For Sage, `1/5`

is a rational number, whereas `1/5.`

is a floating point number:

```
sage: (1/5).parent()
Rational Field
sage: (1/5.).parent()
Real Field with 53 bits of precision
```

Now, when you put them in a numpy array, they are converted to a type named `numpy.float64`

:

```
sage: type(numpy.array([1/5])[0])
<type 'numpy.float64'>
sage: type(numpy.array([1/5.])[0])
<type 'numpy.float64'>
```

As you can see, this conversion do not lead to the same approximation for `1/5`

and `1/5.`

, they seem not rounded to the same direction:

```
sage: numpy.float64(1/5)
0.19999999999999998
sage: numpy.float64(1/5.)
0.20000000000000001
sage: numpy.float64(0.2)
0.20000000000000001
```

This explains your problem. There are various issues about the rounding direction in Sage and the tools its uses, and i agree there should be a general audit.

Now, if you want 1/5 to be rounded in the right direction, you can first round it using `RealField()`

which seems more consistent than `numpy.float64`

:

```
sage: numpy.array([RR(1/5)])[0] == 0.2
True
```