Easy TensorFlow Logging

Note: This is mainly designed for TensorFlow 1; the logging API in TensorFlow 2 is significantly easier than in TensorFlow 1. This module is compatible with TensorFlow 2, but some features may not work with eager execution.

Are you prototyping something and want to be able to magically graph some value without going through all the usual steps to set up TensorFlow logging properly?

easy_tf_log is a simple module to do just that.

from easy_tf_log import tflog

then you can do

for i in range(10):
    tflog('really_interesting_variable_name', i)

and you'll find a directory logs that you can point TensorBoard to

$ tensorboard --logdir logs

to get

See demo.py for a full demo.

Based on logging code from OpenAI's baselines.

Installation

pip install easy-tf-log

Note that TensorFlow must be installed separately.

Documentation

easy-tf-log supports logging using either a global logger or an instantiated logger object.

The global logger is good for very quick prototypes, but for anything more complicated, you'll probably want to instantiate your own Logger object.

Global logger

• easy_tf_log.tflog(key, value, step=None)
  • Logs value (int or float) under the name key (string).
  • step (int) sets the step associated with value explicitly. If not specified, the step will increment on each call.
• easy_tf_log.set_dir(log_dir)
  • Sets the global logger to log to the specified directory.
  • log_dir can be an absolute or a relative path.
• easy_tf_log.set_writer(writer)
  • Sets the global logger to log using the specified tf.summary.FileWriter instance.

By default (i.e. if set_dir is not called), the global logger logs to a logs directory automatically created in the working directory.

Logger object

• logger = easy_tf_log.Logger(log_dir=None, writer=None)
  • Create a Logger.
  • log_dir: an absolute of relative path specifying the directory to log to.
  • writer: an existing tf.summary.FileWriter instance to use for logging.
  • If neither log_dir nor writer are specified, the logger will log to a logs directory in the working directory. If both are specified, the constructor will raise a ValueError.
• logger.log_key_value(key, value, step=None)
  • See tflog.
• logger.log_list_stats(key, values_list)
  • Log the minimum, maximum, mean, and standard deviation of values_list (a list of ints or floats).
• logger.measure_rate(key, value)
  • Log the rate at which value (int or float) changes per second.
  • The first call internally stores the time of the first value; the second call logs the change between the second value and the first value divided by the time between the calls; etc.
• logger.set_dir(log_dir)
  • See easy_tf_log.set_dir(log_dir).
• logger.set_writer(writer)
  • See easy_tf_log.set_writer(writer).
• logger.close()
  • Flush logs and close the log file handle.