set the initial θ0​ for the HMC and save the weights and biases

[yich01]

Hi,

Is there any built-in function to save the weights and biases and set the initial θ0​ for the HMC method? I have not found one and suspect that method.make_sampler(log_posterior_fn, self.global_processes.initial_values) uses self.global_processes.initial_values to represent the initial setting for HMC. Is this correct? And I would like to ask, are the weights resulting from all the samples stored in the samples obtained from samples, results = model.run()? Thank you in advance for your response.

[ZongrenZou]

Yes to both questions. You can also adjust the initial values via "variables._initial_values=new_initial_values" for an individual process. Let me know if you have any other questions.

[yich01]

Thank you for the help! I will try it.