A new one created with my basic *datayse.* @*JB*Δ <http://jbigdata.fr>
2018-02-21 13:14 GMT+01:00 Philippe de Rochambeau <phi...@free.fr>: > Hi JB, > which column in the 8 line DS do you regress ? > > > Le 21 févr. 2018 09:47, JB Data <jbdat...@gmail.com> a écrit : > > Hi, > > Interesting discussion, let me add my *shell* point of view. > My focus is only Prediction, to avoid pure DS to "crier aux loups", I warn > how simple my *datayse* of the problem is : > - No use of the button in the model, only page navigation. > - User navigation 're-init' when click on a page ever clicked > - My dataset is only 8 rows large ! > 2018-01-02 12:00:00;OKK;PAG1;1234555 > 2018-01-02 12:01:01;NEX;PAG2;1234555 > 2018-01-02 12:00:02;OKK;PAG1;5556667 > 2018-01-02 12:01:03;NEX;PAG3;5556667 > 2018-01-02 12:01:04;OKK;PAG3;1234555 > 2018-01-02 12:01:04;NEX;PAG1;1234555 > 2018-01-02 12:01:04;NEX;PAG3;1234555 > 2018-01-02 12:01:04;OKK;PAG2;5556667 > > Anyway... After 250 python lines... > *Regression with SKlearn* > Mean squared error: 2.39 > Variance score: -0.54 > *Regression with Keras* > Results: -4.60 (3.83) MSE > > No doubt that increasing *wc -l* will increase *MSE*. > DL is the nowdays magic wand that everyone wants to shake above data > But mastering the wand is not for everyone (myself included), use wand > with parsimony... > > I create this group <https://github.com/dev2score/python/issues/1> as a > prerequisite of -1 feedback :-) > > > @*JB*Δ <http://jbigdata.fr> > > > 2018-02-10 16:28 GMT+01:00 Philippe de Rochambeau <phi...@free.fr>: > > Hi Jörn, > thank you for replying. > By « path analysis », I mean « the user’s navigation from page to page on > the website » and by « clicking trends » I mean « which buttons does > he/she click and in what order ». In other words, I’d like to measure, make > sense out of, and perhaps, predict user behavior. > > Philippe > > > > Le 10 févr. 2018 à 16:03, Jörn Franke <jornfra...@gmail.com> a écrit : > > > > What do you mean by path analysis and clicking trends? > > > > If you want to use typical graph algorithm such as longest path, > shortest path (to detect issues with your navigation page) or page rank > then probably yes. Similarly if you do a/b testing to compare if you sell > more with different navigation or product proposals. > > > > Really depends your analysis. Only if it looks like a graph does not > mean you need to do graph analysis . > > Then another critical element is how to visualize the results of your > graph analysis (does not have to be a graph to visualize, but it could be > also a table with if/then rules , eg if product placed at top right then > 50% more people buy it). > > > > However if you want to do some other analysis such as random forests or > Markov chains then graphx alone will not help you much. > > > >> On 10. Feb 2018, at 15:49, Philippe de Rochambeau <phi...@free.fr> > wrote: > >> > >> Hello, > >> > >> Let’s say a website log is structured as follows: > >> > >> <date and time>;<web item trigram>;<web page trigram>;<user id> > >> > >> eg. > >> > >> 2018-01-02 12:00:00;OKK;PAG1;1234555 > >> 2018-01-02 12:01:01;NEX;PAG1;1234555 > >> 2018-01-02 12:00:02;OKK;PAG1;5556667 > >> 2018-01-02 12:01:03;NEX;PAG1;5556667 > >> > >> where OKK stands for the OK Button on Page 1, NEX, the Next Button on > Page 2, … > >> > >> Is GraphX the appropriate tool to analyse the website users’ paths and > clicking trends, > >> > >> Many thanks. > >> > >> Philippe > >> > >> --------------------------------------------------------------------- > >> To unsubscribe e-mail: user-unsubscr...@spark.apache.org > >> > > > --------------------------------------------------------------------- > To unsubscribe e-mail: user-unsubscr...@spark.apache.org > > > >
