catagorical and ordinal feature specification and encoding

[hjwilli]

Be able to handle data that has categorical or ordinal features and preprocess data appropriately depending on the algorithm being run, see label vs onehot encoding

• @hjwilli Update dataset upload api '/api/v1/datasets' to allow the specification of categorical or ordinal features
• Allow projects.json to specify the strategy that should be used when pre-processing categorical data for a particular algorithm
• Update machine to preprocess ordinal and categorical features as appropriate per the strategy and the dataset field specification, assume that machine will be able to pull a version of the data from the lab api that has already been label encoded

Needed for some features of #119

[hjwilli]

Should the dataset profiles work off of label encoded data? onehot data?

[weixuanfu]

• Use pipeline with ColumnTransformer
• Two steps of ColumnTransformer for catagorical and ordinal features respectively

[weixuanfu]

• Add ordinal map into files API and use it in OrdinalEncoder
• Add "categorical_encoding_strategy" into projects.json for each ML algorithms

[hjwilli]

Hi @weixuanfu, this seems great, the only update I see is that the response from the api to get a dataset the ordinal and categorical features will be in a slightly different format.

Once the code has been updated for this api spec and there is a unit test that mocks the api response with ordinal features, this should be ready for a pull request or direct merge into master.

Example response from a GET to http://lab:5080/api/v1/datasets/$datasetId:

{
    "_id": "5bf4841c9fc83c002cdbf5e6",
    "name": "appendicitis",
    "username": "testuser",
    "files": [
        {
            "_id": "5bf4841c9fc83c002cdbf5e8",
            "_raw_id" : "5bf4841c9fc83c00adfadfe", 
            "filename": "appendicitis.csv",
            "mimetype": "text/csv",
            "dependent_col": "class",
            "categorical_features" : ["cat_feat_1", "cat_feat_2"],
            "ordinal_features" : {"ord_feat_1" : ["MALE", "FEMALE"], "ord_feat_2" : ["FIRST", "SECOND", "THIRD"]},
            "timestamp": 1542751267786
        }
    ],
    "metafeatures": {
       ....
    }
}

[weixuanfu]

@hjwilli ok, I will push a commit for supporting this API format.