back to listing index

schollz/pluck

[web search]
Original source (github.com)
Tags: text
Clipped on: 2018-04-05
Skip to content
  • You have unread notifications
  • Image (Asset 1/6) alt=

    Pluck text in a fast and intuitive way.

    pluck makes text extraction intuitive and fast. You can specify an extraction in nearly the same way you'd tell a person trying to extract the text by hand: "OK Bob, every time you find X and then Y, copy down everything you see until you encounter Z."

    In pluck, X and Y are called activators and Z is called the deactivator. The file/URL being plucked is parsed (or streamed) byte-by-byte into a finite state machine. Once all activators are found, the following bytes are saved to a buffer, which is added to a list of results once the deactivator is found. Multiple queries are extracted simultaneously and there is no requirement on the file format (e.g. XML/HTML), as long as its text.

    Why?

    pluck was made as a simple alternative to xpath and regexp. Through simple declarations, pluck allows complex procedures like extracting text in nested HTML tags, or extracting the content of an attribute of a HTML tag. pluck may not work in all scenarios, so do not consider it a replacement for xpath or regexp.

    Doesn't regex already do this?

    Yes basically. Here is an (simple) example:

    (?:(?:X.*Y)|(?:Y.*X))(.*)(?:Z)
    

    Basically, this should try and match everything before a Z and after we've seen both X and Y, in any order. This is not a complete example, but it shows the similarity.

    The benefit with pluck is simplicity. You don't have to worry about escaping the right characters, nor do you need to know any regex syntax (which is not simple). Also pluck is hard-coded for matching this specific kind of pattern simultaneously, so there is no cost for generating a new deterministic finite automaton from multiple regex.

    Doesn't cascadia already do this?

    Yes, there is already a command-line tool to extract structured information from XML/HTML. There are many benefits to cascadia, namely you can do a lot more complex things with structured data. If you don't have highly structured data, pluck is advantageous (it extracts from any file). Also, with pluck you don't need to learn CSS selection.

    Getting Started

    Install

    If you have Go1.7+

    go get github.com/schollz/pluck
    

    or just download from the latest releases.

    Basic usage

    Lets say you want to find URLs in a HTML file.

    $ wget nytimes.com -O nytimes.html
    $ pluck -a '<' -a 'href' -a '"' -d '"' -l 10 -f nytimes.html
    {
        "0": [
            "https://static01.nyt.com/favicon.ico",
            "https://static01.nyt.com/images/icons/ios-ipad-144x144.png",
            "https://static01.nyt.com/images/icons/ios-iphone-114x144.png",
            "https://static01.nyt.com/images/icons/ios-default-homescreen-57x57.png",
            "https://www.nytimes.com",
            "http://www.nytimes.com/services/xml/rss/nyt/HomePage.xml",
            "http://mobile.nytimes.com",
            "http://mobile.nytimes.com",
            "https://typeface.nyt.com/css/zam5nzz.css",
            "https://a1.nyt.com/assets/homepage/20170731-135831/css/homepage/styles.css"
        ]
    }

    The -a specifies activators and can be specified multiple times. Once all activators are found, in order, the bytes are captured. The -d specifies a deactivator. Once a deactivator is found, then it terminates capturing and resets and begins searching again. The -l specifies the limit (optional), after reaching the limit (10 in this example) it stops searching.

    Advanced usage

    Parse URLs or Files

    Files can be parsed with -f FILE and URLs can be parsed by instead using -u URL.

    $ pluck -a '<' -a 'href' -a '"' -d '"' -l 10 -u https://nytimes.com

    Use Config file

    You can also specify multiple things to pluck, simultaneously, by listing the activators and the deactivator in a TOML file. For example, lets say we want to parse ingredients and the title of a recipe. Make a file config.toml:

    [[pluck]]
    name = "title"
    activators = ["<title>"]
    deactivator = "</title>"
    
    [[pluck]]
    name = "ingredients"
    activators = ["<label","Ingredient",">"]
    deactivator = "<"
    limit = -1

    The title follows normal HTML and the ingredients were determined by quickly inspecting the HTML source code of the target site. Then, pluck it with,

    $ pluck -c config.toml -u https://goo.gl/DHmqmv
    {
        "ingredients": [
            "1 pound medium (26/30) peeled and deveined shrimp, tails removed",
            "2 teaspoons chili powder",
            "Kosher salt",
            "2 tablespoons canola oil",
            "4 scallions, thinly sliced",
            "One 15-ounce can black beans, drained and rinsed well",
            "1/3 cup prepared chipotle mayonnaise ",
            "2 limes, 1 zested and juiced and 1 cut into wedges ",
            "One 14-ounce bag store-bought coleslaw mix (about 6 cups)",
            "1 bunch fresh cilantro, leaves and soft stems roughly chopped",
            "Sour cream or Mexican crema, for serving",
            "8 corn tortillas, warmed "
        ],
        "title": "15-Minute Shrimp Tacos with Spicy Chipotle Slaw Recipe | Food Network Kitchen | Food Network"
    }

    Extract structured data

    Lets say you want to tell Bob "OK Bob, first look for W. Then, every time you find X and then Y, copy down everything you see until you encounter Z. Also, stop if you see U, even if you are not at the end." In this case, W, X, and Y are activators but W is a "Permanent" activator. Once W is found, Bob forgets about looking for it anymore. U is a "Finisher" which tells Bob to stop looking for anything and return whatever result was found.

    You can extract information from blocks in pluck by using these two keywords: "permanent" and "finisher". The permanent number determines how many of the activators (from the left to right) will stay activated forever, once activated. The finisher keyword is a new string that will retire the current plucker when found and not capture anything in the buffer.

    For example, suppose you want to only extract link3 and link4 from the following:

    <h1>Section 1</h1>
    <a href="link1">1</a>
    <a href="link2">2</a>
    <h1>Section 2</h1>
    <a href="link3">3</a>
    <a href="link4">4</a>
    <h1>Section 3</h1>
    <a href="link5">5</a>
    <a href="link6">6</a>

    You can add "Section 2" as an activator and set permanent to 1 so that only the first activator ("Section 2") will continue to remain activated after finding the deactivator. Then you want to finish the plucker when it hits "Section 3", so we can set the finisher keyword as this. Then config.toml is

    [[pluck]]
    activators = ["Section 2","a","href",'"']
    permanent = 1     # designates that the first 1 activators will persist
    deactivator = '"'
    finisher = "Section 3"
    

    will result in the following:

    {
        "0": [
            "link3",
            "link4",
        ]
    }

    More examples

    See EXAMPLES.md for more examples.

    Use as a Go package

    Import pluck as "github.com/schollz/pluck/pluck" and you can use it in your own project. See the tests for more info.

    Development

    $ go get -u github.com/schollz/pluck/...
    $ cd $GOPATH/src/github.com/schollz/pluck/pluck
    $ go test -cover
    

    Current benchmark

    The state of the art for xpath is lxml, based on libxml2. Here is a comparison for plucking the same data from the same file, run on Intel i5-4310U CPU @ 2.00GHz × 4. (Run Python benchmark cd pluck/test && python3 main.py).

    Language Rate
    lxml (Python3.5) 300 / s
    pluck 1270 / s

    A real-world example I use pluck for is processing 1,200 HTML files in parallel, compared to running lxml in parallel:

    Language Rate
    lxml (Python3.6) 25 / s
    pluck 430 / s

    I'd like to benchmark a Perl regex, although I don't know how to write this kind of regex! Send a PR if you do :)

    To Do

    • Allow OR statements (e.g '|").
    • Quotes match to quotes (single or double)?
    • Allow piping from standard in?
    • API to handle strings, e.g. PluckString(s string)
    • Add parallelism

    License

    MIT

    Acknowledgements

    Graphics by: www.vecteezy.com

Press h to open a hovercard with more details.