import tensorflow as tf
import numpy as np
class Logger(object):
class Loggable(object):
def __init__(self, summary, console_str):
self.summary = summary
self.console_str = console_str
def to_tensorboard(self, filewriter, itr):
filewriter.add_summary(self.summary, global_step=itr)
return self
def to_console(self, itr, append=''):
print('{}: {} {}'.format(itr, self.console_str, append))
class Numpy1DFormatter(object):
def __init__(self, wrapper_str='{}', max_elements=None, precision=3, sep=','):
self._array2string = lambda _arr: wrapper_str.format(
np.array2string(
_arr.flatten()[:max_elements],
precision=precision, separator=sep))
def format(self, arr):
return self._array2string(arr)
def __init__(self):
self._summaries = []
self._summary = None # merged
self._sess = None
self._console_format_strs = []
self._console_tensors = []
self._final = False
def add_summaries(self, summaries):
assert isinstance(summaries, list)
self._summaries.extend(summaries)
def add_console_tensor(self, formatter, tensor):
"""
:param formatter: object responding to -format(), e.g., a string with {}, or Numpy1DFormatter
:param tensor:
:return:
"""
self._console_format_strs.append(formatter)
self._console_tensors.append(tensor)
def finalize_with_sess(self, sess):
assert not self._final
self._final = True
self._sess = sess
assert len(self._console_format_strs) == len(self._console_tensors)
self._summary = tf.summary.merge(self._summaries)
def log(self):
summary, console_tensor = self._sess.run([self._summary, self._console_tensors])
return Logger.Loggable(
summary=summary,
console_str=', '.join(
format_str.format(tensor)
for format_str, tensor in zip(self._console_format_strs, console_tensor)))
