Configuration System¶

flyvis uses Hydra for configuration management of network and training. The configuration system is organized hierarchically, allowing for modular configuration of different components.

Configuration Components¶

Network Configuration¶

network/: Core network architecture and parameters
connectome/: Neural connectivity specification
dynamics/: Neural dynamics parameters
edge_config/: Synapse-related configurations (sign, strength, count)
node_config/: Neuron-related configurations (bias, time constants)
stimulus_config/: Input stimulus parameters

Training Configuration¶

optim/: Optimization parameters
penalizer/: Loss function penalties
scheduler/: Learning rate scheduling
task/: Task-specific settings and parameters

Top-level Configuration¶

solver.yaml: Training loop parameters

Default Configuration Structure¶

The default configuration is structured as follows:

config
├── network                         # Network configuration
│   ├── connectome                  # Connectome configurations
│   │   └── connectome.yaml         # Default connectome configuration
│   ├── dynamics                    # Dynamics configurations
│   │   └── dynamics.yaml           # Default dynamics configuration
│   ├── edge_config                 # Edge configuration
│   │   ├── edge_config.yaml        # Default edge configuration
│   │   ├── sign                    # Signaling configurations
│   │   │   └── sign.yaml           # Default signaling configuration
│   │   ├── syn_count               # Synaptic count configurations
│   │   │   └── syn_count.yaml      # Default synaptic count configuration
│   │   └── syn_strength            # Synaptic strength configurations
│   │       └── syn_strength.yaml   # Default synaptic strength configuration
│   ├── network.yaml                # Default network configuration
│   ├── node_config                 # Node configuration
│   │   ├── bias                    # Bias configurations
│   │   │   └── bias.yaml           # Default bias configuration
│   │   ├── node_config.yaml        # Default node configuration
│   │   └── time_const              # Time constant configurations
│   │       └── time_const.yaml     # Default time constant configuration
│   └── stimulus_config             # Stimulus configuration
│       └── stimulus_config.yaml    # Default stimulus configuration
├── optim                           # Optimizer configuration
│   └── optim.yaml                  # Default optimizer configuration
├── penalizer                       # Penalizer configuration
│   └── penalizer.yaml              # Default penalizer configuration
├── scheduler                       # Scheduler configuration
│   └── scheduler.yaml              # Default scheduler configuration
├── solver.yaml                     # Default solver configuration
└── task                            # Task configuration
    └── task.yaml                   # Default task configuration

15 directories, 16 files

Customizing Configurations¶

Overriding Default Parameters¶

For quick parameter changes you can use hydra command-line overrides.

# Example: Change the scale of the synaptic strength to 0.5
flyvis train-single network.edge_config.syn_strength.scale=0.5 ensemble_and_network_id=0042/007 task_name=flow

# Example: Add a custom parameter to an existing config
flyvis train-single +network.edge_config.syn_strength.custom_param=100 ensemble_and_network_id=0042/007 task_name=flow

More information on hydra command-line overrides can be found here.

Using Custom Configurations¶

In many situations you will want to create custom configuration files, to either break the existing config structure for new code or to keep track of different sets of parameters.

This particularly applies when installing flyvis as a package instead of using the developer mode, where one can directly modify the source code configuration.

To create and maintain your own configuration files proceed as follows:

Initialize a local config directory:

# Copy the default config structure into flyvis_config
flyvis init-config

Create your custom configurations in the generated flyvis_config directory.
Use your custom configs:
```
# Using the full custom config directory
flyvis train-single --config-path $(pwd)/flyvis_config [other options]

# Using a specific custom config
flyvis train-single network/edge_config/syn_strength=custom_syn_strength --config-dir $(pwd)/flyvis_config
```
Notice that config-path overrides the entire default config directory, while config-dir only adds another search path for resolving configs.

Important: It is recommended to use different file names than the defaults. This is because for partial usage with config-dir hydra resolves the config names in the default config directory first. I.e., your changes reflected in files of the same name in the custom config directory might not have the intended effect. E.g., name your config file custom_syn_strength.yaml instead of syn_strength.yaml.

More information on hydra configuration can be found here.