mihirchouhan/data-center-federated-learning

How mihirchouhan/data-center-federated-learning works

This is a resumable M100 forecasting and research-training template. The user-facing workflow downloads and verifies Dataset 5, prepares chronological forecasting windows, trains recurrent models, then evaluates held-out forecasts and writes artifacts. Alongside that baseline, the repository contains centralized and simulated FedAvg experiments, historical-coverage audits, and missing-data policy experiments. The graph includes shared data, model, evaluation, resource, and durable-I/O layers. Relationships beyond README-described stages are limited to references recorded in the source index; unsampled implementations are not assumed to have additional call paths.

Main entry points: federated_rounds.py Round Orchestration · pipeline.py Pipeline CLI · experiment_eval.py Experiment Evaluation · history_data_v4.py Historical Data Access

Components 29

M100 Workflow

Pipeline CLI pipeline.py
Dataset Download src/download.py
Window Preparation src/preprocess.py
Training CLI main.py

Experiment Framework

Experiment Runner experiments.py
Frozen Experiment Data src/experiment_data.py
Experiment Models src/experiment_models.py
Experiment Evaluation src/experiment_eval.py
Resource Monitoring src/experiment_resources.py
Centralized SGD src/centralized_sgd.py
Centralized Evaluation src/centralized_evaluate.py

Federated Training

FedAvg Runner federated.py
Federated Partitions src/federated_data.py
Round Orchestration src/federated_rounds.py
Client Updates src/federated_train.py
Round Commits src/federated_io.py
FedAvg Evaluation src/federated_evaluate.py

Data Research

Historical Data Access src/history_data_v4.py
Coverage Audit src/history_coverage_v4.py
Cohort Builder src/v5_missing_data/build_training_cohorts.py
Missing-Data Training src/v5_missing_data/run_policy_experiments.py
Policy Evaluation src/v5_missing_data/evaluate_policy_models.py
Policy Training Data src/v5_missing_data/training_data.py
Policy Run I/O src/v5_missing_data/training_io.py

Shared Runtime

Forecast Model src/model.py
Forecast Plotting src/forecast_plot.py
Durable I/O src/io_utils.py
Multi-node Data src/multinode.py

Other components

Researcher

Connections 44

Researcher → Pipeline CLI: runs README.md
Pipeline CLI → Dataset Download: downloads data README.md
Pipeline CLI → Window Preparation: prepares data README.md
Pipeline CLI → Training CLI: trains README.md
Pipeline CLI → Forecast Plotting: plots forecasts pipeline.py
Pipeline CLI → Durable I/O: uses pipeline.py
Training CLI → Forecast Model: uses main.py
Training CLI → Durable I/O: uses main.py
Pipeline CLI → Forecast Model: uses pipeline.py
Experiment Runner → Frozen Experiment Data: uses experiments.py
Experiment Runner → Experiment Models: uses experiments.py
Experiment Runner → Experiment Evaluation: uses experiments.py
Experiment Runner → Resource Monitoring: uses experiments.py
Experiment Models → Frozen Experiment Data: uses src/experiment_models.py
Experiment Models → Forecast Model: uses src/experiment_models.py
Experiment Evaluation → Frozen Experiment Data: reads src/experiment_eval.py
Experiment Evaluation → Forecast Plotting: plots src/experiment_eval.py
Experiment Evaluation → Durable I/O: writes src/experiment_eval.py
FedAvg Runner → Frozen Experiment Data: uses federated.py
FedAvg Runner → Round Orchestration: runs rounds federated.py
Round Orchestration → Federated Partitions: uses src/federated_rounds.py
Round Orchestration → Client Updates: uses src/federated_rounds.py
Round Orchestration → Round Commits: commits src/federated_rounds.py
Round Orchestration → Experiment Evaluation: uses src/federated_rounds.py
Round Orchestration → Resource Monitoring: monitors src/federated_rounds.py
Round Orchestration → Forecast Model: initializes src/federated_rounds.py
FedAvg Evaluation → Round Commits: reads src/federated_evaluate.py
FedAvg Evaluation → Round Orchestration: uses src/federated_evaluate.py
Centralized SGD → Federated Partitions: uses src/centralized_sgd.py
Centralized SGD → Round Commits: uses src/centralized_sgd.py
Centralized SGD → Round Orchestration: uses src/centralized_sgd.py
Centralized Evaluation → FedAvg Evaluation: uses src/centralized_evaluate.py
Coverage Audit → Historical Data Access: uses src/history_coverage_v4.py
Historical Data Access → Multi-node Data: uses src/history_data_v4.py
Historical Data Access → Window Preparation: uses src/history_data_v4.py
Cohort Builder → Coverage Audit: uses src/v5_missing_data/build_training_cohorts.py
Cohort Builder → Historical Data Access: uses src/v5_missing_data/build_training_cohorts.py
Missing-Data Training → Policy Training Data: uses src/v5_missing_data/run_policy_experiments.py
Missing-Data Training → Policy Run I/O: uses src/v5_missing_data/run_policy_experiments.py
Missing-Data Training → Policy Evaluation: scores with src/v5_missing_data/run_policy_experiments.py
Policy Evaluation → Policy Training Data: uses src/v5_missing_data/evaluate_policy_models.py
Policy Evaluation → Policy Run I/O: uses src/v5_missing_data/evaluate_policy_models.py
Policy Training Data → Historical Data Access: uses src/v5_missing_data/training_data.py
Policy Run I/O → Durable I/O: uses src/v5_missing_data/training_io.py

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·