What is an era in TensorFlow and what is it used for?
https://github.com/tensorflow/tensorflow/blob/754048a0453a04a761e112ae5d99c149eb9910dd/tensorflow/contrib/learn/python/learn/datasets/mnist.py#L173
The era in Machine Learning is the complete processing by the learning algorithm of the entire set of trains.
The MNIST train set consists of 55,000 samples. After the algorithm has processed all these 55,000 samples, the era will pass.
The era is a complete iteration over the samples. Number of eras - how many times the algorithm will be launched. The number of eras directly affects (or not) the outcome of the training phase (with only a few eras you can only achieve a local minimum, but with more eras you can reach a global minimum, or at least a better local minimum).
The era is not an integral element of the TensorFlow structure. It has been used for a long time in machine learning.
One era is a time step that increases every time it passes through all the samples in the training set.
An era is one iteration of learning, so all samples are repeated once in one iteration. When you call the trend function of tensor flows and determine the value for the parameter epochs, you determine how many times your model should be trained with your sample data (usually at least several times).