Efficient comparison of 1 million vectors containing (float, integer) tuples - database

I am working in a chemistry/biology project. We are building a web-application for fast matching of the user's experimental data with predicted data in a reference database. The reference database will contain up to a million entries. The data for one entry is a list (vector) of tuples containing a float value between 0.0 and 20.0 and an integer value between 1 and 18. For instance (7.2394 , 2) , (7.4011, 1) , (9.9367, 3) , ... etc.
The user will enter a similar list of tuples and the web-app must then return the - let's say - top 50 best matching database entries.
One thing is crucial: the search algorithm must allow for discrepancies between the query data and the reference data because both can contain small errors in the float values (NOT in the integer values). (The query data can contain errors because it is derived from a real-life experiment and the reference data because it is the result of a prediction.)
Edit - Moved text to answer -
How can we get an efficient ranking of 1 query on 1 million records?

You should add a physicist to the project :-) This is a very common problem to compare functions e.g. look here:
http://en.wikipedia.org/wiki/Autocorrelation
http://en.wikipedia.org/wiki/Correlation_function
In the first link you can read: "The SEQUEST algorithm for analyzing mass spectra makes use of autocorrelation in conjunction with cross-correlation to score the similarity of an observed spectrum to an idealized spectrum representing a peptide."

An efficient linear scan of 1 million records of that type should take a fraction of a second on a modern machine; a compiled loop should be able to do it at about memory bandwidth, which would transfer that in a two or three milliseconds.
But, if you really need to optimise this, you could construct a hash table of the integer values, which would divide the job by the number of integer bins. And, if the data is stored sorted by the floats, that improves the locality of matching by those; you know you can stop once you're out of tolerance. Storing the offsets of each of a number of bins would give you a position to start.
I guess I don't see the need for a fancy algorithm yet... describe the problem a bit more, perhaps (you can assume a fairly high level of chemistry and physics knowledge if you like; I'm a physicist by training)?
Ok, given the extra info, I still see no need for anything better than a direct linear search, if there's only 1 million reference vectors and the algorithm is that simple. I just tried it, and even a pure Python implementation of linear scan took only around three seconds. It took several times longer to make up some random data to test with. This does somewhat depend on the rather lunatic level of optimisation in Python's sorting library, but that's the advantage of high level languages.
from cmath import *
import random
r = [(random.uniform(0,20), random.randint(1,18)) for i in range(1000000)]
# this is a decorate-sort-undecorate pattern
# look for matches to (7,9)
# obviously, you can use whatever distance expression you want
zz=[(abs((7-x)+(9-y)),x,y) for x,y in r]
zz.sort()
# return the 50 best matches
[(x,y) for a,x,y in zz[:50]]

Can't you sort the tuples and perform binary search on the sorted array ?
I assume your database is done once for all, and the positions of the entries is not important. You can sort this array so that the tuples are in a given order. When a tuple is entered by the user, you just look in the middle of the sorted array. If the query value is larger of the center value, you repeat the work on the upper half, otherwise on the lower one.
Worst case is log(n)

If you can "map" your reference data to x-y coordinates on a plane there is a nifty technique which allows you to select all points under a given distance/tolerance (using Hilbert curves).
Here is a detailed example.

One approach we are trying ourselves which allows for the discrepancies between query and reference is by binning the float values. We are testing and want to offer the user the choice of different bin sizes. Bin sizes will be 0.1 , 0.2 , 0.3 or 0.4. So binning leaves us with between 50 and 200 bins, each with a corresponding integer value between 0 and 18, where 0 means there was no value within that bin. The reference data can be pre-binned and stored in the database. We can then take the binned query data and compare it with the reference data. One approach could be for all bins, subtract the query integer value from the reference integer value. By summing up all differences we get the similarity score, with the the most similar reference entries resulting in the lowest scores.
Another (simpler) search option we want to offer is where the user only enters the float values. The integer values in both query as reference list can then be set to 1. We then use Hamming distance to compute the difference between the query and the reference binned values. I have previously asked about an efficient algorithm for that search.
This binning is only one way of achieving our goal. I am open to other suggestions. Perhaps we can use Principal Component Analysis (PCA), as described here

Related

Algorithm - How to select one number from each column in an array so that their sum is as close as possible to a particular value

I have an m x n matrix of real numbers. I want to choose a single value from each column such that the sum of my selected values is as close as possible to a pre-specified total.
I am not an experienced programmer (although I have an experienced friend who will help). I would like to achieve this using Matlab, Mathematica or c++ (MySQL if necessary).
The code only needs to run a few times, once every few days - it does not necessarily need to be optimised. I will have 16 columns and about 12 rows.
Normally I would suggest dynamic programming, but there are a few features of this situation suggesting an alternative approach. First, the performance demands are light; this program will be run only a couple times, and it doesn't sound as though a running time on the order of hours would be a problem. Second, the matrix is fairly small. Third, the matrix contains real numbers, so it would be necessary to round and then do a somewhat sophisticated search to ensure that the optimal possibility was not missed.
Instead, I'm going to suggest the following semi-brute-force approach. 12**16 ~ 1.8e17, the total number of possible choices, is too many, but 12**9 ~ 5.2e9 is doable with brute force, and 12**7 ~ 3.6e7 fits comfortably in memory. Compute all possible choices for the first seven columns. Sort these possibilities by total. For each possible choice for the last nine columns, use an efficient search algorithm to find the best mate among the first seven. (If you have a lot of memory, you could try eight and eight.)
I would attempt a first implementation in C++, using std::sort and std::lower_bound from the <algorithm> standard header. Measure it; if it's too slow, then try an in-memory B+-tree (does Boost have one?).
I spent some more time thinking about how to implement what I wrote above in the simplest way possible. Here's an approach that will work well for a 12 by 16 matrix on a 64-bit machine with roughly 4 GB of memory.
The number of choices for the first eight columns is 12**8. Each choice is represented by a 4-byte integer between 0 and 12**8 - 1. To decode a choice index i, the row for the first column is given by i % 12. Update i /= 12;. The row for the second column now is given by i % 12, et cetera.
A vector holding all choices requires roughly 12**8 * 4 bytes, or about 1.6 GB. Two such vectors require 3.2 GB. Prepare one for the first eight columns and one for the last eight. Sort them by sum of the entries that they indicate. Use saddleback search to find the best combination. (Initialize an iterator into the first vector and a reverse iterator into the second. While neither iterator is at its end, compare the current combination against the current best and update the current best if necessary. If the current combination sums to than the target, increment the first iterator. If the sum is greater than the target, increment the second iterator.)
I would estimate that this requires less than 50 lines of C++.
Without knowing the range of values that might fill the arrays, how about something generic like this:
divide the target by the number of remaining columns.
Pick the number from that column closest to that value.
Repeat from 1. Until each column picked.

Need algorithm for fast storage and retrieval (search) of sets and subsets

I need a way of storing sets of arbitrary size for fast query later on.
I'll be needing to query the resulting data structure for subsets or sets that are already stored.
===
Later edit: To clarify, an accepted answer to this question would be a link to a study that proposes a solution to this problem. I'm not expecting for people to develop the algorithm themselves.
I've been looking over the tuple clustering algorithm found here, but it's not exactly what I want since from what I understand it 'clusters' the tuples into more simple, discrete/aproximate forms and loses the original tuples.
Now, an even simpler example:
[alpha, beta, gamma, delta] [alpha, epsilon, delta] [gamma, niu, omega] [omega, beta]
Query:
[alpha, delta]
Result:
[alpha, beta, gama, delta] [alpha, epsilon, delta]
So the set elements are just that, unique, unrelated elements. Forget about types and values. The elements can be tested among them for equality and that's it. I'm looking for an established algorithm (which probably has a name and a scientific paper on it) more than just creating one now, on the spot.
==
Original examples:
For example, say the database contains these sets
[A1, B1, C1, D1], [A2, B2, C1], [A3, D3], [A1, D3, C1]
If I use [A1, C1] as a query, these two sets should be returned as a result:
[A1, B1, C1, D1], [A1, D3, C1]
Example 2:
Database:
[Gasoline amount: 5L, Distance to Berlin: 240km, car paint: red]
[Distance to Berlin: 240km, car paint: blue, number of car seats: 2]
[number of car seats: 2, Gasoline amount: 2L]
Query:
[Distance to berlin: 240km]
Result
[Gasoline amount: 5L, Distance to Berlin: 240km, car paint: red]
[Distance to Berlin: 240km, car paint: blue, number of car seats: 2]
There can be an unlimited number of 'fields' such as Gasoline amount. A solution would probably involve the database grouping and linking sets having common states (such as Gasoline amount: 240) in such a way that the query is as efficient as possible.
What algorithms are there for such needs?
I am hoping there is already an established solution to this problem instead of just trying to find my own on the spot, which might not be as efficient as one tested and improved upon by other people over time.
Clarifications:
If it helps answer the question, I'm intending on using them for storing states:
Simple example:
[Has milk, Doesn't have eggs, Has Sugar]
I'm thinking such a requirement might require graphs or multidimensional arrays, but I'm not sure
Conclusion
I've implemented the two algorithms proposed in the answers, that is Set-Trie and Inverted Index and did some rudimentary profiling on them. Illustrated below is the duration of a query for a given set for each algorithm. Both algorithms worked on the same randomly generated data set consisting of sets of integers. The algorithms seem equivalent (or almost) performance wise:
I'm confident that I can now contribute to the solution. One possible quite efficient way is a:
Trie invented by Frankling Mark Liang
Such a special tree is used for example in spell checking or autocompletion and that actually comes close to your desired behavior, especially allowing to search for subsets quite conveniently.
The difference in your case is that you're not interested in the order of your attributes/features. For your case a Set-Trie was invented by Iztok Savnik.
What is a Set-Tree? A tree where each node except the root contains a single attribute value (number) and a marker (bool) if at this node there is a data entry. Each subtree contains only attributes whose values are larger than the attribute value of the parent node. The root of the Set-Tree is empty. The search key is the path from the root to a certain node of the tree. The search result is the set of paths from the root to all nodes containing a marker that you reach when you go down the tree and up the search key simultaneously (see below).
But first a drawing by me:
The attributes are {1,2,3,4,5} which can be anything really but we just enumerate them and therefore naturally obtain an order. The data is {{1,2,4}, {1,3}, {1,4}, {2,3,5}, {2,4}} which in the picture is the set of paths from the root to any circle. The circles are the markers for the data in the picture.
Please note that the right subtree from root does not contain attribute 1 at all. That's the clue.
Searching including subsets Say you want to search for attributes 4 and 1. First you order them, the search key is {1,4}. Now startin from root you go simultaneously up the search key and down the tree. This means you take the first attribute in the key (1) and go through all child nodes whose attribute is smaller or equal to 1. There is only one, namely 1. Inside you take the next attribute in the key (4) and visit all child nodes whose attribute value is smaller than 4, that are all. You continue until there is nothing left to do and collect all circles (data entries) that have the attribute value exactly 4 (or the last attribute in the key). These are {1,2,4} and {1,4} but not {1,3} (no 4) or {2,4} (no 1).
Insertion Is very easy. Go down the tree and store a data entry at the appropriate position. For example data entry {2.5} would be stored as child of {2}.
Add attributes dynamically Is naturally ready, you could immediately insert {1,4,6}. It would come below {1,4} of course.
I hope you understand what I want to say about Set-Tries. In the paper by Iztok Savnik it's explained in much more detail. They probably are very efficient.
I don't know if you still want to store the data in a database. I think this would complicate things further and I don't know what is the best to do then.
How about having an inverse index built of hashes?
Suppose you have your values int A, char B, bool C of different types. With std::hash (or any other hash function) you can create numeric hash values size_t Ah, Bh, Ch.
Then you define a map that maps an index to a vector of pointers to the tuples
std::map<size_t,std::vector<TupleStruct*> > mymap;
or, if you can use global indices, just
std::map<size_t,std::vector<size_t> > mymap;
For retrieval by queries X and Y, you need to
get hash value of the queries Xh and Yh
get the corresponding "sets" out of mymap
intersect the sets mymap[Xh] and mymap[Yh]
If I understand your needs correctly, you need a multi-state storing data structure, with retrievals on combinations of these states.
If the states are binary (as in your examples: Has milk/doesn't have milk, has sugar/doesn't have sugar) or could be converted to binary(by possibly adding more states) then you have a lightning speed algorithm for your purpose: Bitmap Indices
Bitmap indices can do such comparisons in memory and there literally is nothing in comparison on speed with these (ANDing bits is what computers can really do the fastest).
http://en.wikipedia.org/wiki/Bitmap_index
Here's the link to the original work on this simple but amazing data structure: http://www.sciencedirect.com/science/article/pii/0306457385901086
Almost all SQL databases supoort Bitmap Indexing and there are several possible optimizations for it as well(by compression etc.):
MS SQL: http://technet.microsoft.com/en-us/library/bb522541(v=sql.105).aspx
Oracle: http://www.orafaq.com/wiki/Bitmap_index
Edit:
Apparently the original research work on bitmap indices is no longer available for free public access.
Links to recent literature on this subject:
Bitmap Index Design Choices and Their Performance
Implications
Bitmap Index Design and Evaluation
Compressing Bitmap Indexes for Faster Search Operations
This problem is known in the literature as subset query. It is equivalent to the "partial match" problem (e.g.: find all words in a dictionary matching A??PL? where ? is a "don't care" character).
One of the earliest results in this area is from this paper by Ron Rivest from 19761. This2 is a more recent paper from 2002. Hopefully, this will be enough of a starting point to do a more in-depth literature search.
Rivest, Ronald L. "Partial-match retrieval algorithms." SIAM Journal on Computing 5.1 (1976): 19-50.
Charikar, Moses, Piotr Indyk, and Rina Panigrahy. "New algorithms for subset query, partial match, orthogonal range searching, and related problems." Automata, Languages and Programming. Springer Berlin Heidelberg, 2002. 451-462.
This seems like a custom made problem for a graph database. You make a node for each set or subset, and a node for each element of a set, and then you link the nodes with a relationship Contains. E.g.:
Now you put all the elements A,B,C,D,E in an index/hash table, so you can find a node in constant time in the graph. Typical performance for a query [A,B,C] will be the order of the smallest node, multiplied by the size of a typical set. E.g. to find {A,B,C] I find the order of A is one, so I look at all the sets A is in, S1, and then I check that it has all of BC, since the order of S1 is 4, I have to do a total of 4 comparisons.
A prebuilt graph database like Neo4j comes with a query language, and will give good performance. I would imagine, provided that the typical orders of your database is not large, that its performance is far superior to the algorithms based on set representations.
Hashing is usually an efficient technique for storage and retrieval of multidimensional data. Problem is here that the number of attributes is variable and potentially very large, right? I googled it a bit and found Feature Hashing on Wikipedia. The idea is basically the following:
Construct a hash of fixed length from each data entry (aka feature vector)
The length of the hash must be much smaller than the number of available features. The length is important for the performance.
On the wikipedia page there is an implementation in pseudocode (create hash for each feature contained in entry, then increase feature-vector-hash at this index position (modulo length) by one) and links to other implementations.
Also here on SO is a question about feature hashing and amongst others a reference to a scientific paper about Feature Hashing for Large Scale Multitask Learning.
I cannot give a complete solution but you didn't want one. I'm quite convinced this is a good approach. You'll have to play around with the length of the hash as well as with different hashing functions (bloom filter being another keyword) to optimize the speed for your special case. Also there might still be even more efficient approaches if for example retrieval speed is more important than storage (balanced trees maybe?).

Motivation for k-medoids

Why would one use kmedoids algoirthm rather then kmeans? Is it only the fact that
the number of metrics that can be used in kmeans is very limited or is there something more?
Is there an example of data, for which it makes much more sense to choose the best representatives
of cluster from the data rather then from R^n?
The problem with k-means is that it is not interpretable. By interpretability i mean the model should also be able to output the reason that why it has resulted a certain output.
lets take an example.
Suppose there is food review dataset which has two posibility that there is a +ve review or a -ve review so we can say we will have k= 2 where k is the number of clusters. Now if you go with k-means where in the algorithm the third step is updation step where you update your k-centroids based on the mean distance of the points that lie in a particular cluster. The example that we have chosen is text problem, so you would also apply some kind of text-featured vector schemes like BagOfWords(BOW), word2vec. now for every review you would get the corresponding vector. Now the generated centroid c_i that you will get after running the k-means would be the mean of the vectors present in that cluster. Now with that centroid you cannot interpret much or rather i should say nothing.
But for same problem you apply k-medoids wherein you choose your k-centroids/medoids from your dataset itself. lets say you choose x_5 point from your dataset as first medoid. From this your interpretability will increase beacuse now you have the review itself which is termed as medoid/centroid. So in k-medoids you choose the centroids from your dataset itself.
This is the foremost motivation of introducing k-mediods
Coming to the metrics part you can apply all the metrics that you apply for k-means
Hope this helps.
Why would we use k-medoids instead of k-means in case of (squared) Euclidean distance?
1. Technical justification
In case of relatively small data sets (as k-medoids complexity is greater) - to obtain a clustering more robust to noise and outliers.
Example 2D data showing that:
The graph on the left shows clusters obtained with K-medoids (sklearn_extra.cluster.KMedoids method in Python with default options) and the one on the right with K-means for K=2. Blue crosses are cluster centers.
The Python code used to generate green points:
import numpy as np
import matplotlib.pyplot as plt
rng = np.random.default_rng(seed=32)
a = rng.random((6,2))*2.35 - 3*np.ones((6,2))
b = rng.random((50,2))*0.25 - 2*np.ones((50,2))
c = rng.random((100,2))*0.5 - 1.5*np.ones((100,2))
d = rng.random((7,2))*0.55
points = np.concatenate((a, b, c, d))
plt.plot(points[:,0],points[:,1],"g.", markersize=8, alpha=0.3) # green points
2. Business case justification
Here are some example business cases showing why we would prefer k-medoids. They mostly come down to the interpretability of the results and the fact that in k-medoids the resulting cluster centers are members of the original dataset.
2.1 We have a recommender engine based only on user-item preference data and want to recommend to the user those items (e.g. movies) that other similar people enjoyed. So we assign the user to his/her closest cluster and recommend top movies that the cluster representant (actual person) watched. If the cluster representant wasn't an actual person we wouldn't possess the history of actually watched movies to recommend. Each time we'd have to search additionally e.g. for the closest person from the cluster. Example data: classic MovieLens 1M Dataset
2.2 We have a database of patients and want to pick a small representative group of size K to test a new drug with them. After clustering the patients with K-medoids, cluster representants are invited to the drug trial.
Difference between is that in k-means centroids(cluster centrum) are calculated as average of vectors containing in the cluster, and in k-medoids the medoid (cluster centrum) is record from dataset closest to centroid, so if you need to represent cluster centrum by record of your data you use k-medoids, otherwise i should use k-means (but concept of these algorithms are same)
The K-Means algorithm uses a Distance Function such as Euclidean Distance or Manhattan Distance, which are computed over vector-based instances. The K-Medoid algorithm instead uses a more general (and less constrained) distance function: aka pair-wise distance function.
This distinction works well in contexts like Complex Data Types or relational rows, where the instances have a high number of dimensions.
High dimensionality problem
In standard clustering libraries and the k-means algorithms, the distance computation phase can spend a lot of time scanning the entire vector of attributes that belongs to an instance; for instance, in the context of documents clustering, using the standard TF-IDF representation. During the computation of the cosine similarity, the distance function scans all the possible words that appear in the whole collection of documents. Which in many cases can be composed by millions of entries. This is why, in this domain, some authors [1] suggests to restrict the words considered to a subset of N most frequent word of that language.
Using K-Kedoids there is no need to represent and store the documents as vectors of word frequencies.
As an alternative representation for the documents is possible to use the set of words appearing at least twice in the document; and as a distance measure, there can be used Jaccard Distance.
In this case, vector representation is long as the number of words in your dictionary.
Heterogeneousity and Complex Data Types.
There are many domains where is considerably better to abstract the implementation of an instance:
Graph's nodes clustering;
Car driving behaviour, represented as GPS routes;
Complex data type allows the design of ad-hoc distance measures which can fit better with the proper data domain.
[1] Christopher D. Manning, Prabhakar Raghavan, and Hinrich Schütze. 2008. Introduction to Information Retrieval. Cambridge University Press, New York, NY, USA.
Source: https://github.com/eracle/Gap

Generate pseudo sample of population given probabilities

I would like to generate pseudo data that conforms to the distribution of actual sampled data. Looking for an efficient and accurate method in C/Obj-C for iphone development. Currently the occurrance of 60 different categories in 1000 sampled events has been assigned a probability (0-1). I want to generate 1000 new events which conform to the same probabilities.
Clarification {
I have a categorical distribution of set {1,2,...,60}. I understand that samples from this distribution will conform to the probabilities of each category. Therefore I need to take 1000 samples from this distribution. I have determined (thanks to answers so far) that I need to:
Normalize this distribution by summing the values and dividing each
by the sum.
Order them.
Create a CDF by replacing each value with the sum of all previous values.
Then I can generate a uniform random number between 0 and 1, and find the greatest number in the CDF whose value is less than or equal to the number just chosen, and return the category corresponding to this CDF value.
}
Q1. Is this the correct way to solve the problem?
Q2. The caveat still holds that I'm using NSDecimals to store the category probabilities. Are there any libraries available or functions in Cocoa or Math.h, etc. that I can use to do this simply? I'm open to trying new libraries, currently only have Core-Plot and the standard Cocoa libraries in this project. Thanks.
Your problem description is unclear. But it sounds like you're looking for inverse transform sampling.
Basically, you first need to generate a cumulative distribution function (CDF) corresponding to your original data; call it F(x). You then generate uniform random data in the range 0->1, and then transform it using the inverse CDF, i.e F-1(x).
Here's my suggestion. This assumes that when you say "normalized probability" you mean the sum of the probability of all types is 1. (If not, you'll need to rescale so that's the case.)
Make up some order for your 60 types. (Say, alphabetic.)
Generate a random number between 0 and 1. (Call it your "target".)
Create an accumulator, initially at 0.
Loop through your 60 types. For each type:
Add the probability of that type of event to your accumulator.
If your accumulator is >= your target, generate an event of that type and stop.
If you do that 1000 times, I believe you'll get the distribution you're looking for.

Data structure for finding nearby keys with similar bitvalues

I have some data, up to a between a million and a billion records, each which is represented by a bitfield, about 64 bits per key. The bits are independent, you can imagine them basically as random bits.
If I have a test key and I want to find all values in my data with the same key, a hash table will spit those out very easily, in O(1).
What algorithm/data structure would efficiently find all records most similar to the query key? Here similar means that most bits are identical, but a minimal number are allowed to be wrong. This is traditionally measured by Hamming distance., which just counts the number of mismatched bits.
There's two ways this query might be made, one might be by specifying a mismatch rate like "give me a list of all existing keys which have less than 6 bits that differ from my query" or by simply best matches, like "give me a list of the 10,000 keys which have the lowest number of differing bits from my query."
You might be temped to run to k-nearest-neighbor algorithms, but here we're talking about independent bits, so it doesn't seem likely that structures like quadtrees are useful.
The problem can be solved by simple brute force testing a hash table for low numbers of differing bits. If we want to find all keys that differ by one bit from our query, for example, we can enumerate all 64 possible keys and test them all. But this explodes quickly, if we wanted to allow two bits of difference, then we'd have to probe 64*63=4032 times. It gets exponentially worse for higher numbers of bits.
So is there another data structure or strategy that makes this kind of query more efficient?
The database/structure can be preprocessed as much as you like, it's the query speed that should be optimized.
What you want is a BK-Tree. It's a tree that's ideally suited to indexing metric spaces (your problem is one), and supports both nearest-neighbour and distance queries. I wrote an article about it a while ago.
BK-Trees are generally described with reference to text and using levenshtein distance to build the tree, but it's straightforward to write one in terms of binary strings and hamming distance.
This sounds like a good fit for an S-Tree, which is like a hierarchical inverted file. Good resources on this topic include the following papers:
Hierarchical Bitmap Index: An Efficient and Scalable Indexing Technique for Set-Valued Attributes.
Improved Methods for Signature-Tree Construction (2000)
Quote from the first one:
The hierarchical bitmap index efficiently supports dif-
ferent classes of queries, including subset, superset and similarity queries.
Our experiments show that the hierarchical bitmap index outperforms
other set indexing techniques significantly.
These papers include references to other research that you might find useful, such as M-Trees.
Create a binary tree (specifically a trie) representing each key in your start set in the following way: The root node is the empty word, moving down the tree to the left appends a 0 and moving down the right appends a 1. The tree will only have as many leaves as your start set has elements, so the size should stay manageable.
Now you can do a recursive traversal of this tree, allowing at most n "deviations" from the query key in each recursive line of execution, until you have found all of the nodes in the start set which are within that number of deviations.
I'd go with an inverted index, like a search engine. You've basically got a fixed vocabulary of 64 words. Then similarity is measured by hamming distance, instead of cosine similarity like a search engine would want to use. Constructing the index will be slow, but you ought to be able to query it with normal search enginey speeds.
The book Introduction to Information Retrieval covers the efficient construction, storage, compression and querying of inverted indexes.
"Near-optimal hashing algorithms for approximate nearest neighbor in high dimensions", from 2008, seems to be the best result as of then. I won't try to summarize since I read it over a year ago and it's hairy. That's from a page on locality-sensitive hashing, along with an implementation of an earlier version of the scheme. For more general pointers, read up on nearest neighbor search.
This kind of question has been asked before: Fastest way to find most similar string to an input?
The database/structure can be
preprocessed as much as you like
Well...IF that is true. Then all you need is a similarity matrix of your hamming distances. Make the matrix sparse by pruning out large distances. It doesn't get any faster and not that much of a memory hog.
Well, you could insert all of the neighbor keys along with the original key. That would mean that you store (64 choose k) times as much data, for k differing bits, and it will require that you decide k beforehand. Though you could always extend k by brute force querying neighbors, and this will automatically query the neighbors of your neighbors that you inserted. This also gives you a time-space tradeoff: for example, if you accept a 64 x data blowup and 64 times slower you can get two bits of distance.
I haven't completely thought this through, but I have an idea of where I'd start.
You could divide the search space up into a number of buckets where each bucket has a bucket key and the keys in the bucket are the keys that are more similar to this bucket key than any other bucket key. To create the bucket keys, you could randomly generate 64 bit keys and discard any that are too close to any previously created bucket key, or you could work out some algorithm that generates keys that are all dissimilar enough. To find the closest key to a test key, first find the bucket key that is closest, and then test each key in the bucket. (Actually, it's possible, but not likely, for the closest key to be in another bucket - do you need to find the closest key, or would a very close key be good enough?)
If you're ok with doing it probabilistically, I think there's a good way to solve question 2. I assume you have 2^30 data and cutoff and you want to find all points within cutoff distance from test.
One_Try()
1. Generate randomly a 20-bit subset S of 64 bits
2. Ask for a list of elements that agree with test on S (about 2^10 elements)
3. Sort that list by Hamming distance from test
4. Discard the part of list after cutoff
You repeat One_Try as much as you need while merging the lists. The more tries you have, the more points you find. For example, if x is within 5 bits, you'll find it in one try with about (2/3)^5 = 13% probability. Therefore if you repeat 100 tries you find all but roughly 10^{-6} of such x. Total time: 100*(1000*log 1000).
The main advantage of this is that you're able to output answers to question 2 as you proceed, since after the first few tries you'll certainly find everything within distance not more than 3 bits, etc.
If you have many computers, you give each of them several tries, since they are perfectly parallelizable: each computer saves some hash tables in advance.
Data structures for large sets described here: Detecting Near-Duplicates for Web Crawling
or
in memory trie: Judy-arrays at sourceforge.net
Assuming you have to visit each row to test its value (or if you index on the bitfield then each index entry), then you can write the actual test quite efficiently using
A xor B
To find the difference bits, then bit-count the result, using a technique like this.
This effectively gives you the hamming distance.
Since this can compile down to tens of instructions per test, this can run pretty fast.
If you are okay with a randomized algorithm (monte carlo in this case), you can use the minhash.
If the data weren't so sparse, a graph with keys as the vertices and edges linking 'adjacent' (Hamming distance = 1) nodes would probably be very efficient time-wise. The space would be very large though, so in your case, I don't think it would be a worthwhile tradeoff.

Resources