tsts 0.4.0

toolset for time series forecasting

(Docs)(Benchmark)

Introduction

tsts is an open-source easy-to-use toolset for time series forecasting.

What's New

Sep, 10

✨ New Model

Informer and related modules were added. See results for how to use it.

Installation

pip install tsts

Task Details

Time series forecasting is the task to predict the values of the time series on Horizon given the values of the time series on Lookback Period. Note that data can be multivariate.

Available Modules

Following modules are supported.

Architectures	Losses	Metrics	Optimizers	Scalers	Schedulers
Informer NBeats Seq2Seq	DILATE MAPE MSE	MAE MSE RMSE MAPE	Adam SGD	MinMaxScaler StandardScaler	CosineAnnealing StepScheduler

Getting Started

Following example shows how to train a model on sine curve dataset. See Docs for the details.

Training

Define config and start training with it.

# cfg.yml
LOGGER:
  # Log file and parameters are saved here
  LOG_DIR: "my-first-tsts-model"

import torch
from tsts.solvers import TimeSeriesForecaster

# Define trainining + validation datasets (they are divided inside)
sin_dataset = torch.sin(torch.arange(0.0, 100.0, 0.1))
sin_dataset = sin_dataset.unsqueeze(-1)

# Run training
forecaster = TimeSeriesForecaster("cfg.yml")
forecaster.fit([sin_dataset])

See results in my-first-tsts-model directory.

Inference

For inference, it needs to load parameters from a log directory generated in training.

import torch
from tsts.scalers import StandardScaler
from tsts.solvers import TimeSeriesForecaster

# Initialize scaler with training dataset
sin_dataset = torch.sin(torch.arange(0, 100, 0.1))
sin_dataset = sin_dataset.unsqueeze(-1)
X_scaler = StandardScaler()
# NOTE: 0.75 is default training validation dataset ratio (training: 0.75, validation: 0.25)
num_train_samples = int(0.75 * len(sin_dataset))
X_scaler.fit(sin_dataset)

# Define test dataset
X = torch.sin(torch.arange(100.0, 110.0, 0.1))
X = X.unsqueeze(-1)
X = X_scaler.transform(X)

# Run inference
forecaster = TimeSeriesForecaster("cfg.yml")
Z = forecaster.predict(X)

# Initialize target to compare prediction with it
y = torch.sin(torch.arange(110.0, 110.8, 0.1)).unsqueeze(1)
y = X_scaler.transform(y)

# Result
print(Z)
print(y)

Examples

See Benchmark for advanced usage.