tf.argmax tensorflow is achieved with np.argmax numpy, it can give an object index tensor which is a maximum value in one dimension of the data is located, used for calculating Metric (eg acc) of
tf.argmax () function has a parameter axis (axis), calculated in accordance with the parameter can specify which dimension.
As the structure of the matrix, axis may be set to 0 or 1, respectively
0: calculated column by column, 1: Calculation line
The following is a specific example, for intuitive understanding (the same numpy argmax due tf.argmax () usage of the axis, for convenience, as used herein, the numpy.argmax () for example)
1.
When the axis = 0 when , in units of columns, the first column is the maximum value of the index is 83, so that a first value of 3, and so
2. 
Axis. 1 =
When the axis = 1, in units of maximum index of the first row 32, so that the first result is 2
