In Python 2 I could do the following:
import numpy as np
f = lambda x: x**2
seq = map(f, xrange(5))
seq = np.array(seq)
print seq
# prints: [ 0 1 4 9 16]
In Python 3 it does not work anymore:
import numpy as np
f = lambda x: x**2
seq = map(f, range(5))
seq = np.array(seq)
print(seq)
# prints: <map object at 0x10341e310>
How do I get the old behaviour (converting the map results to numpy array)?
Edit: As @jonrsharpe pointed out in his answer this could be fixed if I converted seq to a list first:
seq = np.array(list(seq))
but I would prefer to avoid the extra call to list.
One more alternative, other than the valid solutions @jonrsharpe already pointed out is to use np.fromiter:
>>> import numpy as np
>>> f = lambda x: x**2
>>> seq = map(f, range(5))
>>> np.fromiter(seq, dtype=np.int)
array([ 0, 1, 4, 9, 16])
Although you refer to it as seq, the map object in Python 3 is not a sequence (it's an iterator, see what's new in Python 3). numpy.array needs a sequence so the len can be determined and the appropriate amount of memory reserved; it won't consume an iterator. For example, the range object, which does support most sequence operations, can be passed directly;
seq = np.array(range(5))
print(seq)
# prints: [0 1 2 3 4]
To restore the previous behaviour, as you're aware, you can explicitly convert the map object back to a sequence (e.g. list or tuple):
seq = np.array(list(seq)) # should probably change the name!
However, as the documentation puts it:
a quick fix is to wrap map() in list(), e.g. list(map(...)), but a better fix is often to use a list comprehension (especially when the original code uses lambda)
So another option would be:
seq = [f(x) for x in range(5)]
or just:
seq = [x**2 for x in range(5)]
Alternatively, actually use numpy from the start:
import numpy as np
arr = np.arange(5)
arr **= 2
print(arr)
# prints [ 0 1 4 9 16] in 2.x and 3.x
