search

gzerveas / mvts_transformer

Multivariate Time Series Transformer, public version
MIT License
718 stars 169 forks source link

Where I can I get the WeldData csv dataset? #62

Open VirtualVsMachine opened 7 months ago

VirtualVsMachine commented 7 months ago

I am struggling to get a working data class set up using the template. I believe it may be the structure of my data. I was hoping to get the WeldData csv to see how my format may be different.

Any help is appreciated.

Banging my head against the wall trying to debug: File "C:\Users\steve\OneDrive\Desktop\VsMachine\EXPERIMENTAL_MODELS\MVTS TRANSFORMER\src\datasets\dataset.py", line 115, in collate_superv X[i, :end, :] = features[i][:end, :] IndexError: too many indices for tensor of dimension 1

TTAyanlade commented 4 months ago

Hey @VirtualVsMachine , Did you find this?

VirtualVsMachine commented 4 months ago

I did notOn Feb 21, 2024, at 10:10 PM, Timilehin Ayanlade @.***> wrote: Hey @VirtualVsMachine , Did you find this?

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: @.***>

gzerveas commented 4 months ago

For the WeldData example, and any other custom dataset you wish to define, you can consider that each .csv file would have the following structure:

sample_ID timestamp feature 1 feature 2 label
02778 2020-06-02 00:05:23 12.4 0.002 1
02778 2020-06-02 00:05:24 12.5 0.095 1
02778 2020-06-02 00:05:25 12.2 0.098 1
14440 2020-06-01 14:05:18 9.15 -0.1 0
14440 2020-06-01 14:05:19 10.2 -0.098 0

This would be one file that contains 2 samples (time series), each with its own unique ID. The first sample here has a sequence length of 3, and the second one a sequence length of 2. Rows with the same ID correspond to the same sample (time series).

You can have several such .csv files, e.g. stored in the same directory (or accessible through a regex pattern), and their contents will be loaded and concatenated into the same dataframe.