test.md

first launch

modification on 'train_ista.py'

# from models.predictive_coding_single import DynPredNet as SingleNet
...
if two_level:
    model = DynPredNet(params, device).to(device)
    optimizer = [
        optim.SGD(model.spatial_decoder.parameters(), params.learning_rate_s),
        optim.Adam([model.temporal], params.learning_rate_t),
        optim.Adam(model.hypernet.parameters(), params.learning_rate_t)
    ]
else:
    print("ERROR")
    # model = SingleNet(params, device).to(device)
    # optimizer = [
    #     optim.SGD(model.spatial_decoder.parameters(), params.learning_rate_s),
    #     optim.Adam(model.temporal.parameters(), params.learning_rate_t),
    # ]

experiments/two_mnist/params.json

"seed": 55,
"model": "two",

first run

python train_ista.py --model_dir experiments/two_mnist --data_dir data/mnist

analys mnist_motion_cluster.py

python mnist_motion_cluster.py --model_dir experiments/two_mnist --data_dir data/mnist