3

I know that there are already some similar questions here on stackoverflow but none of them solved my problem. In a python script, I have to train a keras model multiple times and I want to do this with GPU support. Each time I get a bunch of lines in the output console which is disturbing because I can not see the useful information anymore. Here is a part of the output.

2021-04-08 19:43:40.804324: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2021-04-08 19:43:40.804368: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1746] Adding visible gpu devices: 0
2021-04-08 19:43:40.804409: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1159] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-04-08 19:43:40.804418: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1165]      0 
2021-04-08 19:43:40.804424: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1178] 0:   N 
2021-04-08 19:43:40.804495: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1304] Created TensorFlow device (/device:GPU:0 with 1356 MB memory) -> physical GPU (device: 0, name: NVIDIA GeForce MX250, pci bus id: 0000:06:00.0, compute capability: 6.1)
2021-04-08 19:45:27.918402: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0 with properties: 
name: NVIDIA GeForce MX250 major: 6 minor: 1 memoryClockRate(GHz): 1.582
pciBusID: 0000:06:00.0

I tried some methods I found on the internet including the following, but nothing worked for me.

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"

import logging
logging.getLogger("tensorflow").setLevel(logging.ERROR)
logging.getLogger("tensorflow").addHandler(logging.NullHandler(logging.ERROR))

Every suggestion is welcome.

About my setup:

 - python 3.6.8 
 - keras 2.3.1 
 - tensorflow 2.0.0 
 - cudatoolkit 10.0.130
Innat
  • 16,113
  • 6
  • 53
  • 101
wsm
  • 69
  • 1
  • 10
  • i find those log lines by tf useful, maybe u can print some marking to see the values next to it: print("="*80) – Dee Apr 09 '21 at 10:32
  • 1
    Thanks for the suggestion, but I would prefer to get rid of these outputs. – wsm Apr 09 '21 at 13:34

2 Answers2

3

I use this simple approach, works for me.

# tacke future warning, deprecated warning, bla bla 
import warnings

# ------------ tackle some noisy warning
def warn(*args, **kwargs):
    pass
warnings.warn = warn
warnings.simplefilter(action="ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category = DeprecationWarning)

'''
TF_CPP_MIN_LOG_LEVEL = 0 to all logs .
TF_CPP_MIN_LOG_LEVEL = 1 to filter out INFO logs 
TF_CPP_MIN_LOG_LEVEL = 2 to additionall filter out WARNING 
TF_CPP_MIN_LOG_LEVEL = 3 to additionally filter out ERROR.
'''
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

# now import tf 
import tensorflow as tf
Innat
  • 16,113
  • 6
  • 53
  • 101
0

This tutorial in the Tensorflow Documenation discusses using metadata DB(MLMD) APIs to prevent the printing of outputs. This might apply to your application.

Amanda Kimball
  • 21
  • 2
  • I can not find a solution there. Can you please describe what you mean in more detail? – wsm Apr 09 '21 at 13:38