def _time_metric_writer(logdir):
return summary.SummaryWriter(logdir, filename_base="9999999999.time")
def _image_writer(logdir, digest):
timestamp = _next_tfevent_timestamp(logdir)
return summary.SummaryWriter(logdir,
filename_base="%0.10d.image" % timestamp,
filename_suffix="." + digest)
from guild import summary
with summary.SummaryWriter(".") as writer:
writer.add_scalar("x", 1.123, step=1)
writer.add_scalar("x", 2.234, step=2)
writer.add_scalar("x", float("inf"), step=3)
writer.add_scalar("y", float("-inf"), step=1)
writer.add_scalar("y", 1, step=2)
writer.add_scalar("y", 2, step=3)
writer.add_scalar("z", float("nan"), step=1)
def _hparams_writer(logdir):
return summary.SummaryWriter(logdir, filename_base="0000000000.hparams")
import numpy as np
from guild import summary
noise = 0.1
def f(x):
return np.sin(5 * x) * (1 - np.tanh(x**2)) + np.random.randn() * noise
min_loss = None
writer = summary.SummaryWriter(".")
for step, x in enumerate(np.arange(-3.0, 3.0, 0.1)):
loss = f(x)
min_loss = min(loss, min_loss) if min_loss is not None else loss
writer.add_scalar("loss", loss, step)
writer.close()
print("min_loss: %f" % min_loss)
