Deployed app takes forever to load / is unstable

Community Cloud
1

Hi there,

I am currently trying to deploy a Streamlit app based on an xgboost prediction (link to deployed app / link to GitHub repo).

The problem is that the app does not properly deploy as it’s stuck on a blank screen while displaying the ‘RUNNING…’ info on the upper right. Sometimes the app starts to shows its real content after reloading the tab, but then, it freezes after giving user input via the sidebar slider.

Other topics suggested that deleting Streamlit from the requirements.txt should greatly enhance performance, however this couldn’t solve the case for this app. Also, I do wonder if fitting model on the data via cloud (as in my app) could be a reason for performance issues, despite that xgboost is usually not too computationally expensive?

Best Regards and thanks!

2

Hi @kevin-kohler! Would you be able to post your code so we can take a look at it? Could it be the model is slow?

3

Hi @TylerS, thank you for reaching out! Code is the following:

# Import required libraries
import numpy as np 
import pandas as pd
import streamlit as st
from xgboost import XGBClassifier, plot_importance
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# Retrieve training and test data
# Data set: https://www.kaggle.com/datasets/uciml/pima-indians-diabetes-database
df = pd.read_csv('data/diabetes.csv')

cols_missing_vals = ['Glucose','BloodPressure','SkinThickness','Insulin','BMI','Age']
df[cols_missing_vals] = df[cols_missing_vals].replace(to_replace=0, value=np.NaN)
df[cols_missing_vals] = df[cols_missing_vals].fillna(value=df.mean())

X = df.iloc[:,0:8]
y = df.iloc[:,8]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.20, random_state = 1992)

model = XGBClassifier(n_estimators=200, max_depth=25, learning_rate=0.1, subsample=0.5)
model.fit(X_train, y_train)

st.set_page_config(page_title='dora', page_icon='img/favicon.png', layout="wide", initial_sidebar_state="expanded", menu_items=None)

st.title('Diabetes Occurrence Risk Assessment')
st.subheader('Estimate your risk of developing Type II Diabetes based on your biomarker values, using a trusted machine learning algorithm.')

st.sidebar.image('img/logo.png', use_column_width='auto', output_format='PNG')

st.sidebar.subheader('Insert your biomarker values here:')

def user_input():
    num_pregn = st.sidebar.slider('Number of Pregnancies',0,20,4)
    glucose = st.sidebar.slider('Glucose Level (mg/dL @ 2-Hour GTT)',0,240,121)
    blood_press = st.sidebar.slider('Diastolic Blood Pressure (mm Hg)',0,150,69)
    skin_thickn = st.sidebar.slider('Triceps Skin Fold Thickness (mm)',0,100,21)
    insulin = st.sidebar.slider('2-Hour Serum Insulin (mu U/ml)',0,850,80)
    bmi = st.sidebar.slider('Body Mass Index (kg/m^2)',0,70,32)
    diab_pedigr = st.sidebar.slider('Diabetes Pedrigree Value',0.0,2.5,0.5)
    age = st.sidebar.slider('Age (years)',0,85,33)

    user_input = {
        'Pregnancies': num_pregn,
        'Glucose': glucose,
        'BloodPressure': blood_press,
        'SkinThickness': skin_thickn,
        'Insulin': insulin,
        'BMI': bmi,
        'DiabetesPedigreeFunction': diab_pedigr,
        'Age': age
    }

    user_data = pd.DataFrame(user_input, index=[0])
    return user_data

tab1, tab2, tab3 = st.tabs(['Overall Probability', 'Influencing Factors', 'Forecast Accuracy'])

with tab1:
    st.subheader('What is your risk of developing Type II Diabetes?')
    user_outcome = model.predict_proba(user_input())
    st.metric(label='Probability', value=(st.write('{:.1%}'.format(user_outcome[0,1]))), delta_color='inverse',
    help='This is your calculated risk of developing Type II Diabetes.')
    conclusion = ''
    if user_outcome[0,1] <= 0.3:
        conclusion = 'You **ARE NOT** at risk of developing Type II Diabetes.'
    else:
        conclusion = 'You **ARE** at risk of developing Type II Diabetes. Seek medical consultation.'
    st.write(conclusion)

with tab2:
    st.subheader('Which factors contribute most to your risk of Type II Diabetes?')

    # Plot most important features in desc order by 
    plot_feats = plot_importance(model, importance_type='gain', title=None, show_values=False, xlabel='Knowledge Gain', ylabel='Influencing Factor', color='#7E22CE', grid=False)
    st.pyplot(plot_feats.figure)

with tab3:
    st.subheader('How accurate is the prediction of your Type II Diabetes risk?')
    st.bar_chart(df)

    if False:
        # Return main df info
        df.info()

        # Return head of df
        df.head()

        # Count zero values per columns
        for column_name in df.columns:
            column = df[column_name]
            count_zeros = (column == 0).sum()
            print('Zero values per column: ', column_name, count_zeros)

        # Plot boxplot showing df value distribution
        df.plot(kind='box', figsize=(20,10))

        y_pred = (model.predict_proba(X_test)[:,1] >= 0.3).astype(bool) # Set threshold as 0.3

        test_accuracy = accuracy_score(y_test, y_pred)
        print('Accuracy score: ', test_accuracy)

        print (f'Shape of Train Data : {X_train.shape}')
        print (f'Shape of Test Data : {X_test.shape}')

st.caption('dora (Diabetes Occurrence Risk Assessment) is a Streamlit app which uses XGBoost on the Pima Indians Dataset in order to predict user\'s risks for Diabetes Type II. This is a test app that cannot replace professional medical advice. In case of health issues, seek help from experts. No liability is assumed for the outcomes of this app.')

On my local machine, the model does not take long to fit and predict. I guess that another option would be to train the model locally and then export its coefficients to the app (as done here)? But I first have to check how this works.

4

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.