We Keep Coding

speed up function based on list comprehension

I'm trying to get the 15 most relevant item for each users but every functions i tried took an eternity. (more than 6 hours i shutdown it after that ...)
I have 418 unique users, 3718 unique items.
U2tfifd dict has as well 418 entry and there is 32645 words in tfidf_feature_names.
Shape of my interactions_full_df is (40733, 3)
i tried :
def index_tfidf_users(user_id) :
return [users for users in U2tfifd[user_id].flatten().tolist()]
def get_relevant_items(user_id):
return sorted(zip(tfidf_feature_names, index_tfidf_users(user_id)), key=lambda x: -x[1])[:15]
def get_tfidf_token(user_id) :
return [words for words, values in get_relevant_items(user_id)]
then interactions_full_df["tags"] = interactions_full_df["user_id"].apply(lambda x : get_tfidf_token(x))
or
def get_tfidf_token(user_id) :
tags = []
v = sorted(zip(tfidf_feature_names, U2tfifd[user_id].flatten().tolist()), key=lambda x: -x[1])[:15]
for words, values in v :
tags.append(words)
return tags
or
def get_tfidf_token(user_id) :
v = sorted(zip(tfidf_feature_names, U2tfifd[user_id].flatten().tolist()), key=lambda x: -x[1])[:15]
tags = [words for words in v]
return tags
U2tfifd is a dict with keys = user_id, values = an array

There are several things going on which could cause poor performance in your code. The impact of each of these will depend on things like your Python version (2.x or 3.x), your RAM speed, and whatnot. You'll need to experiment and benchmark the various potential improvements yourself.
1. TFIDF Sparsity (~10x speedup depending on sparsity)
One glaring potential problem is that TFIDF naturally returns sparse data (e.g. a paragraph doesn't use anywhere near as many unique words as an entire book), and working with dense structures like numpy arrays is a strange choice when the data is probably zero almost everywhere.
If you'll be doing this same analysis in the future, it might be helpful to make/use a version of TFIDF with sparse array outputs so that when you extract your tokens you can skip over the zero values. This would likely have the secondary benefit of the entire sparse array for each user fitting in the cache and preventing costly RAM access in your sorts and other operations.
It might be worth sparsifying your data anyway. On my potato, a quick benchmark on data which should be similar to yours indicates that the process can be done in ~30s. The process replaces much of the work you're doing with a highly optimized routine coded in C and wrapped for use in Python. The only real cost is the second pass through the non-zero entries, but unless that pass is pretty efficient to begin with you should be better off working with sparse data.
2. Duplicated Efforts and Memoization (~100x speedup)
If U2tfifd has 418 entries and interactions_full_df has 40733 rows then at least 40315 (or 99.0%) of your calls to get_tfidf_token() are wasted since you've already computed the answer. There are tons of memoization decorators out there, but you don't need anything very complicated for your use case.
def memoize(f):
_cache = {}
def _f(arg):
if arg not in _cache:
_cache[arg] = f(arg)
return _cache[arg]
return _f
#memoize
def get_tfidf_token(user_id):
...
Breaking this down, the function memoize() returns another function. The behavior of that function is to check a local cache for the expected return value before computing it and storing it if necessary.
The syntax #memoize... is short for something like the following.
def uncached_get_tfidf_token(user_id):
...
get_tfidf_token = memoize(uncached_get_tfidf_token)
The # symbol is used to signify that we want the modified, or decorated, version of get_tfidf_token() instead of the original. Depending on your application, it might be beneficial to chain decorators together.
3. Vectorized Operations (varying speedup, benchmarking necessary)
Python doesn't really have a notion of primitive types like other languages, and even integers take 24 bytes in memory on my machine. Lists aren't usually be packed, so you can incur costly cache misses as you're plowing through them. No matter how little work the CPU is doing for sorting and whatnot, clobbering a whole new chunk of memory to turn your array into a list and only using that brand new, expensive memory once is going to incur a performance hit.
Many of the things you are trying to do have fast (SIMD vectorized, parallelized, memory-efficient, packed memory, and other fun optimizations) numpy equivalents AND avoid unnecessary array copies and type conversions. It seems you're already using numpy anyway, so you won't have any extra imports or dependencies.
As one example, zip() creates another list in memory in Python 2.x and still does unnecessary work in Python 3.x when you really only care about the indices of tfidf_feature_names. To compute those indices, you can use something like the following, which avoids an unnecessary list creation and uses an optimized routine with slightly better asymptotic complexity as an added bonus.
def get_tfidf_token(user_id):
temp = U2tfifd[user_id].flatten()
ind = np.argpartition(temp, len(temp)-15)[-15:]
return tfidf_feature_names[ind] # works if tfidf_feature_names is a numpy array
return [tfidf_feature_names[i] for i in ind] # always works
Depending on the shape of U2tfifd[user_id], you could avoid the costly .flatten() computation by passing an axis argument to np.argsort() and flattening the 15 obtained indices instead.
4. Bonus
The sorted() function supports a reverse argument so that you can avoid extra computations like throwing a negative on every value. Simply use
sorted(..., reverse=True)
Even better, since you really don't care about the sort itself but just the 15 largest values you can get away with
sorted(...)[-15:]
to index the largest 15 instead of reversing the sort and taking the smallest 15. That doesn't really matter if you're using a better function for the application like np.argpartition(), but it could be helpful in the future.
You can also avoid some function calls by replacing .apply(lambda x : get_tfidf_token(x)) with .apply(get_tfidf_token) since get_tfidf_token is already a function which has the intended behavior. You don't really need the extra lambda.
As far as I can see though, most additional gains are fairly nitpicky and system-dependent. You can make most things faster with Cython or straight C with enough time for example, but you already have reasonably fast routines which do what you want out of the box. The extra engineering effort probably isn't worth any potential gains.

Python - Versioned list instead of immutable list?

Update:
As of CPython 3.6, dictionaries have a version (thank you pylang for showing this to me).
If they added the same version to list and made it public, all 3 asserts from my original post would pass! It would definitely meet my needs. Their implementation differs from what I envisioned, but I like it.
As it is, I don't feel I can use dictionary version:
It isn't public. Jake Vanderplas shows how to expose it in a post, but he cautions: definitely not code you should use for any purpose beyond simply having fun. I agree with his reasons.
In all of my use cases, the data is conceptually arrays of elements each of which has the same structure. A list of tuples is a natural fit. Using a dictionary would make the code less natural and probably more cumbersome.
Does anyone know if there are plans to add version to list?
Are there plans to make it public?
If there are plans to add version to list and make it public, I would feel awkward putting forward an incompatible VersionedList now. I would just implement the bare minimum I need and get by.
Original post below
Turns out that many of the times I wanted an immutable list, a VersionedList would have worked almost as well (sometimes even better).
Has anyone implemented a versioned list?
Is there a better, more Pythonic, concept that meets my needs? (See motivation below.)
What I mean by a versioned list is:
A class that behaves like a list
Any change to an instance or elements in the instance results in instance.version() being updated. So, if alist is a normal list:
a = VersionedList(alist)
a_version = a.version()
change(a)
assert a_version != a.version()
reverse_last_change(a)
If a list was hashable, hash() would achieve the above and meet all the needs identified in the motivation below. We need to define 'version()' in a way that doesn't have all of the same problems as 'hash()'.
If identical data in two lists is highly unlikely to ever happen except at initialization, we aren't going to have a reason to test for deep equality. From (https://docs.python.org/3.5/reference/datamodel.html#object.hash) The only required property is that objects which compare equal have the same hash value. If we don't impose this requirement on 'version()', it seems likely that 'version()' won't have all of the same problems that makes lists unhashable. So unlike hash, identical contents doesn't mean the same version
#contents of 'a' are now identical to original, but...
assert a_version != a.version()
b = VersionedList(alist)
c = VersionedList(alist)
assert b.version() != c.version()
For VersionList, it would be good if any attempt to modify the result of __get__ automatically resulted in a copy instead of modifying the underlying implementation data. I think that the only other option would be to have __get__ always return a copy of the elements, and this would be very inefficient for all of the use cases I can think of. I think we need to restrict the elements to immutable objects (deeply immutable, for example: exclude tuples with list elements). I can think of 3 ways to achieve this:
Only allow elements that can't contain mutable elements (int, str, etc are fine, but exclude tuples). (This is far too limiting for my cases)
Add code to __init__, __set__, etc to traverse inputs to deeply check for mutable sub-elements. (expensive, any way to avoid this?)
Also allow more complex elements, but require that they are deeply immutable. Perhaps require that they expose a deeply_immutable attribute. (This turns out to be easy for all the use cases I have)
Motivation:
If I am analyzing a dataset, I often have to perform multiple steps that return large datasets (note: since the dataset is ordered, it is best represented by a List not a set).
If at the end of several steps (ex: 5) it turns out that I need to perform different analysis (ex: back at step 4), I want to know that the dataset from step 3 hasn't accidentally been changed. That way I can start at step 4 instead of repeating steps 1-3.
I have functions (control-points, first-derivative, second-derivative, offset, outline, etc) that depend on and return array-valued objects (in the linear algebra sense). The base 'array' is knots.
control-points() depends on: knots, algorithm_enum
first-derivative() depends on: control-points(), knots
offset() depends on: first-derivative(), control-points(), knots, offset_distance
outline() depends on: offset(), end_type_enum
If offset_distance changes, I want to avoid having to recalculate first-derivative() and control-points(). To avoid recalculation, I need to know that nothing has accidentally changed the resultant 'arrays'.
If 'knots' changes, I need to recalculate everything and not depend on the previous resultant 'arrays'.
To achieve this, knots and all of the 'array-valued' objects could be VersionedList.
FYI: I had hoped to take advantage of an efficient class like numpy.ndarray. In most of my use cases, the elements logically have structure. Having to mentally keep track of multi-dimensions of indexes meant implementing and debugging the algorithms was many times more difficult with ndarray. An implementation based on lists of namedtuples of namedtuples turned out to be much more sustainable.

Private dicts in 3.6
In Python 3.6, dictionaries are now private (PEP 509) and compact (issue 27350), which track versions and preserve order respectively. These features are presently true when using the CPython 3.6 implementation. Despite the challenge, Jake VanderPlas demonstrates in his blog post a detailed demonstration of exposing this versioning feature from CPython within normal Python. We can use his approach to:
determine when a dictionary has been updated
preserve the order
Example
import numpy as np
d = {"a": np.array([1,2,3]),
"c": np.array([1,2,3]),
"b": np.array([8,9,10]),
}
for i in range(3):
print(d.get_version()) # monkey-patch
# 524938
# 524938
# 524938
Notice the version number does not change until the dictionary is updated, as shown below:
d.update({"c": np.array([10, 11, 12])})
d.get_version()
# 534448
In addition, the insertion order is preserved (the following was tested in restarted sessions of Python 3.5 and 3.6):
list(d.keys())
# ['a', 'c', 'b']
You may be able to take advantage of this new dictionary behavior, saving you from implementing a new datatype.
Details
For those interested, the latter get_version()is a monkey-patched method for any dictionary, implemented in Python 3.6 using the following modified code derived from Jake VanderPlas' blog post. This code was run prior to calling get_version().
import types
import ctypes
import sys
assert (3, 6) <= sys.version_info < (3, 7) # valid only in Python 3.6
py_ssize_t = ctypes.c_ssize_t
# Emulate the PyObjectStruct from CPython
class PyObjectStruct(ctypes.Structure):
_fields_ = [('ob_refcnt', py_ssize_t),
('ob_type', ctypes.c_void_p)]
# Create a DictStruct class to wrap existing dictionaries
class DictStruct(PyObjectStruct):
_fields_ = [("ma_used", py_ssize_t),
("ma_version_tag", ctypes.c_uint64),
("ma_keys", ctypes.c_void_p),
("ma_values", ctypes.c_void_p),
]
def __repr__(self):
return (f"DictStruct(size={self.ma_used}, "
f"refcount={self.ob_refcnt}, "
f"version={self.ma_version_tag})")
#classmethod
def wrap(cls, obj):
assert isinstance(obj, dict)
return cls.from_address(id(obj))
assert object.__basicsize__ == ctypes.sizeof(PyObjectStruct)
assert dict.__basicsize__ == ctypes.sizeof(DictStruct)
# Code for monkey-patching existing dictionaries
class MappingProxyStruct(PyObjectStruct):
_fields_ = [("mapping", ctypes.POINTER(DictStruct))]
#classmethod
def wrap(cls, D):
assert isinstance(D, types.MappingProxyType)
return cls.from_address(id(D))
assert types.MappingProxyType.__basicsize__ == ctypes.sizeof(MappingProxyStruct)
def mappingproxy_setitem(obj, key, val):
"""Set an item in a read-only mapping proxy"""
proxy = MappingProxyStruct.wrap(obj)
ctypes.pythonapi.PyDict_SetItem(proxy.mapping,
ctypes.py_object(key),
ctypes.py_object(val))
mappingproxy_setitem(dict.__dict__,
'get_version',
lambda self: DictStruct.wrap(self).ma_version_tag)

Why are methods so slow?

Ok, I understand in languages like C++ why calling virtual method defined in a class is slower than calling a non-virtual method (you have to go through the dynamic dispatch table to lookup the correct implementation to call).
But in Python, if I have:
list_of_sets = generate_a_list_containg_a_bunch_of_sets()
intersection_of_all = reduce(list_of_sets[0].intersection, list_of_sets)
This is dramatically (in my experiments about 40%) slower than:
list_of_sets = generate_a_list_containg_a_bunch_of_sets()
intersection_of_all = reduce(set.intersection, list_of_sets)
What I don't get is why that should be so much slower, the method lookup (I would think) would happen on the call to reduce, so the inside of reduce where the intersection method is actually called shouldn't have to be looked up again (it just just reuse the same method reference).
Could someone illuminate where my understanding is flawed?

This is completely unrelated to method binding etc. The first version computes the intersection of three sets in each iteration, while the second version only intersects two sets. This is easy to see if we use the explicit loops instead.
Variant 1:
intersection = list_of_sets[0]
for s in list_of_sets[1:]:
intersection = list_of_sets[0].intersection(intersection, s)
Variant 2:
intersection = list_of_sets[0]
for s in list_of_sets[1:]:
intersection = set.intersection(intersection, s)
(Would you agree now Guido has a point?)
Note that this will probably be even faster:
intersection = list_of_sets[0]
for s in list_of_sets[1:]:
intersection.intersection_update(s)

Which is faster and why? Set or List?

Lets say that I have a graph and want to see if b in N[a]. Which is the faster implementation and why?
a, b = range(2)
N = [set([b]), set([a,b])]
OR
N= [[b],[a,b]]
This is obviously oversimplified, but imagine that the graph becomes really dense.

Membership testing in a set is vastly faster, especially for large sets. That is because the set uses a hash function to map to a bucket. Since Python implementations automatically resize that hash table, the speed can be constant (O(1)) no matter the size of the set (assuming the hash function is sufficiently good).
In contrast, to evaluate whether an object is a member of a list, Python has to compare every single member for equality, i.e. the test is O(n).

It all depends on what you're trying to accomplish. Using your example verbatim, it's faster to use lists, as you don't have to go through the overhead of creating the sets:
import timeit
def use_sets(a, b):
return [set([b]), set([a, b])]
def use_lists(a, b):
return [[b], [a, b]]
t=timeit.Timer("use_sets(a, b)", """from __main__ import use_sets
a, b = range(2)""")
print "use_sets()", t.timeit(number=1000000)
t=timeit.Timer("use_lists(a, b)", """from __main__ import use_lists
a, b = range(2)""")
print "use_lists()", t.timeit(number=1000000)
Produces:
use_sets() 1.57522511482
use_lists() 0.783344984055
However, for reasons already mentioned here, you benefit from using sets when you are searching large sets. It's impossible to tell by your example where that inflection point is for you and whether or not you'll see the benefit.
I suggest you test it both ways and go with whatever is faster for your specific use-case.

Set ( I mean a hash based set like HashSet) is much faster than List to lookup for a value. List has to go sequentially to find out if the value exists. HashSet can directly jump and locate the bucket and look up for a value almost in a constant time.

What is a flexible, hybrid python collection object?

As a way to get used to python, I am trying to translate some of my code to python from Autohotkey_L.
I am immediately running into tons of choices for collection objects. Can you help me figure out a built in type or a 3rd party contributed type that has as much as possible, the functionality of the AutoHotkey_L object type and its methods.
AutoHotkey_L Objects have features of a python dict, list, and a class instance.
I understand that there are tradeoffs for space and speed, but I am just interested in functionality rather than optimization issues.

Don't write Python as <another-language>. Write Python as Python.
The data structure should be chosen just to have the minimal ability you need to use.
list — an ordered sequence of elements, with 1 flexible end.
collections.deque — an ordered sequence of elements, with 2 flexible ends (e.g. a queue).
set / frozenset — an unordered sequence of unique elements.
collections.Counter — an unordered sequence of non-unique elements.
dict — an unordered key-value relationship.
collections.OrderedDict — an ordered key-value relationship.
bytes / bytearray — a list of bytes.
array.array — a homogeneous list of primitive types.
Looking at the interface of Object,
dict would be the most suitable for finding a value by key
collections.OrderedDict would be the most suitable for the push/pop stuff.
when you need MinIndex / MaxIndex, where a sorted key-value relationship (e.g. red black tree) is required. There's no such type in the standard library, but there are 3rd party implementations.

It would be impossible to recommend a particular class without knowing how you intend on using it. If you are using this particular object as an ordered sequence where elements can be repeated, then you should use a list; if you are looking up values by their key, then use a dictionary. You will get very different algorithmic runtime complexity with the different data types. It really does not take that much time to determine when to use which type.... I suggest you give it some further consideration.
If you really can't decide, though, here is a possibility:
class AutoHotKeyObject(object):
def __init__(self):
self.list_value = []
self.dict_value = {}
def getDict(self):
return self.dict_value
def getList(self):
return self.list_value
With the above, you could use both the list and dictionary features, like so:
obj = AutoHotKeyObject()
obj.getList().append(1)
obj.getList().append(2)
obj.getList().append(3)
print obj.getList() # Prints [1, 2, 3]
obj.getDict()['a'] = 1
obj.getDict()['b'] = 2
print obj.getDict() # Prints {'a':1, 'b':2}