深度学习神经网络训练技巧汇总

I. Loss not changing problems:

Issue: Model predicts one of the 2 (or more) possible classes for all data it sees*

Confirming issue is occurring: Method 1: accuracy for model stays around 0.5 while training (or 1/n where n is number of classes). Method 2: Get the counts of each class in predictions and confirm it’s predicting all one class.

Fixes/Checks (in somewhat of an order):

• Double Check Model Architecture: use model.summary(), inspect the model.
• Check Data Labels: make sure the labelling of your train data hasn’t got mixed up somewhere in the preprocessing etc. (it happens!)
• Check Train Data Feeding Is Randomised: make sure you are not feeding your train data to the model one class at a time. For instance if using ImageDataGenerator().flow_from_directory(PATH), check that param shuffle=Trueand that batch_size is greater than 1.
• Check Pre-Trained Layers Are Not Trainable:** If using a pre-trained model, ensure that any layers that use pre-trained weights are NOT initially trainable. For the first epochs, only the newly added (randomly initialised) layers should be trainable; for layer in pretrained_model.layers: layer.trainable = False should be somewhere in your code.
• Ramp Down Learning Rate: Keep reducing your learning rate by factors of 10 and retrying. Note you will have to fully reinitialize the layers you are trying to train each time you try a new learning rate. (For instance, I had this issue that was only solved once I got down to lr=1e-6, so keep going!

* Normalize the images;

* change of the optimizer, from Adam to SGD;

* Change of learning rate, from 1e-1 to 1e-6, or even more;

标签：layers,技巧,trainable,汇总,your,神经网络,learning,model,Check
来源： https://blog.csdn.net/bmwlwg/article/details/100932882