Simulated Data from Compensation of Hyperexcitability with Simulation-Based Inference

File description:amortized_samples.npz : Simulator parameters (theta), drawn from the prior distribution, and their corresponding simulator output (x). 12,930,000 simulations in total. These can be used to train an amortized density estimator.amortized_estimator.pickle : A trained amortized density estimator trained with the sbi Python package. Also includes the simulations that were used to train it. See sbi docs for details.truncated_sequential_estimator_baseline.pickle : Sequentially trained neuronal density estimators to estimate the parameters given the baseline condition. Also contains the amortized and the sequential samples used to train it. The estimator is trained after each iteration.truncated_sequential_estimator_hyperexctiable.pickle : Sequentially trained neuronal density estimators to estimate the parameters given the hyperexcitable condition. Also contains the amortized and the sequential samples used to train it. The estimator is trained after each iteration.

Usage:

To load amortized_samples.npz:

import numpy as np
data = np.load('amortized_samples.npz')
x, theta = data['x'], data['theta']

To load the amortized density estimator:

import pickle
with open('amortized_estimator', "rb") as f:
amortized_estimator = pickle.load(f)

To load the sequential density estimators:

import pickle
def loadall(filename):
with open(filename, "rb") as f:
while True:
try:
yield pickle.load(f)
except EOFError:
break

sequential_estimators = loadall('truncated_sequential_estimator_baseline.pickle')

NOTE: Pickling and unpickling of estimators is highly sensitive to the version of sbi used. The estimators in this repository were trained with a 0.22 version of sbi. Let me know if you encounter problems and I can give you an sbi version I keep for compatibility.