My rough ideas on making a jsonstat object

[MansMeg]

Hi @hmalmedal and @ajschumacher !

Here is a small pull request with the ideas I have on how to extend the usage of the rjstat package. I suggest to have a new json-stat object and from this it is possible to inherit the json stat type classes (dataset etc.).

This makes it possible to work directly with jsonstat object, print them and add methods or relevance. It is also possible to check that the json object is a valid json-stat object (now this validation is really light-weight - hopefully we can use a json schema validator later on if it gets implemented). My next step was to implement a "get_data.frame" method or similar using the parse_dataset function.

What do you think?

[hmalmedal]

You propose a new API that can co-exist with the old API? I think that is a sound approach.

[MansMeg]

Alright! Now I'm done with my first additions and all tests passes. As you mention @hmalmedal the old API is still working. I've added a new API where one can work directly with jsonstat objects since this is of interest for me right now. The data.frame parsing is though plugged in to parse_dataset() function.

Some example code you could try out if you would like to see what I have added to the package:

js <- as.jsonstat("http://json-stat.org/samples/oecd.json")
js
js[1,2:5,]
js[[1,2:5,]]
as.data.frame(js[1,2:5,])
as.array(js[1,2:5,])
js[1,1,1] <- 100
as.data.frame(js[1,2:5,])

A testsuite for the new functionality is added as well.

[hmalmedal]

Looks nice.

Something I found difficult when I worked on this, was to keep track of which elements that should use unbox. Just using auto_unbox won't work.

[MansMeg]

Hi! Thanks!

Thats a good suggestion. I guess that problem should be in the as.character() method? I have a unit test for this and it does not fail?

[hmalmedal]

> dataset <- as.jsonstat("tests/testthat/dataset.json")
> as.character(dataset)
{
  "version": "2.0",
  "class": "dataset",
  "label": "A dataset",
  "value": 1,
  "id": "testdimension",
  "size": 1,
  "dimension": {
    "testdimension": {
      "label": "A dimension",
      "category": {
        "index": "testcategory",
        "label": {
          "testcategory": "A category"
        }
      }
    }
  }
} 

Everything is unboxed.

Compare with the original file:

{
  "version": "2.0",
  "class": "dataset",
  "label": "A dataset",
  "value": [
    1
  ],
  "id": [
    "testdimension"
  ],
  "size": [
    1
  ],
  "dimension": {
    "testdimension": {
      "label": "A dimension",
      "category": {
        "index": [
          "testcategory"
        ],
        "label": {
          "testcategory": "A category"
        }
      }
    }
  }
}

[MansMeg]

Alright! Thanks! Ill put that into a unit test!

How do you solve this? Do you have a function to "box" correctly?

[MansMeg]

I now have this problem as a failing unit test. Thanks again!

[hmalmedal]

You have to use the unbox function from jsonlite on the list elements in question.

http://search.r-project.org/library/jsonlite/html/unbox.html

[MansMeg]

Great! I now fixed the unboxing (a little tricky) by creating a unbox_jsonstat() function. this can be improved further on using real inheritage. Now mostly a workaround.

[hmalmedal]

Good. Let me know when you think it is ready for merging.

[MansMeg]

Sure! I'll be working some more today. I'll let you know.

[MansMeg]

Alright @hmalmedal ! Now you can merge the pull request. I won't work much more the next couple of days.