Gynnii/tft-scheduler-runtime

TFT Scheduler Runtime

Model Overview

Model Name: TFT Scheduler Runtime Developed by: Abdul Sittar Domains: Climate, COVID, Technology Model Type: Temporal Fusion Transformer (TFT) based sequence forecasting Frameworks: PyTorch / PyTorch Lightning License: Apache 2.0

This repository contains TFT models for multi-domain sequence prediction. The models are trained to predict actions, posts, and user interactions over time for three domains: Climate, COVID, and Technology.

Model Card

Model Description

The TFT Scheduler Runtime is a collection of Temporal Fusion Transformer models for multi-domain social simulation. It predicts the temporal evolution of user actions and post interactions for agent-based simulations.

• —Climate domain: Predicts climate-related agent interactions
• —COVID domain: Predicts COVID-related agent interactions
• —Technology domain: Predicts technology-related agent interactions
• —Model type: TFT (Temporal Fusion Transformer)
• —Checkpoint format: PyTorch .ckpt
• —Encoders: Numpy .npy files for user, post, and action embeddings
• —Intended use: Research, simulation, and domain-specific temporal forecasting

Dataset Used

We used the Social Graph Inference Reddit dataset:

DOI / Link: https://zenodo.org/records/18082502

Authors/Creators:

• —Sittar, Abdul
• —Guček, Alenka
• —Češnovar, Miha

Description: A large-scale, empirically grounded dataset from Reddit to support agent-based social simulations. Includes:

• —33 technology-focused agents
• —14 climate-focused agents
• —7 COVID-related agents
• —Each domain includes over one million posts and comments

The dataset defines agent categories, derives inter-agent relationships, and builds directed, weighted networks reflecting real user interactions.

License

This repository is released under Apache 2.0 License, which allows:

• —Commercial and non-commercial use
• —Modification and redistribution
• —Attribution required

Apache 2.0 License Details

Repository Files

Climate

• —models/tft_climate_*.ckpt – TFT model checkpoints
• —encoders/*.npy – User, post, and action embeddings
• —last_sequence/last_sequence.npy – Last sequence state

COVID

• —models/tft_covid_*.ckpt – TFT model checkpoints
• —encoders/*.npy – Encoders
• —last_sequence/last_sequence.npy

Technology

• —models/tft_tech_*.ckpt – TFT model checkpoints
• —encoders/*.npy – Encoders
• —last_sequence/last_sequence.npy

All large files are tracked via Git LFS.

Usage Example

import torch
from pathlib import Path
import numpy as np

# Load Climate TFT model
model_path = Path("Climate/models/tft_climate_target_post_encoded.ckpt")
model = torch.load(model_path, map_location="cpu")

# Load encoders
user_encoder = np.load("Climate/encoders/user_encoder.npy")
post_encoder = np.load("Climate/encoders/post_encoder.npy")
action_encoder = np.load("Climate/encoders/action_encoder.npy")

# Use model for sequence prediction or simulation