Time complexity of dict vs attributes

[heras]

I've simplified the code some more, hoping to make it more clear. This time I profiled both running over 20.000 values. Most of the time is spent in the simple_moving_average method of the dict version. Maybe a lot of dict look-ups are slow somehow, but why would it get exponentially slower? What can I do to narrow it down some more?

#!/usr/bin/env python3

class DataFrame:
    def __init__(self):
        self.data = {}

    def add_value(self, label, value):
        try:
            self.data[label].append(value)
        except KeyError:
            self.data[label] = [value]

    def simple_moving_average(self, src, dest, window):
        try:
            cumsum = self.data[dest+'_cumsum'][-1] + self.data[src][-1]
        except KeyError:
            cumsum = self.data[src][0]
        self.add_value(dest+'_cumsum', cumsum)

        len_cumsum = len(self.data[dest+'_cumsum'][:])
        if len_cumsum <= window:
            sma = self.data[dest+'_cumsum'][-1] / len_cumsum
        else:
            sma = (self.data[dest+'_cumsum'][-1] \
                  -self.data[dest+'_cumsum'][-window - 1]) / window
        self.add_value(dest, sma)


def read_line(fname):
    with open(fname) as f:
        for data in f:
            yield float(data)

df = DataFrame()
fname = 'data2.txt'
for value in read_line(fname):
    df.add_value('data', value)
    df.simple_moving_average('data'  , 'sma250', 250)
    df.simple_moving_average('sma250', 'sma500', 500)

$ python3 -m cProfile with_dict.py 
         300053 function calls in 14.282 seconds

   Ordered by: standard name

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    0.000    0.000 _bootlocale.py:23(getpreferredencoding)
        1    0.000    0.000    0.000    0.000 codecs.py:259(__init__)
        1    0.000    0.000    0.000    0.000 codecs.py:308(__init__)
       23    0.000    0.000    0.000    0.000 codecs.py:318(decode)
    40000   13.913    0.000   14.067    0.000 with_dict.py:13(simple_moving_average)
    20001    0.108    0.000    0.108    0.000 with_dict.py:29(read_line)
        1    0.091    0.091   14.282   14.282 with_dict.py:3(<module>)
        1    0.000    0.000    0.000    0.000 with_dict.py:3(DataFrame)
        1    0.000    0.000    0.000    0.000 with_dict.py:4(__init__)
   100000    0.123    0.000    0.142    0.000 with_dict.py:7(add_value)
       23    0.000    0.000    0.000    0.000 {built-in method _codecs.utf_8_decode}
        1    0.000    0.000    0.000    0.000 {built-in method _locale.nl_langinfo}
        1    0.000    0.000    0.000    0.000 {built-in method builtins.__build_class__}
        1    0.000    0.000   14.282   14.282 {built-in method builtins.exec}
    40000    0.028    0.000    0.028    0.000 {built-in method builtins.len}
        1    0.000    0.000    0.000    0.000 {built-in method io.open}
    99995    0.020    0.000    0.020    0.000 {method 'append' of 'list' objects}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

#!/usr/bin/env python3

class DataFrame:
    def add_value(self, label, value):
        try:
            attr = getattr(self, label)
            attr.append(value)
        except AttributeError:
            setattr(self, label, [value])

    def simple_moving_average(self, src, dest, window):
        try:
            src_attr  = getattr(self, src)
            dest_attr = getattr(self, dest+'_cumsum')
            cumsum = src_attr[-1] + dest_attr[-1]
        except AttributeError:
            cumsum = src_attr[0]
        self.add_value(dest+'_cumsum', cumsum)

        dest_attr = getattr(self, dest+'_cumsum')
        len_cumsum = len(dest_attr)

        if len_cumsum <= window:
            sma = dest_attr[-1] / len_cumsum
        else:
            sma = (dest_attr[-1] - dest_attr[-window - 1]) / window
        self.add_value(dest, sma)

def read_line(fname):
    with open(fname) as f:
        for data in f:
            yield float(data)

df = DataFrame()
fname = 'data2.txt'
for value in read_line(fname):
    df.add_value('data', value)
    df.simple_moving_average('data'  , 'sma250', 250)
    df.simple_moving_average('sma250', 'sma500', 500)

$ python3 -m cProfile with_attr.py 
         520057 function calls in 0.330 seconds

   Ordered by: standard name

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    0.000    0.000 _bootlocale.py:23(getpreferredencoding)
        1    0.000    0.000    0.000    0.000 codecs.py:259(__init__)
        1    0.000    0.000    0.000    0.000 codecs.py:308(__init__)
       23    0.000    0.000    0.000    0.000 codecs.py:318(decode)
    40000    0.126    0.000    0.257    0.000 with_attr.py:11(simple_moving_average)
    20001    0.017    0.000    0.017    0.000 with_attr.py:29(read_line)
        1    0.036    0.036    0.330    0.330 with_attr.py:3(<module>)
        1    0.000    0.000    0.000    0.000 with_attr.py:3(DataFrame)
   100000    0.063    0.000    0.104    0.000 with_attr.py:4(add_value)
       23    0.000    0.000    0.000    0.000 {built-in method _codecs.utf_8_decode}
        1    0.000    0.000    0.000    0.000 {built-in method _locale.nl_langinfo}
        1    0.000    0.000    0.000    0.000 {built-in method builtins.__build_class__}
        1    0.000    0.000    0.330    0.330 {built-in method builtins.exec}
   220000    0.067    0.000    0.067    0.000 {built-in method builtins.getattr}
    40000    0.007    0.000    0.007    0.000 {built-in method builtins.len}
        5    0.000    0.000    0.000    0.000 {built-in method builtins.setattr}
        1    0.000    0.000    0.000    0.000 {built-in method io.open}
    99995    0.013    0.000    0.013    0.000 {method 'append' of 'list' objects}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}