Evaluation

NeuroInfra helps you quickly run experiments and scale them to large multi-GPU deployments.

Single-GPU
Quick experimentsChoose one model and one dataset task to start fast.
Multi-GPU
Large-scale queuesRun many models, datasets, and tasks across GPUs.
22
Channel systemsReusable EEG channel setups for experiments.
1,047
Channel configurationsChannel Names, Channel Systems, Channel Coordinates, Brain regions.
Single-GPU debugMulti-GPU queues

Single-GPU Debug

Choose a model and dataset, then start a simple, fast experiment.

Choose model type

Pick the model family first, then choose the exact registered asset.

Model type
# Select your experiment configuration.# A runnable command will be generated here automatically.## Example EEGNet debug run:conda run -n <env> python -u run_single_experiment.py \  --model eegnet \  --dataset bcic_iv_2b \  --task motor_imagery_left_right_2class \  --cache_dir artifacts/cache \  --pipeline datasets/pipelines/classical_pipeline_with_z_score.py \  --pipeline_param resample=200 lower_bound=0.1 upper_bound=75 notch=50 \  --split_mode cs \  --fold 4 \  --epochs 1 \  --batch_size 16 \  --channel_system 10-20 \  --output_dir artifacts/debug_eegnet_bcic_iv_2b

Multi-GPU Queues

Queue large-scale multi-model, multi-dataset, and multi-task runs across multiple GPUs.

Search model assets

Filter the model registry before selecting the assets for this queue.

# Select your experiment settings.# Batch selection is supported for model assets and dataset tasks.# Complete the Multi-GPU Queues steps to generate the runnable command and JSON suite data.
Live JSON previewUpdates with the current selections.
{  "preview_state": {    "active_step": "models",    "active_stage": "search",    "ready": false,    "completed": {      "models": false,      "datasets": false,      "splits": false,      "channels": false,      "pipelines": false,      "gpus": false,      "paths": true    },    "estimated_jobs": 0  },  "launch": {    "config": "artifacts/generated/evaluation_suite.json",    "gpus": [],    "processes_per_gpu": 1,    "output_dir": "artifacts/generated_run",    "resume": false,    "retry_failed": false  },  "selection": {    "models": [],    "datasets": [],    "split_modes": [],    "channel_systems": [],    "pipelines": []  },  "suite": {    "common_args": {      "epochs": 1,      "batch_size": 16,      "num_workers": 0,      "seed": 0,      "cache_dir": "artifacts/cache",      "save_checkpoints": false    },    "tasks": [],    "model_overrides": {}  }}
N