TimeLinear
Description
This API calls the TimeLinear model, which can be used for time series forecasting.
TimeLinear Model Architecture

A model that combines a simple yet efficient module, Time Stamp Forecaster (TimeSter), for effectively encoding time-related features with a linear model.
TimeLinear Model Structure:
-
TimeSter (Time Stamp Forecaster)
- Encodes time-related features (e.g., year, month, day, hour, minute) to learn cyclical and seasonal patterns in time series data.
- Input
- Time-related features extracted from time information (e.g., [2024, 12, 12, 13, 45]).
- Output
- A representation generated from time-related features.
-
BonSter
- Uses past observations of multivariate time series data (value features) to generate predictions.
- Input
- Value features of the data.
- Characteristics
- Utilizes a linear model by default, but can be replaced with any backbone model if needed.
-
Add DeNorm (Combining Module)
- Combines the results from TimeSter and BonSter to generate the final prediction.
- Formula
- Beta × BonSter Output + (1 - Beta) × TimeSter Output.
- Beta is a parameter controlling the weight.
API module path
from api.v2.model.TimeLinear import TimeLinear
Parameters
seq_len
- Specifies the length of the input sequence.
- Example
- If data's shape (x,60,x).
- seq_len = 60
pred_len
- Specifies the prediction length.
- Example
- pred_len = 1
time_dim
- Specifies the time feature length.
- Example
- time_dim = 4
c_out
- Specifies the output time feature length.
- It must same pred_len
- Example
- c_out = 1
rda
- Specifies the First reduction factor.
- Example
- rda = 1
rdb
- Specifies the Second reduction factor.
- Example
- rdb = 1
Ksize
- Specifies the conv1d kernel size.
- Example
- ksize = 3
beta
- Specifies the prediction adjustment weight.
- Example
- beta = 0.8
Example Sample Code (Python)
Results

Check the entire module code.
datahub/api/v2/model/TimeLinear.py at main · machbase/datahub
All Industrial IoT DataHub with data visualization and AI source - machbase/datahub