krzentner

from dataclasses import dataclass
import numpy as np


def bayesian_update(x: np.ndarray,
                    k: int, mu: float, alpha: float, beta: float):
    """
    Graphical model is like this:
    tau ~ IGa(alpha, beta)
    x_i ~ N(mu, tau)

krzentner

{
  "$type": "garage.experiment.local_runner.LocalRunner",
  "_algo": {
    "$type": "garage.torch.algos.pearl.PEARL",
    "_batch_size": 256,
    "_context_replay_buffers": {
      "0": {
        "$type": "garage.replay_buffer.path_buffer.PathBuffer",
        "_buffer": {},
        "_capacity": 1000000,

krzentner

class PickledCall:
    def __new__(cls, to_call, args=(), kwargs=None, __pickle_target=None):
        if __pickle_target is None:
            result = super(cls, PickledCall).__new__(cls)
            result.to_call = to_call
            result.args = args
            result.kwargs = kwargs or {}
            return result
        else:
            return __pickle_target(*args, **kwargs)

krzentner

#!/usr/bin/env python3
import copy
import types

def configure_with(config_type):
    '''
    Attach a configuration type to the provided type.
    This will allow calling instances of the configuration type to construct
    instances of the provided type, by copying fields from the configuration
    object.