app.py

import gradio as gr
import numpy as np
#from PIL import Image
#import requests
import random
import hopsworks
import joblib
import pandas as pd
from sklearn.preprocessing import OneHotEncoder, LabelEncoder, StandardScaler



project = hopsworks.login()
fs = project.get_feature_store()


mr = project.get_model_registry()
model = mr.get_model("titanic_modal_RF", version=1)
model_dir = model.download()
model = joblib.load(model_dir + "/titanic_model_RF.pkl")

def encode_features(df):
  non_numeric_features = ['embarked', 'sex', 'deck', 'title', 'family_size_grouped']
  for feature in non_numeric_features:
    df[feature] = LabelEncoder().fit_transform(df[feature])

  
  cat_features = ['pclass', 'sex', 'deck', 'embarked', 'title', 'family_size_grouped']
  encoded_features = []
  for feature in cat_features:
    encoded_feat = OneHotEncoder().fit_transform(df[feature].values.reshape(-1, 1)).toarray()
    n = df[feature].nunique()
    cols = ['{}_{}'.format(feature, n) for n in range(1, n + 1)]
    encoded_df = pd.DataFrame(encoded_feat, columns=cols)
    encoded_df.index = df.index
    encoded_features.append(encoded_df)
  df = pd.concat([df, *encoded_features], axis=1)
  drop_cols = ['deck', 'embarked', 'family', 'family_size', 'family_size_grouped',
             'name', 'parch', 'pclass', 'sex', 'sibsp', 'ticket', 'title']

  df.drop(columns=drop_cols, inplace=True)
  return df #Return a df with encoded labels


def titanic(age, deck, embarked, name, family_size, family_size_grouped, fare, is_married, parch, sibsp, sex, ticket_number, title):

    pclass = 0
    if deck == "ABC":
        pclass = 1

    elif deck == "DE":
        pclass = random.choice([2,3])
        
    elif deck == "FG":
        pclass = random.choice([2,3])
    else:
        pclass = 3
    
    df = pd.DataFrame({ 
         "age" : [age],
         "deck" : [deck],
         "embarked" : [embarked],
         "family" : [name.split()[1]],
         "family_size" : [family_size],
         "family_size_grouped" : [family_size_grouped],
         "fare" : [fare],
         "is_married" : [is_married],
         "name" : [name.split()[1] + ", " + name.split()[0]],
         "parch" : [parch],
         "pclass" : [pclass],
         "sex" : [sex],
         "sibsp" : [sibsp],
         "ticket" : [str(ticket_number)],
         "ticket_frequency" : [1],
         "title" : [title] 
         })
    return df

def predictor(age, deck, embarked, name, family_size, family_size_grouped, fare, is_married, parch, sibsp, sex, ticket_number, title):
    feature_view = fs.get_feature_view(name="titanic_modal_2", version=1)
    batch_data = feature_view.get_batch_data()
    row = titanic(age, deck, embarked, name, family_size, family_size_grouped, fare, is_married, parch, sibsp, sex, ticket_number, title)
    batch_data = batch_data.append(row)
    encoded = encode_features(batch_data)
    y_pred = model.predict(encoded)
    prediction = y_pred[y_pred.size-1]
    if prediction == 0:
        survived = "This person survived"
    if prediction == 1:
        survived = "This person did not survive"
    return survived

demo = gr.Interface(
    fn=predictor,
    title="Titanic Survivor Predictive Analytics",
    description="Titanic survivor experiment.",
    allow_flagging="never",
    inputs=[
        gr.inputs.Number(label="age (10-99)"),
        gr.inputs.Textbox(label="Deck (ABC, DE, FG, or M)"),
        gr.inputs.Textbox(label="Embarked (S, C, or Q)"),
        gr.inputs.Textbox(label="Name of person"),
        #gr.inputs.Textbox(label="Family name"),
        gr.inputs.Number(label="Size of family  (integer)"),
        gr.inputs.Textbox(label="Family Size (Alone, Small, Medium or Large)"),
        gr.inputs.Number(label="Fare (Integer)"),
        gr.inputs.Number(label="Is the person married (0 for no, 1 for yes)"),
        gr.inputs.Number(label="Number of Parent/Children onboard (Integer)"),
        gr.inputs.Number(label="Number of siblings onboard (Integer)"),
        gr.inputs.Textbox(label="Sex"),
        gr.inputs.Number(label="Ticket number"),
        gr.inputs.Textbox(label="Title (Mr, Miss, etc)"),
        ],
    outputs="text")

demo.launch()