunately, the license is super expensive so it seems that very few people in the "real world" use it. I would suggest that if you are attempting to make yourself more marketable to choose another option. In addition, it is also a little harder to make good visualizations but the more recent versions have made real improvements and the parallel toolbox is so much better than doing stuff in openMPI and openMP.
R has lots of nice statistical packages and good visualization packages. R Studio is a really good IDE that makes package installation and loading/reading files really easy. The help pages vary depending on the author of the packages and can be difficult. The error messages suck, sometimes it seems like the error messages are purposely obtuse just to mess with you. It is slow, but it can be linked to C, C++, or FORTRAN for computationally intensive tasks. I know people who did this and said it was not too bad to implement.
Python seems to be all the rage (most marketable) but I have the least experience in it. Following that, has anyone tried Juypter Labs? It is supposed to give a Juypter Notebook an experience more like an R studio or spyder IDE but still in the form of a notebook. I have played with it a little but it seems a little buggy and a memory hog. Has anyone else had the same experience?
I agree with everything you said in the second paragraph, it is spot on. Failing to control for initial conditions like formations and coaches known tendencies is problematic. You are also correct that this dataset is not large enough to integrate those things into your model even if you wanted to, and one should be skeptical of this data. It is very likely that running breaks tendency and as soon as coaches start "running enough" the success rates of running will drop.
That does not make the data useless, it just requires responsible consumption of the data. I do not think it would be a mistake to start running more on 2 point conversions until the defenses do more to stop it.
Finally, I do disagree with your issue on the lack of controlled randomized experiments. Sure, it would be nice to have that type of data but that is not realistic. Datasets collected from real world outcomes are employed frequently and can have great outcomes. It just requires more careful thought put into the model and conclusions from the model.
I am going to suggest something contrary to most other suggestions. Don't try to incorporate everything into your initial model. You are going to miss things and there are going to be some counterintuitive forces acting in the model. Start with a simple model that is easy to build on and let the data "tell you" where to go next. I also suggest monte-carlo simulations as a more intuitive way to test different ideas on small data sets.
(Full disclosure-most of the modeling I do is Bayesian so I am bias towards monte-carlo)
Although neighborhood can be a good way to estimate the make-up of a classroom, it is not always a good indicator. I have taught an ELD (English Language Development) math class in a very wealthy district where the majority of my class was on free or reduced lunch. Furthermore, since people self-select for charters, a charter will often become known for working well for a particular population. For example, if a charter is known for having a good ELD program, they may attract more immigrant families from out of the area.
http://theozone.net/Ohio-State/Football/2015/News/Michigan-Monday-The-I…
So I think I have seen this posted before, but graduate students get paid for their services. Right now I get fee remission and an additional salary as a graduate TA. We have a union (we are represented by the UAW). We can also tutor/intern/whatever on the side and accept money for it. Between teaching and research, we probably bring in more than our "salary" (which on its own is laughable, I would probably be better off in the fast food industry), but it is understood we are also receiving something of great value. Furthermore, different schools offer different amounts of money. And the "haves" for the most part get the best students. Finally, pay is loosely attached with performance. For example, a master’s student is less likely to get a TAship. Although they continue their fee remission, they don't get a salary for that quarter. We also went on strike this past year. Some of the biggest issues were class sizes and work load. I think people are thinking too much about how "Charles Woodson" would change the landscape of negotiations between the "union" and the NCAA. Most of the members would be people who are not going to get drafted and I think a union could really benefit them. There are a lot of imperfect things in this model, but I don't think it would turn this into a "semi-pro" system. (And knowing what it takes to survive on my "salary", I would still be able to view the athletes as college students.)
So after watching the game, I have a quick question that maybe someone more knowledgable about the game can answer (or I guess someone less knowledgable about the game could answer but I am not sure that would be helpful). Magnuson had a decent game at tackle. If we need to run that formation with Lewan and Schofield on the same side, why not put Magnuson on the other side? Since the bubble is not a possibility because it is well known by some coaches to be the dumbest play ever, at least with mags you could throw a quick slant or something without getting DG destroyed.
There is a rumor that there have been offers extended to Teddy Ruxpin and H.R. Pufnstuf for positions as offensive and defensive coordinators.