/abm-emulator

Primary LanguageJupyter Notebook

Neural Agent Based Modeling

Introduction

As an alternative to Agent-Based simulation Models, a deep learning-based emulator can be developed as a surrogate model that can predict the Agent-Based Simulator's output based on its initial conditions and parameters.

alt text

Experiment Log

v3.11 - Dilated CNN MSE Loss

v3.12 - ConvLSTM MSE Loss

v3.13 - 3D-CNN MSE Loss

v3.21 - Dilated CNN negative log likelihood Loss

v3.22 - ConvLSTM negative log likelihood Loss

v3.23 - 3D-CNN negative log likelihood Loss

v3.31 - Dilated CNN MAE Loss

v3.32 - ConvLSTM MAE Loss

v3.33 - 3D-CNN MAE Loss

v3.41 - Dilated CNN MAE + penalty Loss

v3.42 - ConvLSTM MAE + penalty Loss

v3.43 - 3D-CNN MAE + penalty Loss

v3.51 - Dilated CNN MSE + penalty Loss

v3.52 - ConvLSTM MSE + penalty Loss

v3.53 - 3D-CNN MSE + penalty Loss

v3.61 - Dilated CNN MAPE Loss