def model_sample(cll=None):
# wrap params for use in model -- required
# decoder = pyro.module("decoder", pt_decode)
# sample from prior
z_mu, z_sigma = Variable(torch.zeros(
[1, 20])), Variable(torch.ones([1, 20]))
# sample
z = pyro.sample("latent", DiagNormal(z_mu, z_sigma))
alpha = Variable(torch.ones([1, 10]) / 10.)
if cll.data.cpu().numpy() is None:
bb()
cll = pyro.sample('class', Categorical(alpha))
print('sampling class')
# decode into size of imgx1 for mu
img_mu = pt_decode.forward(z, cll)
# bb()
# img=Bernoulli(img_mu).sample()
# score against actual images
img = pyro.sample("sample", Bernoulli(img_mu))
# return img
return img, img_mu
def run_pyro_nuts(data, pfile, n_samples, params):
# import model, transformed_data functions (if exists) from pyro module
model = import_by_string(pfile + ".model")
assert model is not None, "model couldn't be imported"
transformed_data = import_by_string(pfile + ".transformed_data")
if transformed_data is not None:
transformed_data(data)
nuts_kernel = NUTS(model, step_size=0.0855)
mcmc_run = MCMC(nuts_kernel,
num_samples=n_samples,
warmup_steps=int(n_samples / 2))
posteriors = {k: [] for k in params}
for trace, _ in mcmc_run._traces(data, params):
for k in posteriors:
posteriors[k].append(trace.nodes[k]['value'])
#posteriors["sigma"] = list(map(torch.exp, posteriors["log_sigma"]))
#del posteriors["log_sigma"]
posterior_means = {
k: torch.mean(torch.stack(posteriors[k]), 0)
for k in posteriors
}
bb()
return posterior_means
def render_update(full_geo, viewport, print_render=False):
# get our init state with some info
init_state, geo_states = full_geo[0], full_geo[1:]
img_size = init_state["img_size"]
start_viewport = init_state["viewport"]
vine_json = {
"depth": init_state["depth"],
"pos": {"x": init_state["pos"].x.item(), "y": init_state["pos"].y.item()},
"width": init_state["width"].item(),
"angle": init_state["angle"].item(),
"img_size": list(init_state['img_size']),
"viewport": {"xmin": start_viewport.xmin, "xmax": start_viewport.xmax,
"ymin": start_viewport.ymin, "ymax": start_viewport.ymax},
"vines": [geo_to_json(state) for state in geo_states]
}
# bb()
vv = PCGVineView()
ss = ProtoScreenshot()
from json import dumps
# parse our json into the actual object
Parse(dumps(vine_json), vv)
queue_name = 'task_queue'
print("Getting class information from Unity")
client = BasicClient(queue_name).connect()
# from IPython import embed
# embed()
cs_response = client.sync_call(client.get_message(vv, ss))
bb()
def main(args):
policies = init(args.num_players)
model(policies)
bb()
lr = 1e-2
inference = SVI(model, guide, Adam({"lr": lr}), 'ELBO')
for i in range(args.num_samples):
loss = inference.step()
print("loss = {}".format(loss))
print(pyro.get_param_store()._params)
def compare_models(data, sfile, pfile, n_samples=2000, model_cache=None):
validate_data_def, init_params, model, transformed_data = get_fns_pyro(
pfile)
s_site_means = run_stan_nuts(data,
sfile,
n_samples=n_samples,
model_cache=model_cache)
# breakpoint
initialized_params = {}
tensorize_data(data)
transformed_data(data)
init_params(data, initialized_params)
p_site_means = run_pyro_advi(data, validate_data_def, initialized_params,
model)
#TODO: check that these means are close to each other for each parameter
bb()
self.connection.process_data_events()
return self.msg_wrap.decode_to_proto(self.response)
if __name__ == '__main__':
import sys
queue_name = 'task_queue'
print("Getting class information from Unity")
client = BasicClient(queue_name).connect()
# lets say hello
print("Class info achieved, saying hello")
hello_friend = ProtoHello()
hello_friend.proto_message = 'hi friend from python'
# empty call to serve, wait for response
cs_response = client.sync_call(client.get_message(hello_friend))
print("simple call and respond {}".format(cs_response))
ss_for_me = ProtoScreenShot()
ss_for_me.width = 32
ss_for_me.height = 32
ss_response = client.sync_call(client.get_message(ss_for_me))
np_ss = screenshot_to_np(ss_response[0][1])
bb()
