The steps are as follows:
1. Record the training log
Add a line of parameters to the command during training to record the Log.
GLOG _ logtostderr = 0 GLOG _ log _ directory =/home/Liu Yun/coffee/model /AAA/A 12/Log/ \
/home/ Liu Yun/caffe/build/tools/caffe train-solverexamples/AAA/solver.prototxt-weights. /models/bvlc _ reference _ caffe net/bvlc _ reference _ caffe net . caffe model
Change the directory to the directory of your own system, so that after the training, the Log of each training will be generated in the log folder.
2. Analyze the training log
Copy the first three script files to the log folder and execute:
. /parse _ log . sh caffe . Liu yun-860-088 cn . root . log . info . 20 160830-090533.5367
This will generate a. train file and a. test file in the current folder.
Generate pictures
Execute:
. /plot_training_log.py.example 6? Train _ loss.pngcafe.liu yun-860-088cn.root.log Note: caffe.liu yun-860-088cn.root.log.info.20160830-090533.5367 must be changed to caffe.liu.
Can you lose training during the training process? Vs. Iters curve, where 6 represents the curve type and train_loss.png represents the name of the saved picture.
Caffe supports a variety of curve drawing. You can specify different types of parameters. The specific parameters are as follows:
Remarks:?
1. supports multiple logs. ?
2. The log file name must end in lowercase. ".log". ?
Supported chart types:?
0: Test accuracy? vs. Iters?
1: test accuracy? Vs seconds?
2: Test loss? vs. Iters?
3: Test loss? Vs seconds?
4. Training learning rate? vs. Iters?
5. Training learning rate? Vs seconds?
6: train loss? vs. Iters?
7: The train lost contact? Vs seconds?
Finally, look at the effect: