Effective Amazon Machine Learning

Applying recipes to the Titanic dataset

The recipe generated by Amazon ML for our dataset is as follows:

{    "groups": {        "NUMERIC_VARS_QB_50": "group('sibsp','parch')",        "NUMERIC_VARS_QB_100": "group('age')",        "NUMERIC_VARS_QB_10": "group('fare')"    },    "assignments": {},    "outputs": [        "ALL_BINARY",        "ALL_CATEGORICAL",        "quantile_bin(NUMERIC_VARS_QB_50,50)",        "quantile_bin(NUMERIC_VARS_QB_100,100)",        "quantile_bin(NUMERIC_VARS_QB_10,10)",        "ALL_TEXT"    ]}

All numeric values are quantile binned. No further processing is done on the text, binary, or categorical variables. The output section of the recipe shows that the numeric variables are replaced by the binned equivalent.

Further comments can be made on this recipe:

• The sibsp and parch variables are ...