app.py

import pickle
import pandas as pd
import shap
from shap.plots._force_matplotlib import draw_additive_plot
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt
import xgboost as xgb
from scipy import stats
from gradio import Interface
from gradio.components import Markdown, Row, Column, Slider, Dropdown, CheckboxGroup, Button, Textbox, Dataframe
from category_encoders import TargetEncoder

# Load DataFrames
bb_df = pd.read_csv('beer_brewery_imputed.csv')
bb_df_percentile = pd.read_csv('bb_df_testing.csv')
aslin_example_df_full = pd.read_csv('App_Example_Aslin_Update.csv')
aslin_example_df = aslin_example_df_full.drop(['Number of Ratings Beer'], axis=1)
aslin_example_df['ABV'] = aslin_example_df['ABV']*100

# Load pickle files
with open('unique_brewery_file.pickle', 'rb') as file:
    unique_breweries_list = pickle.load(file)

loaded_model_regressor = pickle.load(open("XGB_Untappd_regressor_FlavorBreakout.pkl", 'rb'))
loaded_model = pickle.load(open("XGB_Untappd_4_classifier_FlavorBreakout.pkl", 'rb'))
loaded_enc_regressor = pickle.load(open("target_encoder_regressor_flavorbreakout.pkl", 'rb'))
loaded_enc_classification = pickle.load(open("target_encoder_classification_flavorbreakout.pkl", 'rb'))

# Define choices
region_choices = ['Far West','Great Lakes','Mideast','Non-Con','Northeast','OTHER','Plains','Rocky Mountain','Southeast','Southwest']
style_choices = ['Altbier', 'Barleywine - American', 'Belgian Blonde', 'Blonde Ale', 'Bock - Doppelbock', 'Brown Ale - American', 'Cream Ale', 'Dark Ale', 'IPA - American', 'Lager - American', 'Pilsner - German', 'Stout - American', 'Wheat Beer - American Pale Wheat']
brewery_style_choices = ['Brew Pub', 'Cidery', 'Contract Brewery','Macro Brewery','Micro Brewery', 'Nano Brewery', 'OTHER', 'Regional Brewery']
state_choices = ['AK', 'AL', 'AR', 'AZ', 'CA', 'CO', 'CT', 'DC', 'DE', 'FL', 'GA', 'HI', 'IA', 'ID', 'IL', 'IN', 'KS', 'KY', 'LA', 'MA', 'MD', 'ME', 'MI', 'MISSING', 'MN', 'MO', 'MS', 'MT', 'NC', 'ND', 'NE', 'NH', 'NJ', 'NM', 'NV', 'NY', 'OH', 'OK', 'OR', 'PA', 'RI', 'SC', 'SD', 'TN', 'TX', 'UT', 'VA', 'VT', 'WA', 'WI', 'WV', 'WY']

# Define flavors and hops
flavors_list = ['Apple', 'Apricot', 'Berry', 'Bitter', 'Caramel', 'Chocolate', 'Citrus', 'Clove', 'Coffee', 'Dry', 'Earthy', 'Fig', 'Floral', 'Fruity', 'Funky', 'Grapefruit', 'Hazelnut', 'Herbal', 'Malt', 'Nutmeg', 'Nutty', 'Peach', 'Pear', 'Peat', 'Pepper', 'Pine', 'Plum', 'Resin', 'Salty', 'Smoky', 'Sour', 'Spicy', 'Strawberry', 'Sweet', 'Tart', 'Toast', 'Toffee', 'Tropical', 'Vanilla']
hops_list = ['Amarillo', 'Cascade', 'Centennial', 'Chinook', 'Citra', 'Columbus', 'Crystal', 'Fuggle', 'Galaxy', 'Golding', 'Hallertau', 'Magnum', 'Mosaic', 'Noble', 'Nugget', 'Saaz', 'Simcoe', 'Tettnang', 'Warrior', 'Willamette']

# Setup SHAP
explainer = shap.Explainer(loaded_model_regressor)

# Function to extract selected items
def extract_selected_items(row, items_list, prefix):
    return [item for item in items_list if row[prefix + item] == 1]

# Function to filter beers
def filter_beers(style, state):
    target_abv = 0.03
    filtered_df = bb_df[(bb_df['Style'] == style) & (bb_df['State'] == state)].copy()
    filtered_df['ABV_diff'] = abs(filtered_df['ABV'] - target_abv)
    filtered_df.loc[filtered_df['ABV_diff'] > target_abv, 'ABV_diff'] = target_abv
    filtered_df = filtered_df[filtered_df['ABV_diff'] <= target_abv]
    sorted_df = filtered_df.sort_values(by='Number of Ratings Beer', ascending=False)
    limited_df = sorted_df.head(5)[['Brewery', 'Beer Name', 'Average Rating Beer', 'Number of Ratings Beer', 'Style', 'ABV', 'IBU', 'State']]
    limited_df = limited_df.rename(columns={'Average Rating Beer': 'Avg Rating', 'Number of Ratings Beer': '# Ratings'})
    limited_df['ABV'] = (limited_df['ABV'] * 100).round(2).astype(str) + '%'  
    limited_df['Avg Rating'] = limited_df['Avg Rating'].round(2)  
    limited_df['IBU'] = limited_df['IBU'].astype(int)
    limited_df['# Ratings'] = limited_df['# Ratings'].apply(lambda x: '{:,}'.format(x))
    return limited_df

# Main function
def main_func(BeerName, ABV, IBU, Style, BreweryStyle, Region, State, Flavor_Group, Hop_Group):
    flavors_selected = [flavor for flavor in flavors_list if flavor in Flavor_Group]
    hops_selected = [hop for hop in hops_list if hop in Hop_Group]
    new_row = pd.DataFrame(columns=['ABV', 'IBU', 'Style', 'Brewery Style', 'Region', 'State'] + flavors_list + hops_list)
    new_row.loc[0] = [float(ABV), float(IBU), Style, BreweryStyle, Region, State] + [1 if flavor in Flavor_Group else 0 for flavor in flavors_list] + [1 if hop in Hop_Group else 0 for hop in hops_list]
    new_row[['ABV', 'IBU']] = new_row[['ABV', 'IBU']].astype(float)
    new_row['ABV'] = new_row['ABV']/100
    
    new_row_class = new_row.copy()
    new_row_regress = new_row.copy()
    new_row_encoded_class = loaded_enc_classification.transform(new_row_class)
    new_row_encoded_regressor = loaded_enc_regressor.transform(new_row_regress)
    
    prob = loaded_model.predict_proba(new_row_encoded_class)
    score_predict = loaded_model_regressor.predict(new_row_encoded_regressor)[0]
    score_predict = round(score_predict, 2)
    score_predict_str = str(score_predict)
    
    shap_values = explainer(new_row_encoded_regressor)
    plot = shap.plots.bar(shap_values[0], max_display=7, order=shap.Explanation.abs, show_data='auto', show=False)
    plt.tight_layout()
    local_plot = plt.gcf()
    plt.close()
    
    similar_beers = filter_beers(Style, State)
    
    nr_state_p = new_row['State'][0]
    nr_style_p = new_row['Style'][0]
    
    overall_df = bb_df_percentile
    state_df = bb_df_percentile[bb_df_percentile['State'] == nr_state_p]
    style_overall_df = bb_df_percentile[bb_df_percentile['Style'] == nr_style_p]
    style_state_df = bb_df_percentile[(bb_df_percentile['Style'] == nr_style_p) & (bb_df_percentile['State'] == nr_state_p)]
    
    percent_100 = 1/1
    percentile_overall = stats.percentileofscore(overall_df['Average Rating Beer'], score_predict).round(1)/100
    percentile_state = stats.percentileofscore(state_df['Average Rating Beer'], score_predict).round(1)/100
    percentile_style_overall = stats.percentileofscore(style_overall_df['Average Rating Beer'], score_predict).round(1)/100
    percentile_style_state = stats.percentileofscore(style_state_df['Average Rating Beer'], score_predict).round(1)/100
    
    percentile_dict0 = {
        #"Percentile Scale": [percent_100],
        "Overall in USA": [percentile_overall],
        f" Overall in {nr_state_p}": [percentile_state],
        f"{nr_style_p}s in USA": [percentile_style_overall],
        f"{nr_style_p}s in {nr_state_p}": [percentile_style_state]
    }

    title_text = f"<h2><center><b>{BeerName} Rating Prediction Percentiles</b></center></h2>"
    
    #return local_plot, similar_beers, score_predict_str, percentile_dict0
    return local_plot, similar_beers, score_predict_str, percentile_dict0, title_text


# Create the UI
title = "<center><b>🍻 **Untappd Beer Rating Predictor**🍻</b></center>"
description1 = """
This app takes user-inputted beer characteristics to predict the most likely Untappd rating the beer would receive if brewed. The ratings are calculated on a 5-point scale using a model based on data pulled in June 2023.

- The model has a <b>Mean Average Error (MAE) of .12</b> and <b> Root Mean Squared Error (RMSE) of .16</b>
- The input variables in this model <b> explain 65% of the variation </b>in the Untappd beer score

"""

theme = gr.themes.Default()

with gr.Blocks(title=title, theme=theme) as demo:
    Markdown(f"# {title}")
    Markdown(description1)
   
    submit_btn1 = gr.Button("Predict")
    with Row():        
        with Column():
            Markdown("<h2><center><b>New Beer Characteristics</b></center></h2>")
            BeerName = Textbox(label='Beer Name', value='New Beer 1')
            ABV = Slider(label="ABV %", minimum=0, maximum=20, value=4.5, step=.1)
            IBU = Slider(label="IBU", minimum=0.0, maximum=200, value=41, step=1)
            Style = Dropdown(choices=style_choices, label="Beer Style", value='IPA - Session')
            BreweryStyle = Dropdown(choices=brewery_style_choices, label="Brewery Style", value='Micro Brewery')
            Region = Dropdown(choices=region_choices, label="USA Region", value='Mideast')
            State = Dropdown(choices=state_choices, label="State", value='VA')
            Flavor_Group = CheckboxGroup(choices=flavors_list, label="Flavors")
            Hop_Group = CheckboxGroup(choices=hops_list, label="Hops")
            
        with gr.Column(visible=True) as output_col:
            Markdown("<h2><center><b>Untappd Rating Prediction</b></center></h2>")
            score_predict_str = gr.Label(label="XGBoost Regressor")
            Markdown("<h2><center><b>Characteristics Affecting Rating Prediction</b></center></h2>")
            local_plot = gr.Plot(label='Shap:')
            percentile_title = Markdown(f"<h2><center><b>Untappd Rating Percentiles</b></center></h2>")
            percentile_dict0 = gr.Label( label='Percentile', show_label=True, value = float )
            
    submit_btn2 = gr.Button("Predict")
    Markdown("""---""")

    with gr.Row():
        with gr.Column():
            Markdown("<h1><center><b>Similar Beers</b></center></h1>")
            similar_beers = Dataframe(label="", type="pandas")
            

    submit_btn1.click(
        main_func,
        [BeerName, ABV, IBU, Style, BreweryStyle, Region, State, Flavor_Group, Hop_Group],
        #[local_plot, similar_beers, score_predict_str, percentile_dict0],
        [local_plot, similar_beers, score_predict_str, percentile_dict0, percentile_title],
        api_name="Untappd_Rating_Model")
    
    submit_btn2.click(
        main_func,
        [BeerName, ABV, IBU, Style, BreweryStyle, Region, State, Flavor_Group, Hop_Group],
        #[local_plot, similar_beers, score_predict_str, percentile_dict0],
        [local_plot, similar_beers, score_predict_str, percentile_dict0, percentile_title],
        api_name="Untappd_Rating_Model1")

    examples = []
    for _, row in aslin_example_df.iterrows():
        example = [
            row['Beer Name'], row['ABV'], row['IBU'], row['Style'],
            row['Brewery Style'], row['Region'], row['State'],
            extract_selected_items(row, flavors_list, ''), 
            extract_selected_items(row, hops_list, '')
        ]
        examples.append(example)
        
    Markdown("<h1><center><b>Example Inputs from Aslin Beer Menu</b></center></h1>")
    gr.Examples(examples[:10],
                [BeerName, ABV, IBU, Style, BreweryStyle, Region, State, Flavor_Group, Hop_Group],
                #[local_plot, similar_beers, score_predict_str, percentile_dict0],
                [local_plot, similar_beers, score_predict_str, percentile_dict0, percentile_title],

                main_func, 
                cache_examples=True, label="Aslin Beer List")

demo.launch()