shvuuuu's picture
Update app.py
5f78b08
raw
history blame
7.1 kB
import gradio as gr
import pickle
def example1():
model = pickle.load(open('model.pkl', 'rb'))
input_model = [[65,1.8,2,0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1]]
pred=model.predict(input_model)
churn = "False"
if pred[0] == 1:
churn = "He Will Churn"
elif pred[0] == 0:
churn = "He Will Not Churn"
return churn
def example2():
model = pickle.load(open('model.pkl', 'rb'))
input_model = [[41,2,2,0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 0, 0]]
pred=model.predict(input_model)
churn = "False"
if pred[0] == 1:
churn = "He Will Churn"
elif pred[0] == 0:
churn = "He Will Not Churn"
return churn
def example3():
model = pickle.load(open('model.pkl', 'rb'))
input_model = [[10,1.1,2,0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0]]
pred=model.predict(input_model)
churn = "False"
if pred[0] == 1:
churn = "He Will Churn"
elif pred[0] == 0:
churn = "He Will Not Churn"
return churn
def example4():
model = pickle.load(open('model.pkl', 'rb'))
input_model = [[7,0.8,5,0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0,0, 0, 1]]
pred=model.predict(input_model)
churn = "False"
if pred[0] == 0:
churn = "She Will Churn"
elif pred[0] == 1:
churn = "She Will Not Churn"
return churn
def greet(Total_Transaction, Total_Ct_Chng_Q4_Q1, Total_Relationship_Count, Education=None, Annual_Income=None, Marital_Status=None, Card_Type=None):
educ, edud, edug, eduh, edup, eduu, ai0, ai40, ai60, ai80, ai120, msd, msm, mss, ctb, ctg, cts = 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
if Annual_Income == "0k-40k":
ai0 = 1
elif Annual_Income == "40k-60k":
ai40 = 1
elif Annual_Income == "60k-80k":
ai60 = 1
elif Annual_Income == "80k-120k":
ai80 = 1
elif Annual_Income == "120k+":
ai120 = 1
if Marital_Status == "Single":
mss = 1
elif Marital_Status == "Married":
msm = 1
elif Marital_Status == "Divorced":
msd = 1
if Card_Type == "Blue":
ctb = 1
elif Card_Type == "Gold":
ctg = 1
elif Card_Type == "Silver":
cts = 1
if Education == "College":
educ = 1
elif Education == "Doctorate":
edud = 1
elif Education == "Graduate":
edug = 1
elif Education == "High-School":
eduh = 1
elif Education == "Post-Graduate":
edup = 1
elif Education == "Uneducated":
eduu = 1
input_model = [[Total_Transaction,Total_Ct_Chng_Q4_Q1,Total_Relationship_Count,educ, edud, edug, eduh, edup, eduu, ai120, ai40, ai60, ai80, ai0, msd, msm, mss,ctb, ctg, cts]]
model = pickle.load(open('model.pkl', 'rb'))
pred=model.predict(input_model)
churn = "False"
if pred[0] == 1:
churn = "True"
elif pred[0] == 0:
churn = "False"
return churn
with gr.Blocks() as demo:
with gr.Row():
with gr.Column(scale=1,min_width=600):
gr.Image("logo2.png").style(height='7')
Total_Transaction = gr.Slider(0, 200,label="Total Transaction Count")
Total_Ct_Chng_Q4_Q1 = gr.Slider(0, 30,label="Transaction Count Q4 vs Q1")
Total_Relationship_Count = gr.Slider(0, 20,step=1,label="Total Relationship Count")
with gr.Column(scale=2,min_width=600):
with gr.Row():
with gr.Column(scale=1,min_width=300):
Annual_Income = gr.Dropdown(["0k-40k","40k-60k","60k-80k","80k-120K","120k+"],label="Annual Income")
with gr.Column(scale=2,min_width=300):
Education = gr.Dropdown(["College","Doctorate","Graduate","High-School","Post-Graduate","Uneducated","Unknown"],label="Education")
with gr.Row():
with gr.Column(scale=3,min_width=300):
Marital_Status = gr.Dropdown(["Single","Married","Divorced","Unknown"],label="Marital Status")
with gr.Column(scale=4,min_width=300):
Card_Type = gr.Dropdown(["Blue","Silver","Gold"],label="Crad Type")
churn = gr.Textbox(value="", label="Churn")
btn = gr.Button("PREDICT").style()
btn.click(fn=greet, inputs=[Total_Transaction,Total_Ct_Chng_Q4_Q1,Total_Relationship_Count,Education,Annual_Income,Marital_Status,Card_Type], outputs=[churn])
gr.Markdown("""# Few Examples Based on Real-World Simulations""")
with gr.Row():
with gr.Column(scale=1,min_width=300):
gr.Image("avatars/1.png")
churn1 = gr.Textbox(value="", label="Churn")
btn1 = gr.Button("PREDICT").style()
exp =1
btn1.click(fn=example1, inputs=[], outputs=[churn1])
gr.Markdown("""
# Corporate Professional!
Total Transaction Count - 45\n
Transaction Count Q4 vs Q1 - 1.3\n
Total Relationship Count - 2\n
Annual Income - 40k-60k\n
Education - Graduate\n
Marital Status - Married\n
Card Type - Silver\n
""")
with gr.Column(scale=2,min_width=300):
gr.Image("avatars/4.png")
churn2 = gr.Textbox(value="", label="Churn")
bt2 = gr.Button("PREDICT").style()
bt2.click(fn=example4, inputs=[], outputs=[churn2])
gr.Markdown("""
# Medical Professional!
Total Transaction Count - 7\n
Transaction Count Q4 vs Q1 - 0.8\n
Total Relationship Count - 5\n
Annual Income - 80k-120k\n
Education - Doctorate\n
Marital Status - Married\n
Card Type - Gold\n
""")
with gr.Column(scale=3,min_width=300):
gr.Image("avatars/2.png")
churn3 = gr.Textbox(value="", label="Churn")
btn3 = gr.Button("PREDICT").style()
btn3.click(fn=example2, inputs=[], outputs=[churn3])
gr.Markdown("""
# Freelance Photographer!
Total Transaction Count - 41\n
Transaction Count Q4 vs Q1 - 2\n
Total Relationship Count - 2\n
Annual Income - 0k-40k\n
Education - High-School\n
Marital Status - Single\n
Card Type - Blue\n
""")
with gr.Column(scale=4,min_width=300):
gr.Image("avatars/3.png")
churn4 = gr.Textbox(value="", label="Churn")
btn4 = gr.Button("PREDICT").style()
btn4.click(fn=example3, inputs=[], outputs=[churn4])
gr.Markdown("""
# Retired Veteran Pensioner!
Total Transaction Count - 10\n
Transaction Count Q4 vs Q1 - 1.1\n
Total Relationship Count - 2\n
Annual Income - 80k-120k\n
Education - Post-Graduate\n
Marital Status - Divorced\n
Card Type - GOld\n
""")
demo.launch()