Ethnicity from baby names NYC

names.py

from sklearn.ensemble import RandomForestClassifier
from sklearn import preprocessing

import pandas as pd
import numpy as np

from functools import partial


def encode_data(df, cols):
    encoders = dict()
    for c in cols:
        le = preprocessing.LabelEncoder()
        le.fit(df[c])
        df[c] = le.transform(df[c])
        encoders[c] = le
    return df, encoders

# encode categorical values as ints
df = pd.read_csv('nyc_babies.csv')
eth_norm = lambda r, race: race if race in r['ETHCTY'] else r['ETHCTY']
df['ETHCTY'] = df.apply(partial(eth_norm, race='ASIAN'), axis=1)
df['ETHCTY'] = df.apply(partial(eth_norm, race='BLACK'), axis=1)
df['ETHCTY'] = df.apply(partial(eth_norm, race='WHITE'), axis=1)

df, encoders = encode_data(df, ['BRTH_YR', 'GNDR', 'ETHCTY', 'NM'])

# duplicate row based on count field
rdf = pd.DataFrame(np.repeat(df.values, df['CNT'].values, axis=0))
rdf.columns = df.columns

# divide up training and test data
rdf['is_train'] = np.random.uniform(0, 1, len(rdf)) <= .75
train, test = rdf[rdf['is_train']], rdf[~rdf['is_train']]

# create, fit, predict
features = ['NM', 'GNDR', 'BRTH_YR']
clf = RandomForestClassifier(n_jobs=4)
clf.fit(train[features], train['ETHCTY'])
pred = clf.predict(test[features])
pred = encoders['ETHCTY'].inverse_transform(pred)
actual_names = encoders['ETHCTY'].inverse_transform(test['ETHCTY'])
ct = pd.crosstab(actual_names,
                 pred,
                 rownames=['actual'],
                 colnames=['preds']).apply(lambda r: r/r.sum(), axis=1)
print(ct)