Transform

Before reading this document, it is recommended to first read the TSDataset Document to understand the design of TSDataset. Simply speaking, TSDataset is a unified time series data structure throughout the modeling lifecycle. It introduces several fundamental but important time series related concepts such as target (including past_target and future_target) and covariates (including known_covariates and observed_covariates). A good understanding of those concepts would be helpful for deep diving into this documentation and building well-performed models.

PaddleTS provides a series of operators for data transformation, which can be applied to fill missing values, normalize the data, or encode the data in columns of TSDataset, etc. Each transformation operator is implemented as a class with three methods: fit, transform, and fit_transform. The fit method is usually used to learn the parameters for data transformation(e.g. mean and std for data normalization), the transform method is used to apply the data transformation based on the learnt parameters, and the fit_transform method combines tha above two steps into a single step.

Currently, PaddleTS supports the following data transformations:

• fill

• ksigma

• min_max_scaler

• standard_scaler

• onehot

• ordinal

• statistical

• time_feature

1. Example

The below demo takes OneHot as an example to illustrates the basic usage of data transformations.

from paddlets import TSDataset
from paddlets.transform import OneHot

# 1 prepare the data
data = TSDataset.load_from_csv("/path/to/data.csv")

# 2 init the onehot encoder
encoder = OneHot(cols=["Gender"])

# 3 fit the encoder
encoder.fit(dataset=data)

# 4 do transformation with the learnt encoder
transformed_data = encoder.transform(data, inplace=False)

Alternatively, one can also call the combined fit_transform to achieve the same purpose:

from paddlets import TSDataset
from paddlets.transform import OneHot

# 1 prepare the data
data = TSDataset.load_from_csv("/path/to/data.csv")

# 2 init the onehot encoder transform instance.
encoder = Onehot(cols=["Gender"])

# 3 fit + transform simultaneously
transformed_data = encoder.fit_transform(dataset=data)