sintel-dev/ml-stars

Primitives and Pipelines for Time Series Data

ml-stars's Issues

'mlstars.custom.timeseries_preprocessing.rolling_window_sequences' target_size = 0

• ml-stars version: 0.1.1
• Python version: 3.8
• Operating System: MacOS

Description

I am trying to create rolling window sequences to feed as input to a reconstruction model (LSTM-AE). Since it's not prediction, you want target_size = 0 so that the last window contains the last point, i.e. all points are captured in the window sequences. However, setting target_size = 0 raises an indexing error at line 108 aka

y_index.append(index[end + offset])

in mlstars/custom/timeseries_preprocessing.py.

What I Did

sample = 'E-10'
train_data = load_signal('multivariate/' + sample)
train_data.head()

window = 100
primitive = load_primitive('mlstars.custom.timeseries_preprocessing.rolling_window_sequences',
                           arguments = {
                               'index': train_data.index,
                               'window_size': window, 
                               'target_size': 0,
                               'step_size': 1,
                               'target_column': None
                           })

X, y, index, target_index = primitive.produce(X=train_data.drop('timestamp', axis=1).to_numpy())

Remove python 3.6

• Remove python 3.6 from support.
• Revert ubuntu image to latest.