运行非局部循环网络GitHub代码遇KeyError: 'INFO'，求调试方法

Let's break down this error and work through debugging and fixes step by step.

Understanding the Error #

The KeyError: 'INFO' happens because your code is trying to use the string 'INFO' as a key to access tf.logging.__dict__, but this approach has one of two issues:

• Either your TensorFlow version no longer exposes log level names like 'INFO' in tf.logging.__dict__ (common in TF 2.x where tf.logging is deprecated),
• Or the logic is flawed to begin with—TensorFlow's log levels (like INFO, DEBUG) are numeric constants, not string keys meant to index the module's dictionary directly.

Debugging Steps #

First, let's gather more context to pinpoint the exact issue:

• Check TensorFlow logging module status
  Add this line right before the error-causing code to see what attributes are available in tf.logging:

  print("Available attributes in tf.logging:", dir(tf.logging))
  

  If 'INFO' doesn't show up in the output, you're likely using TF 2.x and need to use the compatibility version of logging.

• Inspect hparams.verbosity
  Verify the type and value of the verbosity parameter with this print statement:

  print(f"hparams.verbosity: type={type(hparams.verbosity)}, value={hparams.verbosity}")
  

  You’ll almost certainly find it’s a string ('INFO') instead of the numeric constant tf.logging.INFO (which equals 10).

• List all keys in tf.logging.__dict__
  Confirm exactly what keys exist in the module's dictionary:

  print("Keys in tf.logging.__dict__:", list(tf.logging.__dict__.keys()))
  

  If 'INFO' is missing here, that’s the root cause of the KeyError.

Fix Solutions #

Here are several actionable fixes depending on your setup:

1. Map String Verbosity to TensorFlow Numeric Constants #

Create an explicit mapping to avoid direct dictionary indexing:

# Define a mapping of string names to TF log level constants
verbosity_mapping = {
    'DEBUG': tf.logging.DEBUG,
    'INFO': tf.logging.INFO,
    'WARN': tf.logging.WARN,
    'ERROR': tf.logging.ERROR,
    'FATAL': tf.logging.FATAL
}
# Replace the error line with this
verbosity_level = verbosity_mapping[hparams.verbosity] / 10

For TF 2.x, replace tf.logging with tf.compat.v1.logging to access the deprecated logging module.

2. Use Numeric Values for Verbosity Configuration #

If you control the hparams setup, configure verbosity with numeric values instead of strings:

• Set verbosity=10 for INFO, verbosity=20 for WARN, etc.
• Then your code can simplify to:

  verbosity_level = hparams.verbosity / 10
  

3. Switch to Python's Standard logging Module #

For a more stable, TF-version-agnostic solution, use Python's built-in logging:

import logging
# Map string names to standard logging levels
verbosity_mapping = {
    'DEBUG': logging.DEBUG,
    'INFO': logging.INFO,
    'WARNING': logging.WARNING,
    'ERROR': logging.ERROR,
    'CRITICAL': logging.CRITICAL
}
verbosity_level = verbosity_mapping[hparams.verbosity] / 10

4. Compatibility Mode for TF 2.x #

If you must retain TF's logging for legacy code, enable v1 compatibility:

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
# Now use tf.logging as you would in TF 1.x
verbosity_level = tf.logging.__dict__[hparams.verbosity] / 10

Note: This is a temporary fix—long-term, migrating to standard logging is better.

内容的提问来源于stack exchange，提问作者LathaH Narayan