Prepare Recipe Formula

Solved!
bhakuniv
Level 2
Prepare Recipe Formula

Hi

I am trying to extract a Column using 6 related columns, the output column will the input column with the highest associated score if atleast one of the input columns has a value other than (Undefined or Unknown)

Please refer to the attached excel (Col4 & Val4 are the output columns), is there any way to do this other than the formula step in prepare recipe

 

Thanks

VB

0 Kudos
1 Solution
MickaelH
Dataiker

Hi,

Since you need to:

  • define some "complex" rules to extract the values
  • generate 2 columns

I think it's the easiest and fastest way to do that.

If I understood correctly your use case but you won't be able to translate that into simpler processors. Maybe you can share the corresponding formula used in the spreadsheet that you used for the screenshot?

To illustrate the "python function" that could be used to solve your problem here is an example code snippet:

def process(row):
    pairs = [['Col1', 'Val1'], ['Col2', 'Val2'], ['Col3', 'Val3']]
    max_val = None
    max_col = None
    max_val_undef = None  # Used when all the column names are undefined/unknown
    for column_column_name, value_column_name in pairs:
        current_column = get_column_name(column_column_name, row)
        current_value = float(row.get(value_column_name, '0'))
        if current_column is None:
            max_val_undef = max(current_value, max_val_undef)
        else:
            if max_val is None or current_value > max_val:
                max_val = current_value
                max_col = current_column

    if max_col is None:
        computed_column = ''
        computed_value = max_val_undef
    else:
        computed_column = max_col
        computed_value = max_val
    row['Col4'] = computed_column
    row['Val4'] = computed_value
    return row


def get_column_name(column_column_name, row):
    result = row.get(column_column_name, None)
    if result.lower() in ['undefined', 'unknown']:
        return None
    return result

I hope this helps.

Mickaรซl | R&D at Dataiku

View solution in original post

0 Kudos
5 Replies
MickaelH
Dataiker

Hi,

Just to be sure I understood your use case. The input dataset will look like the one in the attached screenshot ?By that I mean that it will be made of the 6 columns and you need to both check for the first defined "ColN" value. If there are multiple "ColN" values, you then compare the "ValN" value and the N that needs to be kept is the one with the greater "ValN" ? (and another particular case where all the "ColN" are undefined/unknown")

You want to produce 2 new columns that will contains for each line the "ColN" and "ValN" values for the computed "N" ?

Mickaรซl | R&D at Dataiku
0 Kudos
bhakuniv
Level 2
Author

Hi Michael 

The input dataset will be the first 6 columns always no need to check on the count there

the output  should be the column and the corresponding value( where the value is the highest) however if any of the columns is undefined or unknown it should be ignored unless all 3 columns are undefined or unknown 

the expected output is highlighted in the attachment for each row 

0 Kudos
MickaelH
Dataiker

So to do that, you could use a "Python function" step in a "Prepare recipe".

You must use the mode "row: return a row for each row".

This way you will able to write all your logic to compute the "Col4" and "Val4" values.

You can then add another step in your prepare recipe to remove all the other columns if they are not useful.

Mickaรซl | R&D at Dataiku
0 Kudos
bhakuniv
Level 2
Author

Thanks Mickael, I assume there is no other way to get this done other than a custom formula

0 Kudos
MickaelH
Dataiker

Hi,

Since you need to:

  • define some "complex" rules to extract the values
  • generate 2 columns

I think it's the easiest and fastest way to do that.

If I understood correctly your use case but you won't be able to translate that into simpler processors. Maybe you can share the corresponding formula used in the spreadsheet that you used for the screenshot?

To illustrate the "python function" that could be used to solve your problem here is an example code snippet:

def process(row):
    pairs = [['Col1', 'Val1'], ['Col2', 'Val2'], ['Col3', 'Val3']]
    max_val = None
    max_col = None
    max_val_undef = None  # Used when all the column names are undefined/unknown
    for column_column_name, value_column_name in pairs:
        current_column = get_column_name(column_column_name, row)
        current_value = float(row.get(value_column_name, '0'))
        if current_column is None:
            max_val_undef = max(current_value, max_val_undef)
        else:
            if max_val is None or current_value > max_val:
                max_val = current_value
                max_col = current_column

    if max_col is None:
        computed_column = ''
        computed_value = max_val_undef
    else:
        computed_column = max_col
        computed_value = max_val
    row['Col4'] = computed_column
    row['Val4'] = computed_value
    return row


def get_column_name(column_column_name, row):
    result = row.get(column_column_name, None)
    if result.lower() in ['undefined', 'unknown']:
        return None
    return result

I hope this helps.

Mickaรซl | R&D at Dataiku
0 Kudos