epoch_ Num and predict_ Conversion of num

for example EPOCH_NUM=18,PREDICT_NUM=20,batch_size=1024,every_n_step=10 ten thousand ,

Number of training samples =1.49 Billion .

EPOCH_NUM Refers to the training data cycle 18 round ,PREDICT_NUM Refers to before forecast 20 A saved model .

Number of training samples *EPOCH_NUM/(batch_size*every_n_step)=checkpoint Number of saved models .

for example , The data above ,every_n_step It is how many steps to save one checkpoint, The model goes one step at a time batch_size, namely 1024 Data has gone 1 Step .batch_size*every_n_step It means to keep one checkpoint How much data will be scanned .