Stationary time series

link to repository
I have lmdb file , what i am trying is to analyze the time series data. I used adf test to check stationarity. when i do test on whole series, adf test give result as “stationary”. but when I make several chunks of series and perform adf test on it, some parts it gives series as “stationary” some parts as “non-stationary”. what should i assume what kind of series is it???
I have attached the files to the data (as lmdb file) and code(adf-test) in the above mentioned link.(link to repository)

@ptrblck hello sir, please take a look at my query, i have seen you helping around people here,thanks in advance!

I’m not familiar enough with the Augmented Dickey-Fuller Test and my nave understanding would be that data splits could indeed show a trend while the overall data could be seen as “stationary”.
I’m also not familiar with your use case and don’t know how you would use this information in PyTorch.
In any case, maybe @KFrank or @tom would have a better idea how to interpret these results as mathematical experts.

Basically data is pressure values on hydraulic supports installed in coal mining, recorded every second .I am using this data for time series forecasting using LSTM, transformer and non-stationary transformer. but results are poor from both transformer models, I believe I need to preprocess the data for it. I have already performed several transformation as shown in the repository.
I would appreciate someone giving me a direction to deal with such kind of data. @KFrank @tom