Revision d22accebbfe21470e7b4f488e46c4aa6ae643b4a authored by Alexey Sergushichev on 07 July 2018, 16:43:01 UTC, committed by Alexey Sergushichev on 07 July 2018, 16:43:01 UTC
1 parent 337324b
developer-tutorial.Rmd
---
title: "Developing new tools"
author: "Daria Zenkova"
date: "`r Sys.Date()`"
output: rmarkdown::html_vignette
vignette: >
%\VignetteIndexEntry{Developing new tools}
%\VignetteEngine{knitr::rmarkdown}
%\VignetteEncoding{UTF-8}
---
## Introduction to project architecture
Project architecture includes two main components:
- R-package that contains implementations of used tools and methods;
- JavaScript code with client-side of the project.
Those two components are connected through OpenCPU as a bridge, and inside
the JavaScript implementation the RPC-call to server is used.
Consequently, in order to add new tool one needs:
- To implement an exported function in R;
- To describe graphical side of the tool in JavaScript;
- To process returned value inside RPC-call callback.
All of these steps will be described in details in this vignette.
## R-side implementation
Each analysis method in this package takes three compulsory arguments:
- `es` -- ExpressionSet object;
- `rows` -- same thing with rows;
- `columns` -- specified indices of columns that are taken into consideration.
Also, some methods require to replace NA values in series matrix in order
to be used, so usually also `replacena` argument is needed.
Other arguments depend on the method's specifics.
Before calculating the method, the considered data from the whole series matrix
needs to be extracted.
For that reason, the package has non-exported method `prepareData`,
that takes as arguments `es`, `columns`, `rows, `replacena`.
After the method is used, the result can be sent back in two ways:
- As a JSON object (use `jsonlite::toJSON` for that),
if the result is adequately small;
- As a file with ProtoBuf-serialized object (`protolite::serialize_pb`),
if it is large.
The approximate code structure is demonstrated here:
```{r, eval = FALSE}
# instead of ellipsis would be specific arguments
method <- function(es, rows = c(), columns = c(), replacea = "mean", ...) {
# Here may be some assertions
# Data preparation
data <- prepareData(es, rows, columns, replacena)
# Using method
res <- ...
# Sending back the result as JSON:
return(jsonlite::toJSON(res))
# Or as ProtoBuf
f <- tempfile(pattern = "pat", tmpdir = getwd(), fileext = ".bin")
writeBin(protolite::serialize_pb(res), f)
jsonlite::toJSON(f)
}
```
Remember, that your tool must be exported, fully documented and tested.
## JavaScript-side implementation
The structure of the tool description:
- `toString` -- method that returns tool's name;
- `gui` -- field's description, represented as an array of JSON-objects,
which have following values:
- `name` -- field's name;
- `value` -- default value;
- `type` -- field's type (most used types are: select, checkbox-list,
bootstrap-list, text);
- `multiple` -- if multiple values may be chosen
(relevant for checkbox-list);
- 'options' -- an array of values to choose from
(relevant for lists and select);
- `init` -- initialization of tool's input fields;
- `execute` -- main method of the tool, which includes:
1) processing of the input (`options.input.fieldName`);
2) processing of additional arguments, if they need to be derived
from the dataset and input;
3) RPC-call to OpenCPU-server
4) Processing the result.
Here is the approximate JavaScript description of the tool:
```{javascript, eval = FALSE}
phantasus.NewTool = function () {
};
phantasus.NewTool.prototype = {
// Tool name:
toString: function () {
return 'new';
},
// Initialization of tool's input fields
init: function (project, form) {
// Here is your initialization code
},
// Description of tool' GUI
gui: function (project) {
return [{
name : 'fieldName',
type : 'type',
options : [],
value : 'default_value'
}];
},
// Main function of the Tool
execute: function (options) {
var project = options.project;
// Getting the input
var field = options.input.fieldName;
// Reading actual dataset
var dataset = project.getSortedFilteredDataset();
// Each dataset has es session as a field
var es = dataset.getESSession();
// Get indices of selected rows and columns if they are selected
var trueIndices = phantasus.Util.getTrueIndices(dataset);
// Further calculation may proceed only when esSession is ready
es.then(function (essession) {
// Function arguments, there also should be method-specific arguments
var args = {
es: essession
};
if (trueIndices.rows.length > 0) {
args.rows = trueIndices.rows;
}
if (trueIndices.columns.length > 0) {
args.columns = trueIndices.columns;
}
// RPC-call to OpenCPU-server
var req = ocpu.call("methodName", args, function (session) {
session.getObject(function (success) {
// success -- returned result, needs to be processed
// Result getting depends on its type
// JSON:
var result = JSON.parse(success);
// after that you can proceed with result
// ProtoBuf:
var r = new FileReader();
var filePath = phantasus.Util.getFilePath(session,
JSON.parse(success)[0]);
r.onload = function (e) {
var contents = e.target.result;
var ProtoBuf = dcodeIO.ProtoBuf;
// message.proto is file with specified protocol
ProtoBuf.protoFromFile("./message.proto",
function (error, success) {
if (error) {
throw new Error(error);
}
var builder = success,
rexp = builder.build("rexp"),
REXP = rexp.REXP,
rclass = REXP.RClass;
var res = REXP.decode(contents);
var data = phantasus.Util.getRexpData(res, rclass);
var names = phantasus.Util.getFieldNames(res, rclass);
// here you can proceed with result
})
};
phantasus.BlobFromPath.getFileObject(filePath, function (file) {
r.readAsArrayBuffer(file);
});
})
}, false, '::' + dataset.getESVariable());
req.fail(function () {
// failed request procession
throw new Error("Method call failed" + req.responseText);
});
});
}
};
```
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...