Skip to content

Latest commit

 

History

History
179 lines (134 loc) · 7.24 KB

README.org

File metadata and controls

179 lines (134 loc) · 7.24 KB
Raw

p-search

p-search is an Emacs tool to find things. It combines concepts from information retrievial and Bayesian search theory to assist a user in finding documents.

Boolean searches (i.e. the document contains the word “X”), while simple and useful, ofthen don’t match the prior beliefs of the searcher concerning where the relevant files are. Often times the searcher has specific ideas as to where the document is located. Like what type of file it is, who authored the document, when the document was created. The searcher often times isn’t fully confident as to how the seach terms appear or whether they appear at all.

./documents/screenshot.png

Instalation

Until p-search is available on ELPA/MELPA, you will have to install this package manually. The only dependency of p-search is heap.

Using Quelpa:

(quelpa '(p-search :repo "zkry/p-search" :fetcher github))

Using Straight:

(use-package p-search :straight (:host github :repo "https://github.com/zkry/p-search.git"))

Using Elpaca:

(use-package p-search :elpaca (:host github :repo "https://github.com/zkry/p-search.git"))

Usage

A search session can be initiated with the p-search command. The command will set up the session to search for files either in the projects directory (see project.el) if a project exists or the current directory. Execute p-search with the prefix C-u to instantiate an empty session (TODO).

The p-search buffer

The p-search session is composed of three main sections: Candidate Generators, Priors, and Search Results.

./documents/p-search-demo-1.png

Candidate Generators

Candidate generators are the parts of the search session that enumerate all possible search candidates. A search candidate is an entity with a set of key/value properties, 'content and 'title being mandetory. Other properties may exist which will allow you to use additional prior functions. In the p-search session run p-search-add-candidate-generator (C)to add a new candidate generator.

You can delete a prior with the command p-search-kill-entry-at-point.

Priors

The Priors section is the part where you add search criteria to your session. Run p-search-add-prior (P) to add a prior function.

First you must select the type of prior you want to add. Then you will have to configure the prior. It will first prompt you for any fields that are mandetory.

After that, a new transient menu will appear, allowing you configure the prior. Each prior function will have its own set of inputs and options, but each one will let you set its importance and whether the complement should be taken.

You can delete a prior with the command p-search-kill-entry-at-point (k).

Calculation

Each candidate document is given a score from each prior function depending on how well the prior function matches.

So for example, suppose you have a text query search. The query will rank each document on a scale from 0 to 1. This score is then modified by the importance. If you assign a high importance, then the probabilities will be pushed to the extremes. A low importance pushes the probabilities to 0.5, thus lowering its impact.

So for example, if a text search query marked a document as highly relevant, 0.7, but was given a low importance, its probability may be modified to 0.55, thus lowering its impact. On the other hand, if a text query matches poorly giving a score of 0.3 but its importance is low, then its probability will be raised to perhaps 0.45.

[CANDIDATE GENERATOR]
  |
  |                    [PRIOR_X]              [PRIOR_Y]
  |
  |\-- DOC_A ->  importance_X(Score_X_A) ✖ importance_Y(Score_Y_A)
  |
  |\-- DOC_B ->  importance_X(Score_X_B) ✖ importance_Y(Score_Y_B)  ...
  |
  \--- DOC_C ->  importance_X(Score_X_C) ✖ importance_Y(Score_Y_C)

Text Search

Text search is a prominent component in p-search. While text search functions the same way as other prior functions (resulting in a score of 0 to 1), the mecahnisms behind it are more complex.

You can create a text query by selecting “text query” in the transient menu when running p-search-add-prior.

You will then be prompted for your query. Depending on the query you write, one or more processes will be created to perform the search.

As mentioned earlier, each search candidate document has a property 'content. The text search is performed on this field. As you can probably immagine, having to search each document on a single Emacs Lisp thread is slow, so each candidate generator function can have a quicker method to perform the search. This is why you see the search tool like :grep or :rg on the FILESYSTEM candidate generator. When performing a text query on documents coming from this, it will rely on this tool to perform the search.

For the text query, each search result is space separated. So if you type teacher student school it will perform three separate searches for the three terms. Each term will generate its own score for each document and they will then be combined to form a final score. You can use quotes to group words to search something as a whole, thus "teacher student school" will perform one search with the words in a sequence.

Unquoted terms will be processed into multiple variants and searched in parallel. So for example teacherStudentSchool will search both “teacherstudentschool” (case insensitive), but also “teacher_student_school”, “teacher-student-school” (with a lower score), and the sepearate terms “teacher”, “student”, and “school” (given even a lower score).

You can boost a term with ^ so that teacher student^ school will give a boost to student. You can also specify a numeric boost, as in teacher student^2 school^3.

You can search for terms that occur near to one another with the (term1 term2 ...)~ syntax. Depending on the value of p-search-default-near-line-length, the items will be required to be within a certain number of lines from one another.

Observation

p-search will only show you the first p-search-top-n values of the search results. If you are not seeing relevant results you may want to consider adding search criteria. You can also run the command p-search-observe to lower the probability of a particular result. Doing so will lower the probability of the item by multiplying it by 0.3. With prefix C-u p-search-observe, you can specify the probability. After you perform the observation the probabilities will be recalculated and the results will update.

p-search-peruse-mode

p-search-peruse-mode is an experimental global minor mode, that when active, will track the percentage of files that you viewed. The view percentage will be updated in the search results section.