bebesparkelsparkel / to-precision Goto Github PK

William,

I'm not going to drag over-bars into the scope of the already broad PR we're working on, but it turns out that there is a "COMBINING_OVERLINE" Unicode character (\u0305) that places an overline over the character it follows. One would have to deal with Python2 vs. 3 Unicode issues, and it wouldn't be useful to people unless the output device their data was headed for supported it (so it would have to be a unique mode option), but at least it is possible.

>>> print("130\u03050")
130̅0
>>>

Food for thought :)

Add this to PyPi. As far as I can tell There is nothing like this that exists on there. It will make it very easy for users to pull your module.

Here is a simple guide.

Hi William,

I was delighted to find to_precision (https://github.com/BebeSparkelSparkel/to-precision) today. Thanks for sharing it.

I hope this isn't a FAQ but I wondered if better behaviour for notation='auto' wouldn't be to restrict standard notation to, say, 0.001 < abs(value) < 1000 so that very small absolute values are returned using scientific notation?

Thanks again,

John

hypothesis.readthedocs.io

This looks very useful. However, is it true that it cannot handle NaNs?

---> 40 return converter(value, precision, filler)
41
42

/Volumes/Apps_and_Docs/JRR_Utils/anaconda/lib/python2.7/site-packages/to_precision.pyc in sci_notation(value, precision, filler)
82 [email protected]
83 '''
---> 84 is_neg, sig_digits, dot_power, ten_power = _sci_notation(value, precision)
85
86 return ('-' if is_neg else '') + _place_dot(sig_digits, dot_power) + filler + str(ten_power)

/Volumes/Apps_and_Docs/JRR_Utils/anaconda/lib/python2.7/site-packages/to_precision.pyc in _sci_notation(value, precision)
123 [email protected]
124 '''
--> 125 sig_digits, power, is_neg = _number_profile(value, precision)
126
127 dot_power = -(precision - 1)

/Volumes/Apps_and_Docs/JRR_Utils/anaconda/lib/python2.7/site-packages/to_precision.pyc in _number_profile(value, precision)
191
192 power = -1 * math.floor(math.log10(value)) + precision - 1
--> 193 sig_digits = str(int(round(abs(value) * 10.0**power)))
194
195 return sig_digits, int(-power), is_neg

ValueError: cannot convert float NaN to integer

This is a really nice library. I think I've found a minor bug:

>>> std_notation(0.989999, 3)
'0.990' # good
>>> std_notation(0.999999, 3)
'1.000' # extra sig fig
>>> std_notation(1.0, 3)
'1.00' # good
>>> std_notation(1.00000, 3)
'1.00' # good
>>> std_notation(0.099999, 3)
'0.1000' # extra sig fig
>>> std_notation(0.100000099999, 3)
'0.100' # good

Values less than 1 that get rounded up appear to be given an extra digit of precision. This seems to be because the calculation for 'power' in _number_profile() yields a value that is too big by 1 in these cases.

This seems to have other consequences as well:

>>> sci_notation(1.00000,3)
'1.00e0'
>>> sci_notation(0.9999,3)
'10.00e-1'

The same issue seems to exist symmetrically for negative numbers. I don't have a mathematically elegant solution, but conditionally decrementing power before determining sig_digits at the end of _number_profile() seems to work, e.g.:

    power = -1 * floor(log10(value)) + precision - 1
    if abs(value) < 1 and (floor(log10(int(round(abs(value) * 10.0**power)))) > floor(log10(int(abs(value) * 10.0**power)))):
        power -= 1
    sig_digits = str(int(round(abs(value) * 10.0**power)))

Cheers,
Tom