|
import gradio as gr |
|
import numpy as np |
|
from PIL import Image |
|
import requests |
|
|
|
import hopsworks |
|
import joblib |
|
|
|
project = hopsworks.login() |
|
fs = project.get_feature_store() |
|
|
|
|
|
mr = project.get_model_registry() |
|
model = mr.get_model("titanic_modal", version=1) |
|
model_dir = model.download() |
|
model = joblib.load(model_dir + "/titanic_model.pkl") |
|
|
|
|
|
def titanic(Pclass, Sex, Age, SibSp, Parch, Embarked): |
|
input_list = [] |
|
|
|
input_list.append(Pclass) |
|
input_list.append(Sex) |
|
input_list.append(Age) |
|
input_list.append(SibSp) |
|
input_list.append(Parch) |
|
input_list.append(Embarked) |
|
|
|
res = model.predict(np.asarray(input_list).reshape(1, -1)) |
|
if res[0]==0: |
|
link ="https://github.com/JeetNimbhorkar/TitanicLab1/raw/d9482baa7cbe47d0a8d5dcbe93e1ce7c0b2538a2/didnotsurvive.png" |
|
else: |
|
link = "https://github.com/JeetNimbhorkar/TitanicLab1/raw/d9482baa7cbe47d0a8d5dcbe93e1ce7c0b2538a2/survived.png" |
|
|
|
|
|
|
|
titanic_url=link |
|
img = Image.open(requests.get(titanic_url, stream=True).raw) |
|
return img |
|
|
|
demo = gr.Interface( |
|
fn=titanic, |
|
title="Titanic survival Predictive Analytics", |
|
description="Enter passanger details to predict survival in Titanic", |
|
allow_flagging="never", |
|
inputs=[ |
|
gr.inputs.Number(default=1.0, label="Pclass (Enter 1,2 or 3)"), |
|
gr.inputs.Number(default=1.0, label="Sex (0 for Male, 1 for Female)"), |
|
gr.inputs.Number(default=1.0, label="Age"), |
|
gr.inputs.Number(default=1.0, label="SibSp (Enter 0,1,2,3,4,5 or 8)"), |
|
gr.inputs.Number(default=1.0, label="Parch (Enter 0,1,2,3,4,5 or 6)"), |
|
gr.inputs.Number(default=1.0, label="Embarked (Enter 0 for C, 1 for Q and 2 for S)") |
|
], |
|
outputs=gr.Image(type="pil")) |
|
|
|
demo.launch() |
|
|
|
|