The Loan Desk — hard Machine Learning in Code problem

The Loan Desk

A small bank wants a model that says "yes" or "no" to loan requests. Its table mixes several kinds of column, and each needs its own treatment.

Column	Kind	What to do
`income`	number, may be `None`	fill with the median, then scale
`debt`	number, may be `None`	fill with the median, then scale
`job`	word	one-hot, ignoring unseen jobs
`has_guarantor`	already `0` or `1`	pass through untouched

Task: write loan_desk(table, labels, new_rows, seed).

table and new_rows are dicts with those four columns. labels holds "yes" or "no" for each row of table.
Make income and debt float so each None becomes np.nan.
Split with train_test_split(X, y, test_size=0.25, random_state=seed, stratify=y).
Build a ColumnTransformer with ("numbers", ..., ["income", "debt"]) and ("words", ..., ["job"]), using remainder="passthrough" so has_guarantor survives.
Put it in a Pipeline with KNeighborsClassifier(n_neighbors=3) last, and fit it on the training rows only.
Return {"width": number of columns the ColumnTransformer outputs, "score": test score rounded to 3 decimal places, "predictions": list of predictions for new_rows}.

Examples

table = {"income": [52, 61, None, 23, 19, 70, 25, 58, None, 21, 66, 27], "debt": [5, 8, 4, 30, 28, None, 25, 6, 9, None, 7, 31], "job": ["nurse", "chef", "nurse", "chef", "driver", "nurse", "driver", "driver", "chef", "nurse", "driver", "chef"], "has_guarantor": [1, 1, 0, 0, 0, 1, 0, 1, 1, 0, 1, 0]}

labels = ["yes", "yes", "yes", "no", "no", "yes", "no", "yes", "yes", "no", "yes", "no"]

new_rows = {"income": [60, 22], "debt": [None, 29], "job": ["chef", "pilot"], "has_guarantor": [1, 0]}

seed = 0

returns {'width': 6, 'score': 1.0, 'predictions': ['yes', 'no']}

table = {"income": [40, 45, 15, 18, 42, 12, 48, 16], "debt": [2, None, 20, 22, 3, 25, 1, None], "job": ["a", "b", "a", "b", "c", "c", "a", "b"], "has_guarantor": [1, 1, 0, 0, 1, 0, 1, 0]}

labels = ["yes", "yes", "no", "no", "yes", "no", "yes", "no"]

new_rows = {"income": [44, 14], "debt": [2, 21], "job": ["c", "d"], "has_guarantor": [1, 0]}

seed = 0

returns {'width': 6, 'score': 1.0, 'predictions': ['yes', 'no']}

Hints

1import pandas as pd

2from sklearn.compose import ColumnTransformer

3from sklearn.impute import SimpleImputer

4from sklearn.model_selection import train_test_split

5from sklearn.neighbors import KNeighborsClassifier

6from sklearn.pipeline import Pipeline

7from sklearn.preprocessing import OneHotEncoder, StandardScaler

10def loan_desk(table, labels, new_rows, seed):

11 pass