Elephants

Memoization utilities with fading memory.


Keywords
memo, memoize, cache
License
MIT
Install
pip install Elephants==1.0.0

Documentation

Elephants

This is a memoize package using StrongDict a class introduced in this package. StrongDict described in detail in the following section uses a linked list of strong references in combination with a weak reference dictionary to enable fading memory storage by either restricting the number of items in the store or imposing a time limit on items stored.

About the name

This package is named Elephants because they are the symbol of remembering, or in a modern Python world, memoizing.

Examples

Simple memoize

from elephants import memo

@memo
def fib(n):
  if n<2: return 1
  return fib(n-1) + fib(n-2)

Memoize with time limit

The following pattern demonstrates the @memo_until decorator to cache data object instances for a given length of time.

from elephants import memo_until

class MyDataObject(object):
  def __init__(self, x):
    self.x = x

  def __str__(self): return str(self.x)

@memo_until(tlimit=3600) # seconds
def get_my_data_obj_instance(x):
  print 'Getting a new instance for x =', x
  return MyDataObject(x)


y = get_my_data_obj_instance(1)
y = get_my_data_obj_instance(2)
y = get_my_data_obj_instance(1)
y = get_my_data_obj_instance(3)
y = get_my_data_obj_instance(1)

print 'The final value of y should be 1.  It is', y

Memoize with length limit

The following pattern demonstrates the @memo_until decorator to cache data object instances for a given length of time.

from elephants import nmemo

class MyDataObject(object):
  def __init__(self, x):
    self.x = x

  def __str__(self): return str(self.x)

@nmemo(limit=2) # Limit the cache to only two items.
def get_my_data_obj_instance(x):
  print 'Getting a new instance for x =', x
  return MyDataObject(x)


y = get_my_data_obj_instance(1)
y = get_my_data_obj_instance(2)
y = get_my_data_obj_instance(3)
y = get_my_data_obj_instance(1)

print 'At this point the cache should only have 1 and 3.'

StrongDict

A dictionary using strong and weak references.

Items are stored in a WeakValueDictionary. They are also stored in a the linked list, the strong cache, holding strong references. Retreival is a fast WeakValueDictionary lookup. Items most recently used are moved to the top of the strong cache list. Options are provided to fade from the strong linked list defined by either a time limit or a list quantity limit. The default is to keep everything in strong linked list forever. If an item is removed from the strong cache and later queried and found in the WeakValuedictionary it will be added back to the strong cache. This can only happen if an reference to the removed item exists outside the cache.

As long as items have a strong reference they will stay in the WeakValueDictionary. However, if an item no longer has a strong reference it is not guaranteed to be in the WeakValueDictionary. The WeakValueDictionary will "forget" about objects not also referenced elsewhere, in this case the StrongDict() strong reference linked list.

The weakref.WeakValueDictionary() cannot hold native types like int or str. But it can hold a python object that has native types. This StrongDict class uses a Link object defined above to hold all values stored in the cache. Therefore the user of StrongDict() does not need to be concerned with this limitation of the weakref package.

CAVEAT EMPTOR: If a class instance is used as key in the
dictionary then the __hash__() method may need to be implmented.

StrongDict Example

from elephants import StrongDict

w = StrongDict(limit=3)

# populate with test cases
for i in range(10): w[i] = i * i

# demonstrate limit is imposed and len() works for StrongDict
assert(len(w) == 3)

assert(6 not in w)

# demonstrate that del works on StrongDict instances
del w[8]
assert(len(w) == 2)

# demonstrate that KeyError is thrown when item is not in the dictionary.
try:
  del w[6]
  assert(False)
except KeyError:
  'It cannot delete w[7] since it is not there.'
  assert(True)

# demonstrate for negation on inclusion test.
del w[9]
assert(9 not in w)

# demonstrate equals operator
x = {}
for i in range(4):
  w[i] = x[i]= i * i
del x[0]
assert(w == x)

# demonstrate conversion to dict
x = dict(w)
print x
print type(x)

Special Note

@memo, @nmemo, and @memo_until have two special arguments.

If a decorated method is called with 'size_elephant_cache' as the argument it will return the number of items in the cache. If it is called with 'clear_elephant_cache' as the argument it will clear the cache as shown in the following example.

Example

from elephants import memo

@memo
def f(x): return x * x

for i in range(10): f(i)

print 'There are', f('size_elephant_cache'), 'items in the cache.'
f('clear_elephant_cache')
print 'There are', f('size_elephant_cache'), 'items in the cache.'

Thank you

A special thank you to Jonathan Arrender Smith for his guidance and education.