LosAngelesPayPredictor.py

import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.model_selection import train_test_split
from fastai.tabular.all import *
from sklearn.ensemble import VotingRegressor
from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.base import BaseEstimator, RegressorMixin
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
import gradio as gr

df = pd.read_csv('City_Employee_Payroll__Current__20240915.csv', low_memory=False)
df = df.replace([np.inf, -np.inf], np.nan)

cat_names = ['EMPLOYMENT_TYPE', 'JOB_STATUS', 'MOU', 'GENDER', 'ETHNICITY', 'JOB_TITLE', 'DEPARTMENT_NO']
cont_names = ['PAY_YEAR', 'REGULAR_PAY', 'OVERTIME_PAY', 'ALL_OTHER_PAY']

df['PAY_RATIO'] = df['REGULAR_PAY'] / (df['OVERTIME_PAY'] + df['ALL_OTHER_PAY'] + 1)
df['TOTAL_NON_REGULAR_PAY'] = df['OVERTIME_PAY'] + df['ALL_OTHER_PAY']
cont_names.extend(['PAY_RATIO', 'TOTAL_NON_REGULAR_PAY'])

X = df[cat_names + cont_names].copy()
y = df['TOTAL_PAY'].copy()
for col in cat_names:
    X[col] = X[col].fillna('Unknown')

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
X_train_sample, _, y_train_sample, _ = train_test_split(X_train, y_train, train_size=0.3, random_state=42)

to = TabularPandas(df, procs=[Categorify, FillMissing, Normalize], cat_names=cat_names, cont_names=cont_names, y_names='TOTAL_PAY', splits=RandomSplitter(valid_pct=0.2)(range_of(df)))
dls = to.dataloaders(bs=64)

learn = tabular_learner(dls, layers=[200, 100, 50], metrics=rmse)
learn.fit_one_cycle(9)

class FastAIWrapper(BaseEstimator, RegressorMixin):
    def __init__(self, learn):
        self.learn = learn
    def fit(self, X, y):
        return self
    def predict(self, X):
        dl = self.learn.dls.test_dl(X)
        preds, _ = self.learn.get_preds(dl=dl)
        return preds.numpy().flatten()

preprocessor = ColumnTransformer(
    transformers=[
        ('num', StandardScaler(), cont_names),
        ('cat', OneHotEncoder(drop='first', sparse=False, handle_unknown='ignore'), cat_names)
    ])

model1 = FastAIWrapper(learn)
model2 = Pipeline([('preprocessor', preprocessor), ('regressor', LinearRegression())])
model3 = Pipeline([('preprocessor', preprocessor), ('regressor', DecisionTreeRegressor())])

ensemble = VotingRegressor(
    estimators=[('fastai', model1), ('lr', model2), ('dt', model3)],
    weights=[2, 1, 1]
)

ensemble.fit(X_train_sample, y_train_sample)

def predict_total_pay(gender, job_title, ethnicity):
    sample = pd.DataFrame({
        'GENDER': [gender],
        'JOB_TITLE': [job_title],
        'ETHNICITY': [ethnicity],
    })
    group = df[(df['GENDER'] == gender) & (df['JOB_TITLE'] == job_title) & (df['ETHNICITY'] == ethnicity)]
    if len(group) > 0:
        sample['EMPLOYMENT_TYPE'] = [group['EMPLOYMENT_TYPE'].mode().iloc[0]]
        sample['JOB_STATUS'] = [group['JOB_STATUS'].mode().iloc[0]]
        sample['MOU'] = [group['MOU'].mode().iloc[0]]
        sample['DEPARTMENT_NO'] = [group['DEPARTMENT_NO'].mode().iloc[0]]
        sample['REGULAR_PAY'] = [group['REGULAR_PAY'].mean()]
        sample['OVERTIME_PAY'] = [group['OVERTIME_PAY'].mean()]
        sample['ALL_OTHER_PAY'] = [group['ALL_OTHER_PAY'].mean()]
    else:
        job_group = df[df['JOB_TITLE'] == job_title]
        if len(job_group) > 0:
            sample['EMPLOYMENT_TYPE'] = [job_group['EMPLOYMENT_TYPE'].mode().iloc[0]]
            sample['JOB_STATUS'] = [job_group['JOB_STATUS'].mode().iloc[0]]
            sample['MOU'] = [job_group['MOU'].mode().iloc[0]]
            sample['DEPARTMENT_NO'] = [job_group['DEPARTMENT_NO'].mode().iloc[0]]
            sample['REGULAR_PAY'] = [job_group['REGULAR_PAY'].mean()]
            sample['OVERTIME_PAY'] = [job_group['OVERTIME_PAY'].mean()]
            sample['ALL_OTHER_PAY'] = [job_group['ALL_OTHER_PAY'].mean()]
        else:
            sample['EMPLOYMENT_TYPE'] = [df['EMPLOYMENT_TYPE'].mode().iloc[0]]
            sample['JOB_STATUS'] = [df['JOB_STATUS'].mode().iloc[0]]
            sample['MOU'] = [df['MOU'].mode().iloc[0]]
            sample['DEPARTMENT_NO'] = [df['DEPARTMENT_NO'].mode().iloc[0]]
            sample['REGULAR_PAY'] = [df['REGULAR_PAY'].mean()]
            sample['OVERTIME_PAY'] = [df['OVERTIME_PAY'].mean()]
            sample['ALL_OTHER_PAY'] = [df['ALL_OTHER_PAY'].mean()]
    sample['PAY_YEAR'] = [df['PAY_YEAR'].max()]
    sample['PAY_RATIO'] = sample['REGULAR_PAY'] / (sample['OVERTIME_PAY'] + sample['ALL_OTHER_PAY'] + 1)
    sample['TOTAL_NON_REGULAR_PAY'] = sample['OVERTIME_PAY'] + sample['ALL_OTHER_PAY']
    categorical_columns = ['GENDER', 'JOB_TITLE', 'ETHNICITY', 'EMPLOYMENT_TYPE', 'JOB_STATUS', 'MOU', 'DEPARTMENT_NO']
    for col in categorical_columns:
        sample[col] = sample[col].astype('object')
    prediction = ensemble.predict(sample)[0]
    return prediction

def gradio_predict(gender, ethnicity, job_title):
    predicted_pay = predict_total_pay(gender, job_title, ethnicity)
    return f"${predicted_pay:.2f}"

genders = df['GENDER'].dropna().unique().tolist()
ethnicities = df['ETHNICITY'].dropna().unique().tolist()
job_titles = sorted(df['JOB_TITLE'].dropna().unique().tolist())

iface = gr.Interface(
    fn=gradio_predict,
    inputs=[
        gr.Dropdown(choices=genders, label="Gender"),
        gr.Dropdown(choices=ethnicities, label="Ethnicity"),
        gr.Dropdown(choices=job_titles, label="Job Title")
    ],
    outputs=gr.Textbox(label="Predicted Total Pay"),
    title="LA City Employee Pay Predictor",
    description="Predict the total pay for LA City employees based on gender, ethnicity, and job title."
)

iface.launch()