tinysearch 0.5.0

Tiny one-phase search engine

TinySearch

TinySearch is a tiny one-phase search engine. It is extremely easy to use and works well with simple lists where the query may not match the document text exactly.

This is a minimal search engine. You don't need to run separate, big instances of search engine when your use case is a few hundreds or thousands small documents.

Example

Input documents:

"Goldilocks and the Three Bears"
"Fuzzy Wuzzy"
"The Bear Went Over The Mountain"
"We're Going on a Bear Hunt"
"Brown Bear, Brown Bear, What Do You See?"

Search query:

bear

Results (ordered by best match):

"Brown Bear, Brown Bear, What Do You See?"
"Goldilocks and the Three Bears"
"The Bear Went Over The Mountain"
"We're Going on a Bear Hunt"

How to use

from tinysearch.search import Search

docs = [
    "Goldilocks and the Three Bears",
    "Fuzzy Wuzzy",
    "The Bear Went Over The Mountain",
    "We're Going on a Bear Hunt",
    "Brown Bear, Brown Bear, What Do You See?",
]
query = "bear"

s = Search(docs, query)

# How many results?
print(s.results.count)

# What is the top result?
print(s.results.matches[0].doc)

# Print all matches. Best results are at the top.
for m in s.results.matches:
    print(m.doc)

Pass your own analyzer

When tinysearch.analyzer.SimpleEnglishAnalyzer does not satisfy your needs, you can write your own analyzer and pass it to the Search object.

An analyzer inherits from tinysearch.analyzer.base.Analyzer. It only need to implement analyze method. The analyze method accepts a string representing the document on the input, and returns a list of strings representing tokens (terms). Everything that you need to make it happen can be implemented there. See the docstring of the Analyzer base class.

You can then pass your analyzer to Search:

my_analyzer = MyOwnAnalyzer()

s = Search(docs, query, analyzer=my_analyzer)
print(s.results.count)

Under the hood

When you pass documents to the Search object, each document is tokenized and transformed for easier search. The same process is applied to the query.

Then each document is scored using the TF-IDF algorithm to find the best match, and matches are returned sorted to the user. The best match is at the top.

Performance

Performance is important since search engines typically respond to user queries, so it should generate results in a few seconds at most. More than that would appear as a significant delay.

The numbers below are dependent on the running machine, so they are just indicative.

gantt
title Search time for different dataset sizes [s]
dateFormat X
axisFormat %s

section 100
0.0, terms=1 : 0, 0.0s
0.0, terms=2 : 0, 0.0s
0.0, terms=3 : 0, 0.0s

section 1000
0.3, terms=1 : 0, 0.3s
0.2, terms=2 : 0, 0.2s
0.3, terms=3 : 0, 0.3s

section 10000
2.7, terms=1 : 0, 2.7s
2.7, terms=2 : 0, 2.7s
2.7, terms=3 : 0, 2.7s

section 52478
15.1, terms=1 : 0, 15.6s
15.4, terms=2 : 0, 15.1s
15.6, terms=3 : 0, 15.2s

Datasets of around 1000 entries might generate reasonable search times, which is the intended use case for TinySearch. Still, there is probably room for improvement.

Can we make it faster?

Most time is spent in analyzer, so improving performance means improving processing time of the analyzer. The default SimpleEnglishAnalyzer has already been highly optimized.

The next step to consider is to split the search into two phases: indexing and searching. Since analyzer needs to process every document, indexing can happen earlier in the process execution and searching when the user requests it. This has an additional benefit of indexing once and searching multiple times.

from tinysearch.index import Index
from tinysearch.search import Search

i = Index(docs)

# ...later...
s = Search(i, query)
print(s.results.matches[0])

License

See LICENSE.