【Python坑系列】

为了节约空间，使用numpy数组时候采用了float16，结果发现精度远远不达标

默认的floar64是最接近原本浮点数的

a

Out[206]:

array([ 0.00010002,0.00010002,0.00010002],dtype=float16)

a = 0.0001 * np.ones(10,np.float32)
a

Out[208]:

array([  9.99999975e-05,9.99999975e-05,9.99999975e-05],dtype=float32)
a = 0.0001 * np.ones(10,np.float64)
a

Out[210]:

array([ 0.0001,0.0001,0.0001])
a = 0.9999 * np.ones(10,np.float64)
a

Out[212]:

array([ 0.9999,0.9999,0.9999])