Can we do “predictive text” for time series?

[albert1480]

Hi all! Can anybody provide a tutorial for this?

[seanlaw]

@artem1480 Instead of an image can you please provide the source for the original document and what you are attempting to accomplish?

[albert1480]

after all I want to input simple audio (sine) signal as ts and predict it

[albert1480]

the main work is finding pattern cross two ts

[seanlaw]

Have you looked at these two existing tutorials yet?

1. https://stumpy.readthedocs.io/en/latest/Tutorial_AB_Joins.html
2. https://stumpy.readthedocs.io/en/latest/Tutorial_Pattern_Searching.html

[seanlaw]

According to the original work (not by us), they state that:

% The tool predicts the next subsequence based on the current subsequence
% If the current subsequence (S_current) is very similar to a subsequence
% in the past (S_past), then our best guess of the next subsequence is what
% followed S_past.
% The best guess is associated with a confidence level. The confidence
% level is the correlation between S_current and S_past.
% Only parameters data and sub_len are compulsory. The rest is optional.
% The code below is heavy on the visualization side.
% The core idea is given a query subsequence and a longer time series,
% compute the best matches with z-normalized Euclidean distance fast.

[albert1480]

ok(