feat: Add Hydra configuration system for ML experiments

README_HYDRA.md

@@ -0,0 +1,96 @@
+# Hydra Configuration System - Quick Start
+
+## 🚀 Installation
+
+```bash
+# Install pip and create virtual environment
+sudo apt install python3-pip python3.12-venv -y
+python3 -m venv venv
+source venv/bin/activate
+
+# Install minimal requirements for Hydra
+pip install -r requirements-minimal.txt
+```
+
+## 📋 Usage Examples
+
+### Basic Configuration
+```bash
+# Use default configuration
+python test_config.py
+
+# Override parameters
+python test_config.py experiments.training.lr=0.001 experiments.model.hidden_dims=[128,64]
+```
+
+### Pre-configured Experiments
+```bash
+# Debug experiment (small model, 10 epochs)
+python test_config.py --config-name experiments/debug
+
+# Baseline experiment (standard settings)
+python test_config.py --config-name experiments/baseline
+```
+
+### Command-line Overrides
+```bash
+# Learning rate override
+python test_config.py experiments.training.lr=0.001
+
+# Model architecture override
+python test_config.py experiments.model.hidden_dims=[32]
+
+# Multiple overrides
+python test_config.py experiments.training.lr=0.001 experiments.model.hidden_dims=[128,64] experiments.training.epochs=300
+```
+
+## 📁 Configuration Structure
+
+```
+configs/
+├── config.yaml                    # Main configuration
+├── model/gcn.yaml                 # GCN model configs
+├── training/default.yaml          # Training configs
+├── data/cora.yaml                 # Dataset configs
+└── experiments/
+    ├── debug.yaml                 # Debug experiment
+    ├── baseline.yaml              # Baseline experiment
+    └── hyperparameter_search.yaml # Parameter sweep
+```
+
+## 🎯 Key Features
+
+- **Easy Parameter Override**: `experiments.training.lr=0.001`
+- **Pre-configured Experiments**: Debug, baseline, hyperparameter search
+- **Automatic Output Organization**: Results saved to timestamped directories
+- **Configuration Tracking**: Full config saved with results
+
+## 📊 Test Results
+
+The system successfully demonstrates:
+- ✅ Configuration loading and merging
+- ✅ Command-line parameter overrides
+- ✅ Pre-configured experiment selection
+- ✅ Automatic output organization
+- ✅ Configuration tracking and logging
+
+## 🔄 Next Steps
+
+1. Install full PyTorch dependencies when needed:
+   ```bash
+   pip install -r requirements.txt  # May take time due to large downloads
+   ```
+
+2. Use the full training script:
+   ```bash
+   python train.py experiments.training.lr=0.001
+   ```
+
+3. Run hyperparameter sweeps:
+   ```bash
+   python train.py --multirun experiments.training.lr=0.001,0.01,0.1
+   ```
+
+---
+
+**The Hydra configuration system is ready for easy experiment management!** 🚀

configs/config.yaml

@@ -0,0 +1,22 @@
+# Default configuration for AstroML training experiments
+defaults:
+  - model: gcn
+  - training: default
+  - data: cora
+  - _self_
+
+# Experiment settings
+experiment:
+  name: "astroml_experiment"
+  seed: 42
+  device: "auto"  # auto, cpu, cuda
+  save_dir: "outputs"
+  log_level: "INFO"
+
+# Hydra settings
+hydra:
+  run:
+    dir: outputs/${experiment.name}/${now:%Y-%m-%d_%H-%M-%S}
+  sweep:
+    dir: outputs/${experiment.name}/multirun
+    subdir: ${hydra.job.override_dirname}

configs/data/cora.yaml

@@ -0,0 +1,23 @@
+# Cora dataset configuration
+_target_: torch_geometric.datasets.Planetoid
+
+name: "Cora"
+root: "data"
+transform:
+  _target_: torch_geometric.transforms.NormalizeFeatures
+
+# Dataset-specific settings
+num_classes: 7
+num_node_features: 1433
+
+# Alternative datasets
+variants:
+  citeseer:
+    name: "CiteSeer"
+    num_classes: 6
+    num_node_features: 3703
+
+  pubmed:
+    name: "PubMed"
+    num_classes: 3
+    num_node_features: 5003

configs/experiments/baseline.yaml

@@ -0,0 +1,26 @@
+# Baseline experiment configuration
+defaults:
+  - /model: gcn
+  - /training: default
+  - /data: cora
+  - _self_
+
+# Baseline settings
+experiment:
+  name: "baseline_experiment"
+  seed: 42
+
+# Model settings
+model:
+  hidden_dims: [64, 32]
+  dropout: 0.5
+
+# Training settings
+training:
+  epochs: 200
+  lr: 0.01
+  weight_decay: 5e-4
+
+# Data settings
+data:
+  name: "Cora"

configs/experiments/debug.yaml

@@ -0,0 +1,29 @@
+# Debug experiment configuration
+defaults:
+  - /model: gcn
+  - /training: default
+  - /data: cora
+  - _self_
+
+# Debug settings
+experiment:
+  name: "debug_experiment"
+  seed: 42
+  device: "auto"
+  save_dir: "outputs"
+  log_level: "INFO"
+
+# Model overrides
+model:
+  hidden_dims: [16]  # Small model for fast debugging
+  dropout: 0.1
+
+# Training overrides
+training:
+  epochs: 10  # Short training for debugging
+  lr: 0.01
+  log_interval: 1
+
+# Data overrides
+data:
+  name: "Cora"

configs/experiments/hyperparameter_search.yaml

@@ -0,0 +1,37 @@
+# Hyperparameter search experiment configuration
+defaults:
+  - /model: gcn
+  - /training: default
+  - /data: cora
+  - _self_
+
+# Hyperparameter search settings
+experiment:
+  name: "hyperparameter_search"
+  seed: 42
+
+# Model parameter grid
+model:
+  hidden_dims:
+    - [32]
+    - [64, 32]
+    - [128, 64]
+  dropout:
+    - 0.2
+    - 0.5
+    - 0.7
+
+# Training parameter grid
+training:
+  lr:
+    - 0.001
+    - 0.01
+    - 0.1
+  weight_decay:
+    - 1e-5
+    - 5e-4
+    - 1e-3
+
+# Data settings
+data:
+  name: "Cora"

configs/model/gcn.yaml

@@ -0,0 +1,28 @@
+# Graph Convolutional Network model configuration
+_target_: astroml.models.gcn.GCN
+
+input_dim: ???
+hidden_dims: [64, 32]
+output_dim: ???
+dropout: 0.5
+activation: "relu"
+batch_norm: false
+residual: false
+
+# Alternative model configurations
+variants:
+  small:
+    hidden_dims: [32]
+    dropout: 0.2
+
+  medium:
+    hidden_dims: [64, 32]
+    dropout: 0.5
+
+  large:
+    hidden_dims: [128, 64, 32]
+    dropout: 0.6
+
+  very_large:
+    hidden_dims: [256, 128, 64]
+    dropout: 0.7

configs/training/default.yaml

@@ -0,0 +1,38 @@
+# Training configuration
+epochs: 200
+lr: 0.01
+weight_decay: 5e-4
+optimizer: "adam"
+scheduler: null
+early_stopping:
+  patience: 50
+  min_delta: 1e-4
+  monitor: "val_loss"
+  mode: "min"
+
+# Training settings
+batch_size: null  # Full batch for graph data
+val_split: 0.1
+test_split: 0.1
+shuffle: true
+
+# Logging
+log_interval: 20
+save_best_only: true
+save_last: true
+
+# Optimizer configurations
+optimizer_configs:
+  adam:
+    betas: [0.9, 0.999]
+    eps: 1e-8
+    amsgrad: false
+
+  sgd:
+    momentum: 0.9
+    nesterov: true
+
+  adamw:
+    betas: [0.9, 0.999]
+    eps: 1e-8
+    weight_decay: 1e-2

docs/_build/.buildinfo

@@ -0,0 +1,4 @@
+# Sphinx build info version 1
+# This file records the configuration used when building these files. When it is not found, a full rebuild will be done.
+config: c77cbac2cb1078e76006b15750c98cfd
+tags: 645f666f9bcd5a90fca523b33c5a78b7