Maybe not the best dataset for the correlation matrix, but this is a must for the initial analysis of any dataset. Getting basic info, descriptive statistics and distribution of each column and correlation matrix.

great sum up

yup, they tend to risk less, so you go for the central path, good notice

Thumbs up for this analysis. It's not something that's just sitting in a column as an integer, so needs some additional work, great job :)

And a great conclusion. It's also important to have this kind of insight because we can maybe exclude it from some specific analysis as an outlier set.

There's a fantastic example on how much you can achieve only with a single column, that majority of people would just neglect and disregard. If you're acquainted with the titanic dataset and the ML prediction tasks, check this notebook, it's a jewel :) https://www.kaggle.com/ccastleberry/titanic-cabin-features/notebook

Great, simple chart, giving a temporal dynamics. Some further idea could be more features involved (winners, rallies, forced errors etc. maybe they can be combined into a single metric etc.)

Nole like things the hard way :) typical Serb :D But it's a good match to represent the "importance" or "relativity" of the statistics.

I like how you handled the outliers. Made reasonable assumptions, documented them and moved on with the analysis.

good observation, also 'detected_ballhit_count' is the output of the 'ball_hit' algorithm so it's probably on the account of the algorithm accuracy. Good check would be to use inferred ball hits from the 'point_description' column.

great remark

great commenting, small remark that it would've probably been easier that the result is on a single plot. In addition, you could emphasize the difference in absolute points or percentage-wise

yup, it's considered one of "those" matches that in regular statistics you would assume that things ended up differently.

Concerning the organizational part, there's a great nbextension for generating automatic table of contents, so check it out. https://jupyter-contrib-nbextensions.readthedocs.io/en/latest/nbextensions/toc2/README.html

And a a collection of community-contributed unofficial extensions: https://jupyter-contrib-nbextensions.readthedocs.io/en/latest/

nice intro :)

Bins are always a good checkpoint for discussions, so here it goes :) Like for any kind of histogram we underepresent the data in order to have a "meaningful" plot and depending of the bins the "story" can go one way or the other. So it's cool to use some specific bin size to represent something, but always a good idea to check that the data "behaves' the same when the bin is smaller. Here in specific, because the data is not that "linear", it would probably be better to distribute them into equal bins (containing roughly the same amount of points) or go with the "domain knowledge" and use that (e.g. short rallies 1-4, medium 4-9, and long >9).

Just to emphasize the things already discussed, it's nice to have frequent comments. Great storytelling and thinking out loud.

Great, that you've picked up the story part. Also, it's nice to see that you can add comments that are facts, like conclusion and also some additional thoughts (my guess). So you're getting 2 things. One, it's much easier to go through a notebook, but also document your own line of thinking. It'll be much easier when/if you return to continue with the project later on.

There's a great nbextension for this part, so check it out. https://jupyter-contrib-nbextensions.readthedocs.io/en/latest/nbextensions/toc2/README.html

And a a collection of community-contributed unofficial extensions: https://jupyter-contrib-nbextensions.readthedocs.io/en/latest/

Nice set of plots, would be cool to include 'unforced errors' (but this is relevant from domain knowledge point of view) It's also interesting when you put it into perspective who won which set (it can be an added info to the plot).

storytelling part: e.g. It's also interesting that despite the statistics (provided here) in the 5th set, Djokovic still managed to take the victory

Not the same format of end data, but just wanted to give you a glimpse of power of using groupby. It also needs some time to get accustomed to the groupby logic (e.g. writing sql queries) like for vectorization as we discussed.

data.groupby(['p1_sets','p2_sets'])['point_duration'].sum().values/60

a more elaborate read: https://pandas.pydata.org/docs/user_guide/groupby.html

It's great flow how you elaborated from services to winners and errors, a few comments/observations on the charts and it's full house.

Nice set of plots, would be cool to include 'unforced errors' (but this is relevant from domain knowledge point of view) It's also interesting when you put it into perspective who won which set (it can be an added info to the plot).

storytelling part: e.g. It's also interesting that despite the statistics (provided here) in the 5th set, Djokovic still managed to take the victory

Not the same format of end data, but just wanted to give you a glimpse of power of using groupby. It also needs some time to get accustomed to the groupby logic (e.g. writing sql queries) like for vectorization as we discussed.

data.groupby(['p1_sets','p2_sets'])['point_duration'].sum().values/60

a more elaborate read: https://pandas.pydata.org/docs/user_guide/groupby.html