lymboy-lstm 1.7.1

Some out-of-the-box lstm-based time series models

Updates

• 2023-03-16
  • Support multivariate time forcast
• 2023-03-08
  • AddCNN_BiLSTM_Attentionmodel

中文文档

Introduce

lymboy-lstm contains several commonly used LSTM models for time series forecasting. Currently only univariate time series forecasting is supported. Currently built-in models are: LSTM BiLSTM CNN_LSTM CNN_BiLSTM CNN_BiLSTM_Attention Other models are under study... (CNN_BiLSTM_Attention, Encoder-Decoder Model, Multivariate Time Prediction Support) Please look forward to it~

Packaging method

python ./setup.py sdist bdist_wheel
pip install dist/lymboy-lstm-[latest-version].tar.gz
# Upload to pypi
# pip install twine
# twine upload dist/*

How to install?

pip install lymboy-lstm

How to use?

Taking LSTM model to predict power consumption as an example

• Import lib

import pandas as pd
import numpy as np
from lstm import LSTM
from lstm.util import plot

• Read dataset

file = './dataset/power/power_consumption_A.csv'
df = pd.read_csv(file, index_col=0)
sequence = df.load

• Modeling

# Use the data of the past 96 times to predict the data of the next 10 times in the future
model = LSTM(n_steps=96, n_output=10)
# Process sequence data as model input, specifying a test set ratio of 20%
model.createXY(sequence, test_size=0.2)
model.fit(epochs=500, verbose=True)
print(model.score()) 

plot(model.y_hat[:,0], model.y_test[:,0])

CNN_BiLSTM predicts transformer oil temperature

• Import lib

import pandas as pd
import numpy as np
from lstm import LSTM, BiLSTM, CNN_BiLSTM
from lstm.util import plot

• Read dataset

file = './dataset/ETT/ETTh1.csv'
df = pd.read_csv(file, index_col=0)
sequence = df.OT

• Modeling

model = CNN_BiLSTM(n_steps=96, n_output=24, n_seq=6)
model.createXY(sequence)
model.fit(epochs=500, verbose=True)
print(model.score())

Prediction results of LSTM model on multiple data sets (multivariate, multi-step prediction)

Parameter Description

• n_steps: training step size, representing the step size of historical data, int
• n_output: predicted output length, int
• n_seq: subsequence, int (note that n_seq should be divisible by n_steps, the minimum is 1)
• learning_rate: learning rate for Adm, float

Other parameters are consistent with tensorflow

Error feedback

[e-mail]