This stage cleans the data for unwanted variables and samples, in addition to encoding the target variable if the case of a classification task.
clean.target: Name of target variable. Must correspond to the name of the
column that contains this variable in your .csv-files.clean.classification: True or False. Specify whether we are dealing
with a classification task or not. If set to True, the various classes will
be encoded to numbers.clean.onehot_encode_target: True or False. Specify whether we want to
one-hot encode the target variable or not. One-hot encoding is only needed
for classification tasks with more than two classes. Without one-hot
encoding, the various classes will be encoded as numbers on a continuous
scale. This usually gives a wrong impression of how the classes relate to
each other, for example by implying that some classes are closer than
others, while in fact they are actually just different categories. One-hot
encoding means that instead of encoding classes as numbers, each class will
get its own output vector that indicate whether a sample is a part of that
class or not. In some cases we might not want to use one-hot encoding even
though we have a multi-class classification task, for example if the
categories correspond to size (small, medium, large). In that case the
classes might actually be well represented as numbers on a continuous scale.clean.combine_files: True or False. Specify whether all files should be
combined into one before subsequences are extracted from the data. If we have
multiple files containing time series data, there is usually a gap in time
between the different files. In this case it does not make sense to
combine the files, because when subsequences are extracted from the data,
some subsequences will be made across this time gap. However, if the data
files contain one, continuous time series, the files can be combined. If your
data only resides in one file, this parameter will have no effect.clean.percentage_zeros_threshold: Fraction from 0.0 to 1.0. If a variable
has a fraction of zero-values higher than this threshold, it will be removed.
If this parameter is set to 1.0, only the variables that only contains zeros
will be removed.clean.input_max_correlation_threshold: Fraction from 0.0 to 1.0. If two
variables have a correlation higher than this threshold, one of them will be
removed. This parameter gives you a way of reducing the number of variables (which
in turn reduces computational costs and model complexity). However, keep in
mind that the model performance often can be reduced by setting this
parameter lower than 1.0. If set to 1.0, only variables that are perfectly
correlated will be affected.During this stage, the following operations will be performed:
clean.classification must be set to True in params.yaml, the target
variable will be encoded to a number.Previous stage: profile
Next stage: featurize