Skip to content

YiranHuangIrene/Large-Scale-Machine-Learning-Modeling-traffic-flow

Repository files navigation

Project 03: Modelling Traffic Flow

Installation

# Use pip-tools to manage dependencies
$ pip install pip-tools

# Install pinned dependencies
$ pip-sync

# Install package in editable mode
$ pip install -e .

Run the Training

Before you run anything, log into your W&B account with wandb login. Then you can run model training, for example, as follows.

./train.py data=example model=knn

Run on SLURM

To submit jobs to the SLURM cluster, set the launcher to slurm.

./train.py hydra/launcher=slurm # ... other arguments

Model

model=
GAT Vanilla Graph Attention Model
GnnRes Vanilla GNNRes with 1 timeslot as input
GnnRes-HF GNNRes with variable number of feature channels
GnnRes-High-Input GNNRes with multiple timeslots as input
GnnRes-Avg GNNRes with global average over training sample
Unet Vanilla U-Net

Parameters

For GNNRes-High-Input:

  • input_size: Number of input time slots
  • n_channels: Number of features channels insides residual blocks
  • inout_skipconn: If there should be an additional GCN on the input and output of the model
  • depth: Number of residual blocks
  • activation: Activation function (ReLU, Sigmoid)

Dataset

data=
t4c Traffic Forecasting Dataset
t4c_avg Traffic Forecasting Dataset including global average

Parameters

  • data_dir: Directory of data files
  • batch_size: Batchsize during training
  • input_size: Number of input time slots
  • city: City of data files
  • threshold: Threshold used for mask creation
  • include_timestamps: If timestamps should be included as circular features

Task

task=
temporal_forecasting Temporal Forecasting Task
temporal_forecasting_avg Temporal Forecating Task including global average
temporal_forecasting_unet Traffic Forecasting Task adapted to U-Net

Parameters

  • lr: Learning rate used during training

Large-Scale-Machine-Learning-Modeling-traffic-flow

Large-Scale-Machine-Learning-Modeling-traffic-flow

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published