CODEWITHSUNDEEP
pythonnumpyvectorizationstring-formatting
Mike T
Mike T

Reputation: 43742

Format numeric array to string-like array

I'd like to have Numpy efficiently convert each element of a numeric array (e.g. float32) to a formatted array (i.e. string-like). I can make this work as I expect by iterating each element to a list:

import numpy as np
a = (10 ** np.arange(-5, 6, 2, dtype='d') * 3.14159).astype('f')
# array([3.14159e-05, 3.14159e-03, 3.14159e-01, 3.14159e+01, 3.14159e+03,
#        3.14159e+05], dtype=float32)

# Good conversion to a list
print([str(x) for x in a])
# ['3.14159e-05', '0.00314159', '0.314159', '31.4159', '3141.59', '314159.0']
print(list(map(lambda x: str(x), a)))  # also does the same

# Expected result: a string-like Numpy array
print(repr(np.array([str(x) for x in a])))
# array(['3.14159e-05', '0.00314159', '0.314159', '31.4159', '3141.59',
#        '314159.0'], dtype='<U11')

However, this example doesn't easily scale to multidimensional arrays, since map() or list comprehensions don't understand how additional dimensions work. I'd like a result provided as a Numpy array with a string-like datatype, as shown above.

Typically, numpy.vectorize could be used to do this, however each of my attempts with Numpy 1.15 do not return the expected result:

# Bad conversions with np.vectorize, all show the same result
f = np.vectorize(lambda x: str(x))
f = np.vectorize('%s'.__mod__)  # equivalent; gives same result
f = np.vectorize(lambda x: '{!s}'.format(x))  # also same, but modern formatter
print(f(a))
# array(['3.141590059385635e-05', '0.003141589928418398',
#        '0.31415900588035583', '31.4158992767334', '3141.590087890625',
#        '314159.0'], dtype='<U21')

(The reason why these results are bad is that it appears that Numpy upgraded the datatype from float32 to Python's native double precision; similar to [str(x) for x in a.tolist()])

Any ideas on how to either use map()/list comprehensions on arbitrary dimension Numpy arrays and/or fix np.vectorize to achieve an equivalent result?

Upvotes: 0

Views: 865

Answers (2)

kuzand
kuzand

Reputation: 9806

You could simply use astype with dtype 'str'

a.astype(dtype=str)

# array(['3.14159e-05', '0.00314159', '0.314159', '31.4159', '3141.59',
#       '314159.0'], dtype='<U32')

Edit: just saw your comment that you have figured it out by yourself. Nevertheless I will keep my answer.

Upvotes: 0

ZisIsNotZis
ZisIsNotZis

Reputation: 1750

How about np.char.mod?

import numpy as np
np.char.mod('%.2f', np.random.rand(8, 8))

It outputs

array([['0.04', '0.86', '0.74', '0.45', '0.30', '0.09', '0.65', '0.58'],
       ['0.96', '0.58', '0.41', '0.29', '0.26', '0.54', '0.01', '0.59'],
       ['0.38', '0.86', '0.37', '0.14', '0.32', '0.57', '0.19', '0.28'],
       ['0.91', '0.80', '0.78', '0.39', '0.67', '0.51', '0.16', '0.70'],
       ['0.61', '0.12', '0.89', '0.68', '0.01', '0.23', '0.57', '0.18'],
       ['0.71', '0.29', '0.08', '0.01', '0.86', '0.03', '0.79', '0.75'],
       ['0.44', '0.84', '0.89', '0.75', '0.48', '0.88', '0.69', '0.20'],
       ['0.36', '0.69', '0.12', '0.60', '0.16', '0.39', '0.15', '0.02']],
      dtype='<U4')

Upvotes: 1

Related Questions