notebooks/simulation_experiments.ipynb
- explore a FNP pretrained on simulated simple/complex cellsnotebooks/visual_experiments.ipynb
- measure performance of a FNP pretrained on real responses
PYTHONPATH="." python tuning_manifold/train.py \
--architecture='{"architecture": [[11, 32], [5, 8], [3, 8], [3, 8], 64], "padding": "same", "norm": "batch", "nb_orientations": 8}' \
--cell_latent_dim=64 --synthetic_rf_dim=8 --image_shape="(16,16,1)"
The main files and directories for this project are
tuning_manifold
-- the python project directorytuning_manifold/train.py
-- the training scripttuning_manifold/fnp_model.py
-- the Factorized Neural Process modeltuning_manifold/synthetic_sampler.py
-- generate TF Datasets of simulated responsestuning_manifold/se2cnn
-- the G-CNN library