163 - An introduction to time series forecasting - Part 3 Using ARIMA in python

[@RaghavaIndra]

Best tutorial on ARIMA forcasting. Thank you.

[@DigitalSreeni]

Glad it was helpful!

[@kabokbl2412]

if you watched last video start this one at 08:30

[@jaikishank]

Thanks, Digital Sreeni for the great video and vivid explanation of the implementation of ARIMA.This greatly helped to clear my doubts in TS data analysis and prediction. Deeply appreciate your efforts.

[@jacobsomer1882]

Best tutorials on youtube

[@RajKumar-ob8wk]

Thank you for making such a valuable video.

[@vivek6389]

Hello Sreeni, in most of your time series videos, we have 1-step ahead forecasts. Can you please make videos for multi-step ahead forecast. That would give us some insights into Direct vs Iterative Forecasting too. Thanks

[@khairulfahim]

Everything is offset by 1. What does that mean? Is the train_prediction model giving something like naive forecasting?

[@marvinvalentin8960]

This is a very helpful tutorial. wonder if this works for temperature measured in an hourly basis for one year. Thanks

[@khairulfahim]

Tried doing the ADFuller test. it didn't work. I'm getting the following error "ModuleNotFoundError: No module named 'pmdarima'"

[@rathnakumarv3956]

Sir are we having nse(nash-sutcliffe-efficiency) as performance metric in math / scicit/keras because in hydrology that is the general metric recommended

[@sf9746]

Why shouldn’t we use auto_arima on the training set rather than the whole set?

[@abhisheksingh65478]

Can you tell how can we do multivariate time series analysis with SARIMA?

[@GoldenMoments100]

how can I use neural networks to predict a chunk data in the future like this? all I ever see with neural networks are these 1-step incremental predictions on existing data. What if I want to predict a whole chunk of future data to see whether the neural network actually recognized the patterns inherent to the dataset? Otherwise the neural networks are essentially just creating a sort of moving average of the data.

[@manuelsteele8030]

Hello, my name is Manuel Steele. I am a PhD student in data science at Arizona State U. in Tempe, AZ (USA). Question: what book(s) or article(s) would you recommend to supplement your video and code? There are many resources out there - easy to get buried in months of reading when I need to generate results quickly (e.g. within a few weeks or so). Your videos are awesome and most clear given my background in python, Julia, R and machine learning. I am focused on time series data right now.

[@maliniv8043]

Aurélien Géron - Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow_ Concepts, Tools, and Techniques to Build Intelligent Systems-O'Reilly Media (2022)

[@akhmadmasyudi5539]

when performing stepwise search to minimize AIC, what is the difference between intercept and non-intercept model?

[@s.e.7268]

very understandable for beginners, thank you.

[@sidc.3817]

I did fine with the X_train prediction. But why am I getting a flat line for the X_test set prediction?

[@sidc.3817]

oops nvm....did something wrong with the d and D parameters during autoarima....its good now

[@saipurnaniharikapakalapati3029]

how to include a data set can u pls explain , i dont understand as im a new bee

[@peiyuli2824]

what if I want to apply Arima to many time series , and do the prediction parallel? How should I set the input and output?

[@maxwellcui3746]

Hi Sir, Thanks for your video, when auto-arima calculate the best model according to the least AIC, I think the result gives us(0,1,1)x(2,1,0,12), rather than (0,1,1)x(2,1,1,12).

[@Dr_Techo]

Quick question: Your start p was 1, how come the best model is giving p=0? SARIMAX(p,d,q)(P,D,Q)s

[@christossylalika6032]

Thank you so much

[@aarifansari9097]

clearly explain. thank you.

[@gernz1000]

Thank you, i cant install the pmdarima package, i think my python version is to new...

[@nagarajrajendiran7396]

memoryerror: unable to allocate 668. mib for an array with shape (27, 27, 120192) and data type float64..how to handle this error

[@DigitalSreeni]

Seems to be an issue with your local system memory. May be this helps: stackoverflow.com/questions/57507832/unable-to-allocate-array-with-shape-and-data-type

[@fathimahida8878]

Thankyou sir

[@classictremonti7997]

Hello...Great video series! Regarding RMSE values, how do you establish that a value is is a "good result"? It seems reasonable that you could compare the performance of models relative to each other and look for lower RMSE values, I am just not sure how to establish whether a value is either "good" or "bad".

[@DigitalSreeni]

RMSE by itself does not mean anything. You need to compare it against the dependent variable. If your dependent variable ranges from 0 to 100, an RMSE value of 0.5 is good. It the range is from 0 to 1 then the same 0.5 is not good. If my model works well on the training data then I compare test RMSE against train RMSE. This should give you a good idea about the magnitude of error. Again, always compare them against the magnitude of your dependent variable.

[@amintaleghani2110]

@DigitalSreeni Thank for your very informative video. I have tried to apply auto_arima (with the same parameters as yours) on my dataset but it got stuck in this stage: ARIMA(2,1,0)(4,1,0)[12] : AIC=5550.167, Time=3.82 sec any Idea what is the problem? thanks

[@DigitalSreeni]

May be it is taking too much time on a specific model and the resources are not enough. Please try with smaller data, you do not need your entire dataset for exploration of parameters and models.

[@kailashj2145]

@@DigitalSreeni a small request sir, pls put a disclaimer about the hardware requirement for some cases, would be helpful.

[@kailashj2145]

I exhausted 30gb/32gb ram in my server for running just 10 fit steps