Suppose I would like to loop over an array and within the loop index the array forward and backward for all of its indices like so:
x = np.random.uniform(size=600)
for i in range(len(x)):
dot = np.dot(x[:-i], x[i:])
Now this doesn't work, because x[:-0] is just like x[:0] which gives [].
I could handle the zero case separately but was wondering whether there's a more pythonic way of doing this.
Use an end of slice value of -i or None. If i is non-zero, then it's just -i, but if it's 0, then -0 is falsy, and it evaluates and returns the second term, None, which means "run to end of sequence". This works because foo[:None] is equivalent to foo[:], when you omit that component of the slice it becomes None implicitly, but it's perfectly legal to pass None explicitly, with the same effect.
So your new line would be:
dot = np.dot(x[:-i or None], x[i:])
True and False, you just care whether a value is true or false for boolean evaluation purposes; truthy or falsy makes it clear you care about the meaning, not the specific value.
len(x) outside the loop and store it off; the overhead for calling built-in functions like that is surprisingly high given how little work they actually do). -i or None only adds a single implicit boolean evaluation (about the cheapest thing CPython can do shy of the pass statement) to the common case, where len(x) - i adds a built-in lookup, built-in call, and subtract operation to every usage (local load and subtract if you precache would cost only a trivial amount more at least).Why don't you just use the length information:
length = len(x)
for i in range(length):
dot = np.dot(x[:length-i], x[i:])
