Thanks Nishiraj..

🔥🔥

In fbProphet, if you are performing a multivariate forecasting, in simple terms let say you are predicting y based on x1, x2, …. xn, then you need to future values of all x variables, one way to achieve is forecasting them individually using univariate forecasting and using those forecasted values to predict y

@@SatyajitPattnaik cant thank you enough as i needed a quick reply for my thesis 💜...again, what you just explained, is this the technique applied for other multivariate time series forecasting models too?

@@tabitajannatul6736 any other ts algo will have some drawbacks in multi variate forecasting, try xgboost regressor

Surely, in few hours.

@@SatyajitPattnaik Thank you!

@@joseluisbeltramone599 Materials are already provided in the description: github.com/pik1989/FBProphet

Yes, can!!

If your data is something like 2010-11, 2010-12, 2011-01... and so on, it is in a monthly format. 8 quarter is 2 years, which is 24 months. So if I understood correctly, you should do this: future_data = model.make_future_dataframe(periods=24, freq = 'M') Or you can aggregate the target for each quarters by summing up every 4 months, and then use quarter format. In that case, your freq would be 'Q' If I made any mistake, please correct me.

??

Multi variate simply means multiple variables, you can choose to predict multiple variables or a single variable which is dependent on multiple variables, both the scenarios comes under multi variate.

​@@SatyajitPattnaik yah the issue with this approach is that we also need future regressor data to predict y but dont know how to get future regressor data without prediction

@@AbdulMusawarSoomro sometimes, we predict them individually and use predicted values

check video description

@@SatyajitPattnaik Thanks, BTW I even don't know how to use github repository. I am chemical engineer (PhD in Materials Engineering) and don't know much about computer science and programming.

@@SatyajitPattnaik I want to learn Machine learning, please suggest any online course which can help me. Thanks

@@Dr.mzeshanAli I am launching my end to end DS program in next 15-30days, stay tuned!!

@@SatyajitPattnaik Thanks Dear

I know fbProphet is bit problematic, try on Google colab, it works perfectly, and in meanwhile, try to solve it by yourself, I solved it myself, and so did many people, so try it out 👍

While predicting temperature, others are used as features (x variables)

@@SatyajitPattnaik I understood they are all independent variables but I want to know while predicting temperature into the future, whether the humidity value used is forecasted one or actual data from our data set, could you please briefly explain it.

@@kirankumarkr4611 Actual values

@@SatyajitPattnaik thank you sir

@@SatyajitPattnaik Since we are forecasting temperature into the future how will we use values of independent variables in the future because it will be unknown right?

Yes

Is there any video of you for vars modeling ?

@@ranjankumarnath4416 not yet, can do a video on it

Send me your file with data at pattnaiksatyajit89@gmail.com , i can check

@@SatyajitPattnaik Thank you so much for your reply. Sent to you the code and the data. Have a great day man

Facebook prophet always does a out of boundary forecasts, in real time industrial implementations, we train a model on backdated day, and then calculate the rmse or mse for the test sampleset.. For ex: if you have Jan to Dec 2020 data, train till sep 2020, and forecast on Oct, Nov, Dec data, and then calculate rmse for these 3 month predictions, do that with all the algos, the lowest rmse model turns out to be the best model. This is the logic. Practically we can use sklearn metrics to calculate rmse, mse or any other regression metrics.

@@SatyajitPattnaik Brother, your tutorial is very comprehensive. Thank you for this!

To add to what Satyajit sir said, here's how you can use sklearn to find the RMSE: from sklearn.metrics import mean_squared_error import math MSE = mean_squared_error(y_true, y_pred) RMSE = math.sqrt(MSE) print("RMSE:", RMSE)

It works!!

@@SatyajitPattnaik Sorry, it was my misunderstanding. It works with regressor though as obvious, now we need the other variables and it is not just ds anymore, therefore model.make_future_dataframe doesn't work which make sense.

pip install pystan pip install prophet

@@SatyajitPattnaik tried that not working

@@SatyajitPattnaik if possible can you make step by step process for it

@@user-zr1jl3ey5d whats the error?

@@SatyajitPattnaik import fbprophet this not working