Onboard NeuralProphet: a hybrid time-series forecasting library

chevron-icon
Back
project-presentation-img
Kevin R. C.
Project Owner

Onboard NeuralProphet: a hybrid time-series forecasting library

Funding Awarded

$14,000 USD

Expert Review
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 0
Community
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 0 (0)

Status

  • Overall Status

    🛠️ In Progress

  • Funding Transfered

    $0 USD

  • Max Funding Amount

    $14,000 USD

Funding Schedule

View Milestones
Milestone Release 1
$4,500 USD Pending TBD
Milestone Release 2
$2,500 USD Pending TBD
Milestone Release 3
$1,000 USD Pending TBD
Milestone Release 4
$2,500 USD Pending TBD
Milestone Release 5
$3,500 USD Pending TBD

Project AI Services

No Service Available

Overview

NeuralProphet is an open-sourced time-series forecasting library. It is a hybrid general additive model that combines the simplicity and usability of Facebook’s Prophet with the performance and flexibility of a neural network. As of date, NP has reached 1.2 million+ downloads and 2,700+ GitHub stars.

Proposal Description

Compnay Name

Temporai

Service Details

NeuralProphet  is an open-sourced time-series forecasting library. It is a hybrid general additive model that combines the simplicity and usability of Facebook’s Prophet

with the performance and flexibility of a neural network. As of writing this proposal, NeuralProphet has surpassed 1.35 million downloads and 2,750 GitHub stars. The team behind this proposal are also the core developers and maintainers of NeuralProphet.

This service acts as an API wrapper to the underlying NeuralProphet library, which includes the NeuralProphet model class. The wrapper automatically configures the hyperparameters that need to be passed into the NeuralProphet model based on the user's input data, as well as preprocessing that data wherever needed. Therefore, all the user need to do is input his/her data directly into this service API, bypassing the need to manually write Python code to use the NeuralProphet library. Nevertheless, the API service allows the user to customize certain hyperparameters, such as known seasonal patterns or events, if necessary.

The NeuralProphet library is created by Oskar T. and is open-sourced under his GitHub account. Lead core developers and maintainers include Kevin R.C. and Richard S. Please see the Related links section at the end for more details.

As for this service, the project lead is Kevin R.C., with Oskar T. and Richard S. as the technical advisers. Please see the Team section for more details.

Solution Description

NeuralProphet is a generalized additive model with the following components: trend, seasonality, events/holidays, autoregression, and lagged regressors/covariates. Its autoregression and lagged regressors/covariate uses an underlying neural network model called AR-Net, with PyTorch Lightning implementation. Its other components use more statistical methods (e.g., Fourier terms for seasonality) hence the hybrid nature.

Project Benefits for SNET AI platform

This NeuralProphet service will be utilized by the SYBIL: The General-Purpose Forecaster service as its hybrid base model. See our separate SYBIL proposal here in DF2 - Pool A [New projects].

Aside from SYBIL, the NeuralProphet service can also be its standalone forecasting service as an ample alternative to the existing Time-Series Analysis service. Although further research and testing need to be done, we expect the NeuralProphet service’s comparative strengths to include the following:

  • Automatic preprocessing steps
  • Handling multivariates as lagged and future regressors
  • Longer forecasting horizons
  • Support and maintenance directly from the NeuralProphet developers

Service License Info

The NeuralProphet library is open-sourced under the MIT Licenselink

This service will be open-sourced under the Apache 2.0 License.

Related Links

NeuralProphet:

Kevin’s demo:

  • Photrek FCNT NeuralProphet Demo with Catalyst Swarm (2022) - YouTube link

Proposal Video

SIBYL: The General-Purpose Forecaster

13 February 2023
  • Total Milestones

    5

  • Total Budget

    $14,000 USD

  • Last Updated

    8 Mar 2024

Milestone 1 - Develop wrapper code

Status
🧐 In Progress
Description

Develop wrapper code for the NP service to preprocess input data and execute the underlying NP model by passing in the necessary parameters.

Deliverables

Budget

$4,500 USD

Link URL

Milestone 2 - Integrate on cloud hosting

Status
😐 Not Started
Description

Have the wrapper code and library successfully integrated as an AWS or cloud service. For example, how to handle time-out and latency from multiple users.

Deliverables

Budget

$2,500 USD

Link URL

Milestone 3 - Deploy on SNET

Status
😐 Not Started
Description

Deploy this cloud service onto the SNET marketplace platform using the SNET APIs.

Deliverables

Budget

$1,000 USD

Link URL

Milestone 4 - Test and Maintain

Status
😐 Not Started
Description

Test the wrapper and service on various time-series datasets, and maintain this service, so it is in good order.All

Deliverables

Budget

$2,500 USD

Link URL

Milestone 5 - API Calls and Hosting

Status
😐 Not Started
Description

25% budget reservation for API calls or hosting.

Deliverables

Budget

$3,500 USD

Link URL

Join the Discussion (0)

Reviews & Rating

New reviews and ratings are disabled for Awarded Projects

Sort by

0 ratings

Summary

Overall Community

0

from 0 reviews
  • 5
    0
  • 4
    0
  • 3
    0
  • 2
    0
  • 1
    0

Feasibility

0

from 0 reviews

Viability

0

from 0 reviews

Desirabilty

0

from 0 reviews

Usefulness

0

from 0 reviews