CODEWITHSUNDEEP
pythonpandasnumpytypes
Jack Overby
Jack Overby

Reputation: 91

How to convert numpy array with numeric and non_numeric entries to all floats

I have a numpy array with a mix of different dtypes: floats, ints and strings. I want to convert all floats and ints to floats, while leaving non-numeric entries untouched. Currently, when I do:

array = np.array(['1', '2', '3', 'string'])
array.astype(np.float64)

I get the following error:

ValueError: could not convert string to float: 'string'

I'd like for the output to look like this:

np.array([1.0, 2.0, 3.0, 'string'])

I've tried pd.is_numeric() as well, but can't figure it out. Is this feasible, or does it violate the rules of numpy arrays?

Upvotes: 2

Views: 3309

Answers (1)

Julien
Julien

Reputation: 15071

Your desired result is impossible since np.arrays can only have one data type (which would usually be one of the numerical types, e.g. float, int,...)... unless you choose the generic type dtype=object, but then you loose all the numpy goodness (i.e. all the optimization working on numerical values). So why do you want to do this?

If this is really what you want then try this:

array = np.array(['1', '2', '3', 'string'])

def safe_float(x):
    try:
        x = float(x)
    except:
        pass
    return x

array = np.array(list(map(safe_float, array)), dtype=object)
print(array)

array([1.0, 2.0, 3.0, 'string'], dtype=object)

Upvotes: 3

Related Questions