Python module for machine learning time series:

Hi David,

I added the patch_sampler(imblearn_sampler_class) function which can be used to derive a dynamically created (and pickable) sampler class compatible with Pype. The derived class implements a transform method which returns the data unchanged. The fit_transform method calls the fit_resample method of the imbalanced-learn sampler which resamples the data. These steps are important to ensure that resampling only applies to training data but not to test data (the example shows that Pype.fit calls the fit_transform method, whereas score calls the transform method).

Cheers, Matthias

Hello, I have an issue about the shape of prediction is not same to the original labels due to the segmentation process. Do you have any function to convert the shape of prediction back to the shape before the segmentation?

Hi David,

I implemented the possibility to use a step size instead of the overlap percentage when segmenting. It has proven to be quite useful for me :).

Cheers, Matthias

Hello, I can't understand the usage of the segment class, in what cases I need to use this transform and how does it help? I also couldn't find an example as how to incorporate contextual variables? When I run it on toy data - it is very unclear what happened, since X is unchanged by y was reduced to a single value:

# Single multivariate time series with 3 samples of 4 variables
X = [np.array([[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11]])]
# Time series target
y = [np.array([True, False, False])]
print("X: " , X)
print("y: " ,y)
segment = Segment(width=3, overlap=1)
X, y, _ = segment.fit_transform(X, y)
print('After segmentation:')
print("X:", X)
print("y: ", y)

X : [array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])]
y : [array([ True, False, False])]
After segmentation:
X: [[[ 0  1  2  3]
  [ 4  5  6  7]
  [ 8  9 10 11]]]
y:  [False]

Hi, I was following your example code (simple regression), but I'm stuck. I have a DataFrame of shape (1017, 15). The last column is the target so I created two dfs, one for X (1017, 14) and one for y (1017). I tried to pass those values to temporal_split but I always get an error no matter what I do (passing the df, passing them as lists). For example, passing them as list gives:

KeyError: "None of [Int64Index([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9,\n ...\n 991, 992, 993, 994, 995, 996, 997, 998, 999, 1000],\n dtype='int64', length=1001)] are in the [columns]"

If, on the other hand, I pass them as df I get:

AttributeError: 'DataFrame' object has no attribute 'ts_data'

The same holds true if I manually split the DataFrames and pass them to seg.fit_transform(X_train, y_train) I tried to put the date column in the df as well as in the index but the error is still there. What's wrong?

Info of the Dataframe:

<class 'pandas.core.frame.DataFrame'>
Int64Index: 1017 entries, 896 to 1912
Data columns (total 13 columns):
 #   Column        Non-Null Count  Dtype         
---  ------        --------------  -----         
 0   date          1017 non-null   datetime64[ns]
 1   id            1017 non-null   object        
 2   price         1017 non-null   float64       
 3   month         1017 non-null   int64         
 4   year          1017 non-null   int64         
 5   event_name_1  1017 non-null   int64         
 6   event_type_1  1017 non-null   int64         
 7   event_name_2  1017 non-null   int64         
 8   event_type_2  1017 non-null   int64         
 9   snap_CA       1017 non-null   int64         
 10  dow           1017 non-null   int64         
 11  is_weekend    1017 non-null   int64         
 12  is_holiday    1017 non-null   int64         
dtypes: datetime64[ns](1), float64(1), int64(10), object(1)
memory usage: 111.2+ KB

I tried to use it with date column or date index or as a list. The same for y: I tried to use it a Series, a Dataframe with date column or date index and list both with or without the date column. As you see there are no NaN values.

Hi David,

as promised in #13, here is the generic FunctionTransformer for applying functions to Xt. Resampling Xt is disallowed and y not changed (and not provided to the supplied function).

Cheers, Matthias

New feature: ReconstructTs

Should go in new postprocessing module and reconstruct time series target labels from predictions on segments and mapped to the original data samples.

This could be implemented using interpolation (nearest neighbor) for categorical targets and anything else for continuous targets.

I don't think this can be integrated in the current pipeline atm. Another option would be to design another pipeline class that has this implemented as its last step.

Hi David,

I added a generic transformer which allows one to arbitrarily transform the data (Xt and yt) before or after segmentation using a custom function. Useful examples which come to mind are:

• selecting specific parts of data (e.g. column filter)
• applying filter functions (e.g. bandpass filter EMG data)

I added a safeguard against unintentional sub- or oversampling of the data as this would affect the training data (which is fine) and the test data (which is not fine) . The safeguard can be deactivated to allow for legitimate use cases similar to segmentation or interpolation of the data.

Could you please verify if my reasoning is correct. I would remove the last two commits if it were not.

Cheers, Matthias

Hi David,

I wrote a simple wrapper to use the sklearn ColumnTransformer on segmented data which is kind of useful when dealing with heterogeneous (multivariate) time series data. I've taken a look into supporting contextual data but did not find an easy way to make the current code work with the TS_Data class. Maybe copying and adapting the whole ColumnTransformer code instead of patching some parts of it could lead to a proper solution to support both. Nevertheless, I hope you find the SegmentedColumnTransformer to be useful.

Cheers, Matthias

When using Pype with scikit-lean version 0.24 I get the following error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/seglearn/pipe.py", line 59, in __init__
    super(Pype, self).__init__(steps, memory)
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/utils/validation.py", line 74, in inner_f
    return f(**kwargs)
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/pipeline.py", line 118, in __init__
    self._validate_steps()
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/pipeline.py", line 157, in _validate_steps
    self._validate_names(names)
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/utils/metaestimators.py", line 70, in _validate_names
    invalid_names = set(names).intersection(self.get_params(deep=False))
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/pipeline.py", line 137, in get_params
    return self._get_params('steps', deep=deep)
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/utils/metaestimators.py", line 29, in _get_params
    out = super().get_params(deep=deep)
  File "/home/wllr/miniconda3/envs/py38/lib/python3.8/site-packages/sklearn/base.py", line 195, in get_params
    value = getattr(self, key)
AttributeError: 'Pype' object has no attribute 'scorer'

Example to reproduce the error:

• python=3.8
• scikit-learn=0.24
• seglearn=1.2.1

from seglearn.transform import SegmentX
from seglearn.pipe import Pype

pipe = Pype([('segment', SegmentX())])   # will crash on creation

From a quick view on the seglearn's source code https://github.com/dmbee/seglearn/blob/9000eee4a1edd7dbb0bc4c3b63970c1b71c77a31/seglearn/pipe.py#L58-L64 one solution could be to move the call of the super's __init__ to the end

def __init__(self, steps, scorer=None, memory=None):
    self.scorer = scorer
    self.N_train = None
    self.N_test = None
    self.N_fit = None
    self.history = None
    super(Pype, self).__init__(steps, memory)

great code thanks may you clarify : will it work for multivariate time series prediction both regression and classification 1 where all values are continues values 2 or even will it work for multivariate time series where values are mixture of continues and categorical values for example 2 dimensions have continues values and 3 dimensions are categorical values

color        weight     gender  height  age  

1 black 56 m 160 34 2 white 77 f 170 54 3 yellow 87 m 167 43 4 white 55 m 198 72 5 white 88 f 176 32

Pype version: 1.2.2 Sklearn version: 1.0.1 Python version: 3.8.10

Problem: init() takes 2 positional arguments but 3 were given

Solution: change to super(Pype, self).init(steps, memory=memory) in seglearn/pipe.py

I have a z different time series with different lengths. For each time series, there are a different number of time points with timestamps and for each time point, there is an m different features and observed float outcome for this time point. My aim is modeling a regressor (given m features what is the outcome). I have trained a regressor by omitting the temporal dimension of a dataset (train on all data points using m features and predict the outcome), but it resulted in a poor result. (Multiple multivariate time series with different length and sampling frequency)

My aim is to add temporal dimension for each time point (like adding new features on rolling fashion, for each time point, mean of past values of features, std of past feature values etc). I could not find any example of adding new features to a data frame of Multiple multivariate time series with different length and sampling frequency. Can you help me?

How can I work with seglearn if I have a data representation that is presented here.

I have two cases. In the first case I have a variable that is time dependent so I would like to extract features from the previous values in order to build the X matrix.

2011-01-01 01:00:00    1.073392
2011-01-01 02:00:00    0.274406
2011-01-01 03:00:00    1.446233
2011-01-01 04:00:00   -0.035727

In the second case I have the same problem but having along one or more dependent (and time dependent) variables that I want to use them in order to predict the third one.

Implement data splitter with rolling splits similar to sklearn.model_selection.TimeSeriesSplit but with compatibility for data sets with more than a single time series and contextual data.