Fortune-Telling with Python: An Intro to Facebook Prophet
Вставка
- Опубліковано 7 вер 2017
- A pythonic tour of Facebook's time series package. Intermediate level with basic statistics and time data familiarity required.
Jonathan Balaban is a senior data scientist, strategy consultant, and entrepreneur with ten years of private, public, and philanthropic experience. He currently teaches business professionals and leaders the art of impact-focused, practical data science at Metis.
Founded in 2003, Chicago Python User Group is one of the world's most active programming language special interest groups with over 1,000 active members and many more prestigious alumni. Our main focus is the Python Programming Language.
~~ Connect with us! ~~
chipymentor.org
@ChicagoPython
chipy.slack.com
chipy.org - Наука та технологія
fantastic package..
Awesome talk! Thanks for sharing this!
great presentation!
Good explanation on the workings
that was really good; thanks!
Cool way to explain the model. Nice one...
his talk is clean like his dataset
Thanks
Just watched this, what a beautiful presentation! I wish everything I do some research on explained things as simply as this xD
I just used fbprophet model right now. To me, there are some limited parts: 1)Time problem:in daily case, one new data comes and use model to fit or predict again with almost same data except new one. 2) model: linear or logistic is not automatic decided, still analysis by human being, in the case we have a bunch of time series. 3) multiple seasonality: additive or keep some leave some out. It is decided by human being too.
would you consider releasing the example code you demo here?
possible to do high granular level - weekly/monthly etc?
Thanks for posting this, Jonathan! Do you have the link to the JP notebook?
You can find a recreated one here: colab.research.google.com/drive/114Hj-4ui4weXGbUerhuqau-au91JOaS-?usp=sharing
A great video, thank you so much.
In the first ex. you have taken log of y so for actual predicted value doesn't need to take antilog ??
YES. Predicted value should be corrected with numpy.exp(y_hat) to get the "human understandable" number :)
@@jalbarracin thank you for your answer.... Actually I am doing the same conversion but didn't find it in the video so asked....
Can u share the link to the juputer notebook you used? Thx
It is present on facebook's official github page for prophet package's documentation.
github.com/ultimatist/ODSC17
can this model take exogenous variables?
Where is the testing data comparing to the prediction data? wheres the error curve? evaluation metric?
Only works on academic data. On real industrial dataset, it fails miserably. In case of clear seasonality, it works somewhat better
Why so? Do you have any examples?
@@hadialkhamees2744 I tried it on some industrial datasets but didn't work. Sometimes the error is even higher than the actual values resulting in negative accuracies.
which package works better on Industrial data ?
The link is given here-
facebook.github.io/prophet/docs/quick_start.html
Now Combine this with Mundane Astrology , You get a More Accurate Prophecy .