meghanap1305/a3-ps003-tc122

How meghanap1305/a3-ps003-tc122 works

This repository implements traffic forecasting with a compact MLP, both centralized and simulated federated training, plus preprocessing and evaluation utilities. The main workflow loads and normalizes Parquet traffic windows, trains a model, and evaluates forecasts; the federated path partitions training rows, trains client copies, aggregates updates, and validates the global model. The README is minimal, so the architecture is grounded mainly in the indexed and excerpted source; demo wiring and other unsampled behavior are not inferred.

Main entry points: data_loader.py Window loader · federated_train.py Federated runner · train.py Central trainer · model.py Traffic MLP

Components 16

Data preparation

Parquet splits
Window loader preprocessing/data_loader.py
Client partitioning partition_data.py

Centralized forecasting

Central trainer src/train.py
Traffic MLP src/model.py

Federated learning

Client batch loader federated/client_data_loader.py
Local client trainer federated/client.py
Federated runner federated/federated_train.py
FedAvg aggregation federated/fedavg.py

Evaluation and demo

Persistence baseline preprocessing/baseline.py
Horizon metrics preprocessing/baseline_horizons.py
Global evaluator federated/evaluation.py
Demo server demo/server.py
Demo interface demo/static/index.html
Demo training demo/src/train.py

Other components

Operator

Connections 22

Operator → Central trainer: runs training src/train.py
Operator → Client partitioning: creates partitions partition_data.py
Operator → Federated runner: runs rounds federated/federated_train.py
Parquet splits → Window loader: reads batches preprocessing/data_loader.py
Window loader → Central trainer: provides batches src/train.py
Window loader → Persistence baseline: provides batches preprocessing/baseline.py
Window loader → Horizon metrics: provides batches preprocessing/baseline_horizons.py
Window loader → Client batch loader: supplies window logic federated/client_data_loader.py
Parquet splits → Client partitioning: reads training rows partition_data.py
Client partitioning → Client batch loader: writes client Parquet
Central trainer → Traffic MLP: trains src/train.py
Client batch loader → Local client trainer: yields batches federated/client.py
Local client trainer → Traffic MLP: trains local copy federated/client.py
Federated runner → Local client trainer: invokes clients federated/federated_train.py
Federated runner → FedAvg aggregation: aggregates updates federated/federated_train.py
Federated runner → Global evaluator: validates global model federated/federated_train.py
Federated runner → Traffic MLP: initializes global model federated/federated_train.py
Global evaluator → Window loader: reads validation batches federated/evaluation.py
Global evaluator → Traffic MLP: evaluates forecasts federated/evaluation.py
Operator → Demo server: opens demo
Demo server → Demo interface: serves interface
Demo training → Central trainer: uses training history

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·