DeepXF: Explainable Forecasting and Nowcasting with State-of-the-art Deep Neural Networks and Dynamic Factor Model
deep-xf
pypi: https://pypi.org/project/deep_xf
Related Blog: https://towardsdatascience.com/interpretable-nowcasting-with-deepxf-using-minimal-code-6b16a76ca52f
Related Blog: https://towardsdatascience.com/learning-similarities-between-biomedical-signals-with-deep-siamese-network-7684648e2ba0
Related Blog: https://ajay-arunachalam08.medium.com/denoising-ecg-signals-with-ensemble-of-filters-65919d15afe9
DeepXF is an open source, low-code python library for forecasting and nowcasting tasks. DeepXF helps in designing complex forecasting and nowcasting models with built-in utility for time series data. One can automatically build interpretable deep forecasting and nowcasting models at ease with this simple, easy-to-use and low-code solution. It enables users to perform end-to-end Proof-Of-Concept (POC) quickly and efficiently. One can build models based on deep neural network such as Recurrent Neural Network (RNN), Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional RNN/LSTM/GRU (BiRNN/BiLSTM/BiGRU), Spiking Neural Network (SNN), Graph Neural Network (GNN), Transformers, Generative Adversarial Network (GAN), Convolutional Neural Network (CNN), and others. It also provides facility to build nowcast model using Dynamic Factor Model.
DeepXF is conceived and developed by Ajay Arunachalam - https://www.linkedin.com/in/ajay-arunachalam-4744581a/
Please Note:- This is still by large a work in progress, so always open to your comments and things you feel to be included. Also, if you want to be a contributor, you are always most welcome. The RNN/LSTM/GRU/BiRNN/BiLSTM/BiGRU are already part of the initial version roll-out, while the latter ones (SNN, GNN, Transformers, GAN, CNN, etc.) are work in progress, and will be added soon once the testing is completed.
- Exploratory Data Analysis with services like profiling, filtering outliers, univariate/multivariate plots, plotly interactive plots, rolling window plots, detecting peaks, etc.
- Data Preprocessing for Time-series data with services like finding missing, imputing missing, date-time extraction, single timestamp generation, removing unwanted features, etc.
- Descriptive statistics for the provided time-series data, Normality evaluation, etc.
- Feature engineering with services like generating time lags, date-time features, one-hot encoding, date-time cyclic features, etc.
- Finding similarity between homogeneous time-series inputs with Siamese Neural Networks.
- Denoising time-series input signals.
- Building Deep Forecasting Model with hyperparameters tuning and leveraging available computational resource (CPU/GPU).
- Forecasting model performance evaluation with several key metrics
- Game theory based method to interpret forecasting model results.
- Building Nowcasting model with Expectation–maximization algorithm
- Explainable Nowcasting
DeepXF is an open-source library ideal for:-
- Citizen Data Scientists who prefer a low code solution.
- Experienced Data Scientists who want to increase model accuracy and improve productivity.
- Data Science Professionals and Consultants involved in building proof-of-concept (poc) projects.
- Researchers for quick poc prototyping and testing.
- Students and Teachers.
- ML Enthusiasts.
- Learners.
- Python 3.6.x
- torch[>=1.4.0]
- NumPy[>=1.9.0]
- SciPy[>=0.14.0]
- Scikit-learn[>=0.16]
- statsmodels[0.12.2]
- Pandas[>=0.23.0]
- Matplotlib
- Seaborn[0.9.0]
- plotly[5.10.0]
- tqdm
- shap
- keras[2.6.0]
- pandas_profiling[3.1.0]
- py-ecg-detectors
sudo bash setup.shUsing pip:
pip install deep-xf or pip3 install deep-xf or pip install git+git://github.com/ajayarunachalam/Deep_XF$ git clone https://github.com/ajayarunachalam/Deep_XF
$ cd Deep_XF
$ python setup.py installUsing notebook:
!pip install deep-xfUsing conda:
$ conda install -c conda-forge deep-xf- FORECASTING DEMO:
# set model config
select_model, select_user_path, select_scaler, forecast_window = Forecast.set_model_config(select_model='rnn', select_user_path='./forecast_folder_path/', select_scaler='minmax', forecast_window=1)
# select hyperparameters
hidden_dim, layer_dim, batch_size, dropout, n_epochs, learning_rate, weight_decay = Forecast.hyperparameter_config(hidden_dim=64, layer_dim = 3, batch_size=64, dropout = 0.2, n_epochs = 30, learning_rate = 1e-3, weight_decay = 1e-6)
# train model
opt, scaler = Forecast.train(df=df_full_features, target_col='value', split_ratio=0.2, select_model=select_model, select_scaler=select_scaler, forecast_window=forecast_window, batch_size=batch_size, hidden_dim=hidden_dim, layer_dim=layer_dim,dropout=dropout, n_epochs=n_epochs, learning_rate=learning_rate, weight_decay=weight_decay)
# forecast for user selected period
forecasted_data, ff_full_features, ff_full_features_ = Forecast.forecast(model_df, ts, fc, opt, scaler, period=25, fq='1h', select_scaler=select_scaler,)
# interpret the forecasting result
Helper.explainable_forecast(df_full_features, ff_full_features_, fc, specific_prediction_sample_to_explain=df_full_features.shape[0]+2, input_label_index_value=0, num_labels=1)__author__ = 'Ajay Arunachalam'
__version__ = '0.0.1'
__date__ = '7.11.2021'
from deep_xf.main import *
from deep_xf.dpp import *
from deep_xf.forecast_ml import *
from deep_xf.forecast_ml_extension import *
from deep_xf.stats import *
from deep_xf.utility import *
from deep_xf.denoise import *
from deep_xf.similarity import *
df = pd.read_csv('../data/PJME_hourly.csv')
print(df.shape)
print(df.columns)
# set variables
ts, fc = Forecast.set_variable(ts='Datetime', fc='PJME_MW')
# get variables
model_df, orig_df = Helper.get_variable(df, ts, fc)
# EDA
ExploratoryDataAnalysis.plot_dataset(df=model_df,fc=fc, title='PJM East (PJME) Region: estimated energy consumption in Megawatts (MW)')
# Feature Engg
df_full_features = Features.generate_date_time_features_hour(model_df, ['hour','month','day','day_of_week','week_of_year'])
# generating cyclic features
df_full_features = Features.generate_cyclic_features(df_full_features, 'hour', 24, 0)
df_full_features = Features.generate_cyclic_features(df_full_features, 'day_of_week', 7, 0)
df_full_features = Features.generate_cyclic_features(df_full_features, 'month', 12, 1)
df_full_features = Features.generate_cyclic_features(df_full_features, 'week_of_year', 52, 0)
# holiday feature
df_full_features = Features.generate_other_related_features(df=df_full_features)
select_model, select_user_path, select_scaler, forecast_window = Forecast.set_model_config(select_model='rnn', select_user_path='./forecast_folder_path/', select_scaler='minmax', forecast_window=1)
hidden_dim, layer_dim, batch_size, dropout, n_epochs, learning_rate, weight_decay = Forecast.hyperparameter_config(hidden_dim=64, layer_dim = 3, batch_size=64, dropout = 0.2, n_epochs = 30, learning_rate = 1e-3, weight_decay = 1e-6)
opt, scaler = Forecast.train(df=df_full_features, target_col='value', split_ratio=0.2, select_model=select_model, select_scaler=select_scaler, forecast_window=forecast_window, batch_size=batch_size, hidden_dim=hidden_dim, layer_dim=layer_dim,dropout=dropout, n_epochs=n_epochs, learning_rate=learning_rate, weight_decay=weight_decay)
forecasted_data, ff_full_features, ff_full_features_ = Forecast.forecast(model_df, ts, fc, opt, scaler, period=25, fq='1h', select_scaler=select_scaler,)
Helper.explainable_forecast(df_full_features, ff_full_features_, fc, specific_prediction_sample_to_explain=df.shape[0]+1, input_label_index_value=0, num_labels=1)- NOWCASTING DEMO:
# set model config
select_model, select_user_path, select_scaler, forecast_window = Forecast.set_model_config(select_model='em', select_user_path='./forecast_folder_path/', select_scaler='minmax', forecast_window=5)
# nowcast for user selected window
nowcast_full_data, nowcast_pred_data = EMModel.nowcast(df_full_features, ts, fc, period=5, fq='1h', forecast_window=forecast_window, select_model=select_model)
# interpret the nowcasting model result
EMModel.explainable_nowcast(df_full_features, nowcast_pred_data, fc, specific_prediction_sample_to_explain=df.shape[0]+2, input_label_index_value=0, num_labels=1)__author__ = 'Ajay Arunachalam'
__version__ = '0.0.1'
__date__ = '7.11.2021'
from deep_xf.main import *
from deep_xf.dpp import *
from deep_xf.forecast_ml import *
from deep_xf.forecast_ml_extension import *
from deep_xf.stats import *
from deep_xf.utility import *
from deep_xf.denoise import *
from deep_xf.similarity import *
df = pd.read_csv('./data/PJME_hourly.csv')
# set variables
ts, fc = Forecast.set_variable(ts='Datetime', fc='PJME_MW')
# get variables
model_df, orig_df = Helper.get_variable(df, ts, fc)
select_model, select_user_path, select_scaler, forecast_window = Forecast.set_model_config(select_model='em', select_user_path='./forecast_folder_path/', select_scaler='minmax', forecast_window=5)
df_full_features = Features.generate_date_time_features_hour(model_df, ['hour','month','day','day_of_week','week_of_year'])
# generating cyclic features
df_full_features = Features.generate_cyclic_features(df_full_features, 'hour', 24, 0)
df_full_features = Features.generate_cyclic_features(df_full_features, 'day_of_week', 7, 0)
df_full_features = Features.generate_cyclic_features(df_full_features, 'month', 12, 1)
df_full_features = Features.generate_cyclic_features(df_full_features, 'week_of_year', 52, 0)
df_full_features = Features.generate_other_related_features(df=df_full_features)
nowcast_full_data, nowcast_pred_data = EMModel.nowcast(df_full_features, ts, fc, period=5, fq='1h', forecast_window=forecast_window, select_model=select_model)
EMModel.explainable_nowcast(df_full_features, nowcast_pred_data, fc, specific_prediction_sample_to_explain=df.shape[0]+3, input_label_index_value=0, num_labels=1)- Find the forecasting demo here : https://github.com/ajayarunachalam/Deep_XF/blob/main/demo_notebook/Forecast-Demo.py
- Find the notebook of the nowcasting demo here : https://github.com/ajayarunachalam/Deep_XF/blob/main/demo_notebook/Nowcast-Demo.ipynb
Copyright 2021-2022 Ajay Arunachalam <[email protected]>
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. © 2021 GitHub, Inc.