I think I am not getting it when you do the rolling forecast origin: So you loop over your test_data index, and every time you do that you SHORTEN your training array (at least that is what I think the minus is doing there). Wouldn't it make more sense to extend your training array? At least that's what I thought you tried to elaborate before that part. Sorry for being late, but maybe somebody could clarify. :)
He is extending the train data in each loop. That minus is for excluding the last date for which we need to predict. But if we keep on adding next day data, won't the training data grow for ever ?
rather helpful, in forex (foreign exchange currency markets) it is handy to be able to predict one ATR (average true range) movement up or down - to guarantee a minimum reward to risk (where the risk is the max loss we are willing to take) - so if we could identify trends where the one ATR move is about 80+% in our favour then just that trend indicator would make us a mint...
Good videos...but the concern is the author does not seem to have enough time to answer questions here? Also the gaps like - why seasonality is not removed ( for stationarity ) before deciding to apply AR?
i don't get it all yet! i mean: what are proofs we need to say a model fits well? i have read that we must plot the acf for the residuals to realize if a model is going to work well for our data but i dont see in your videos those plots in theses examples! i knwo you talked about acf and pacf in other videos but you said nothig of using this plots to proof our model!
For rolling forecast origin, could I know if I need to use the predicted value for last month, or the actual value for the last month? Also, still, confused about what "averaging all the predictions" means?
for rolling forecast origin you use the real value from last month. but, it would also be interesting to look at the predicted value from last month when making this month's prediction because that would show you how the prediction uncertainty grows through time
Hello there, please upload all the contents of time series analysis. Would be a big favour for us, as there is no other youtube channel which focuses on this sectiion.
Thanks for the lectures on RUclips. However there are some explanation does not have caption like this "Evaluating Time Series Models : Time Series Talk" . would you make those to have caption on it?
The series came 3 years ago, still the best playlist for TS.
I don't see the code in your Github, could you please add the code? Thank you
In order to use a AR model, shouldn’t you maybe get rid of the seasonality in the data first?
One of the best playlists to learn High level TS concepts along with code. Great job, Ritvik.
Glad you think so!
Could the python code be uploaded in the Github repo? Thank you for the explanation
I think I am not getting it when you do the rolling forecast origin: So you loop over your test_data index, and every time you do that you SHORTEN your training array (at least that is what I think the minus is doing there). Wouldn't it make more sense to extend your training array? At least that's what I thought you tried to elaborate before that part. Sorry for being late, but maybe somebody could clarify. :)
He is extending the train data in each loop. That minus is for excluding the last date for which we need to predict. But if we keep on adding next day data, won't the training data grow for ever ?
rather helpful, in forex (foreign exchange currency markets) it is handy to be able to predict one ATR (average true range) movement up or down - to guarantee a minimum reward to risk (where the risk is the max loss we are willing to take) - so if we could identify trends where the one ATR move is about 80+% in our favour then just that trend indicator would make us a mint...
Very educative and informational.
Can rolling forecast origin be used for predicting New values out of dataset ?
Yes
I think only the first new value
Good videos...but the concern is the author does not seem to have enough time to answer questions here? Also the gaps like - why seasonality is not removed ( for stationarity ) before deciding to apply AR?
This might be a silly question: He tries to predict from three cycles, is that the same thing as predicting from three lags?
i don't get it all yet! i mean: what are proofs we need to say a model fits well? i have read that we must plot the acf for the residuals to realize if a model is going to work well for our data but i dont see in your videos those plots in theses examples! i knwo you talked about acf and pacf in other videos but you said nothig of using this plots to proof our model!
For rolling forecast origin, could I know if I need to use the predicted value for last month, or the actual value for the last month? Also, still, confused about what "averaging all the predictions" means?
for rolling forecast origin you use the real value from last month. but, it would also be interesting to look at the predicted value from last month when making this month's prediction because that would show you how the prediction uncertainty grows through time
@@ritvikmath thanks!!! Could I know the meaning of "averaging all the predictions"? I really appreciate your help.
@@yalili1600 I guess that means averaging the evaluation metrics (say accuracy ) , for all folds of the time series model.
Wonderful video sir!! Could you please share this notebook.
But for this case won't the last iteration have more data and then always have good results
Where can i find the code to this particular video ??? I don't see the code in your Github repository ??
Cool
I liked the simplified explaniation in each video
can you please provide the code for this lecture?
Hello there, please upload all the contents of time series analysis. Would be a big favour for us, as there is no other youtube channel which focuses on this sectiion.
Awesome video, helped me get the concept easily!
No need to normalize the data before training?
Great video!!!
Thanks for the lectures on RUclips. However there are some explanation does not have caption like this "Evaluating Time Series Models : Time Series Talk" . would you make those to have caption on it?