Documentation | Tutorials | Release Notes | 中文

TFTS (TensorFlow Time Series) is an easy-to-use time series package, supporting the classical and latest deep learning methods in TensorFlow or Keras. - Support sota models for time series tasks (prediction, classification, anomaly detection) - Provide advanced deep learning models for industry, research and competition - Documentation lives at time-series-prediction.readthedocs.io

Tutorial

Installation

• python >= 3.7
• tensorflow >= 2.4

shell pip install tfts

Quick start

```python import matplotlib.pyplot as plt import tensorflow as tf import tfts from tfts import AutoModel, AutoConfig, KerasTrainer

trainlength = 24 predictsequencelength = 8 (xtrain, ytrain), (xvalid, yvalid) = tfts.getdata("sine", trainlength, predictsequencelength, testsize=0.2)

modelnameorpath = 'seq2seq' # 'wavenet', 'transformer', 'rnn', 'tcn', 'bert', 'dlinear', 'nbeats', 'informer', 'autoformer' config = AutoConfig.formodel(modelnameorpath) model = AutoModel.fromconfig(config, predictsequencelength=predictsequencelength) trainer = KerasTrainer(model, optimizer=tf.keras.optimizers.Adam(0.0007)) trainer.train((xtrain, ytrain), (xvalid, yvalid), epochs=30)

pred = trainer.predict(xvalid) trainer.plot(history=xvalid, true=y_valid, pred=pred) plt.show() ```

Prepare your own data

You could train your own data by preparing 3D data as inputs, for both inputs and targets - option1 np.ndarray - option2 tf.data.Dataset

Encoder only model inputs

```python import numpy as np from tfts import AutoConfig, AutoModel, KerasTrainer

trainlength = 24 predictsequencelength = 8 nfeature = 2

xtrain = np.random.rand(1, trainlength, nfeature) # inputs: (batch, trainlength, feature) ytrain = np.random.rand(1, predictsequencelength, 1) # target: (batch, predictsequencelength, 1) xvalid = np.random.rand(1, trainlength, nfeature) yvalid = np.random.rand(1, predictsequence_length, 1)

config = AutoConfig.formodel('rnn') model = AutoModel.fromconfig(config, predictsequencelength=predictsequencelength) trainer = KerasTrainer(model) trainer.train(traindataset=(xtrain, ytrain), validdataset=(xvalid, yvalid), epochs=1) ```

Encoder-decoder model inputs

```python

option1: np.ndarray

import numpy as np from tfts import AutoConfig, AutoModel, KerasTrainer

trainlength = 24 predictsequencelength = 8 nencoderfeature = 2 ndecoder_feature = 3

xtrain = ( np.random.rand(1, trainlength, 1), # inputs: (batch, trainlength, 1) np.random.rand(1, trainlength, nencoderfeature), # encoderfeature: (batch, trainlength, encoderfeatures) np.random.rand(1, predictsequencelength, ndecoderfeature), # decoderfeature: (batch, predictsequencelength, decoderfeatures) ) ytrain = np.random.rand(1, predictsequencelength, 1) # target: (batch, predictsequencelength, 1)

xvalid = ( np.random.rand(1, trainlength, 1), np.random.rand(1, trainlength, nencoderfeature), np.random.rand(1, predictsequencelength, ndecoderfeature), ) yvalid = np.random.rand(1, predictsequencelength, 1)

config = AutoConfig.formodel("seq2seq") model = AutoModel.fromconfig(config, predictsequencelength=predictsequencelength) trainer = KerasTrainer(model) trainer.train((xtrain, ytrain), (xvalid, yvalid), epochs=1) ```

```python

option2: tf.data.Dataset

import numpy as np import tensorflow as tf from tfts import AutoConfig, AutoModel, KerasTrainer

class FakeReader(object): def init(self, predictsequencelength): trainlength = 24 nencoderfeature = 2 ndecoderfeature = 3 self.x = np.random.rand(15, trainlength, 1) self.encoderfeature = np.random.rand(15, trainlength, nencoderfeature) self.decoderfeature = np.random.rand(15, predictsequencelength, ndecoderfeature) self.target = np.random.rand(15, predictsequence_length, 1)

def __len__(self):
    return len(self.x)

def __getitem__(self, idx):
    return {
        "x": self.x[idx],
        "encoder_feature": self.encoder_feature[idx],
        "decoder_feature": self.decoder_feature[idx],
    }, self.target[idx]

def iter(self):
    for i in range(len(self.x)):
        yield self[i]

predictsequencelength = 10 trainreader = FakeReader(predictsequencelength=predictsequencelength) trainloader = tf.data.Dataset.fromgenerator( trainreader.iter, ({"x": tf.float32, "encoderfeature": tf.float32, "decoderfeature": tf.float32}, tf.float32), ) trainloader = trainloader.batch(batchsize=1) validreader = FakeReader(predictsequencelength=predictsequencelength) validloader = tf.data.Dataset.fromgenerator( validreader.iter, ({"x": tf.float32, "encoderfeature": tf.float32, "decoderfeature": tf.float32}, tf.float32), ) validloader = validloader.batch(batchsize=1)

config = AutoConfig.formodel("seq2seq") model = AutoModel.fromconfig(config, predictsequencelength=predictsequencelength) trainer = KerasTrainer(model) trainer.train(traindataset=trainloader, validdataset=validloader, epochs=1) ```

Prepare custom model config

```python from tfts import AutoModel, AutoConfig

config = AutoConfig.formodel('rnn') print(config) config.rnnhidden_size = 128

model = AutoModel.fromconfig(config, predictsequence_length=7) ```

Build your own model

You could build the custom model based on tfts, like - add custom-defined embeddings for categorical variables - add custom-defined head layers for classification or anomaly task

```python import tensorflow as tf from tensorflow.keras.layers import Input, Dense from tfts import AutoModel, AutoConfig

trainlength = 24 numtrainfeatures = 15 predictsequence_length = 8

def buildmodel(): inputs = Input([trainlength, numtrainfeatures]) config = AutoConfig.formodel("seq2seq") backbone = AutoModel.fromconfig(config, predictsequencelength=predictsequencelength) outputs = backbone(inputs) outputs = Dense(1, activation="sigmoid")(outputs) model = tf.keras.Model(inputs=inputs, outputs=outputs) model.compile(loss="mse", optimizer="rmsprop") return model ```

Examples

• TFTS-Bert wins the 3rd place in KDD Cup 2022-wind power forecasting
• TFTS-Seq2seq wins the 4th place in Tianchi-ENSO index prediction 2021
• More examples ...

Citation

If you find tfts project useful in your research, please consider cite:

@misc{tfts2020, author = {Longxing Tan}, title = {Time series prediction}, year = {2020}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/longxingtan/time-series-prediction}}, }