My streamlit app is so slow

🎈 Using Streamlit
1

I tried running my streamlit app both locally and when deployed but it is so slow and it can’t process its supposed function. I’d like to ask for help to know what’s wrong :smiling_face_with_tear:

Here’s my app:

import streamlit as st
import pandas as pd
import numpy as np

st.title("Merge Master Sample (Products to Merge)")

# Upload CSV
csv_file = st.file_uploader("Upload csv file here:")

if csv_file is not None:
    csv_file = pd.read_csv(csv_file)
    st.subheader("Preview of Uploaded File")
    st.dataframe(csv_file)
    
    # Applicability Symbol Input
    applicability_symbol = st.text_input("Input used applicability symbol:")
    
    # Step 1 -- Total of Applicable Parts per Product -- Additional Column
    def count_occurrences(row, value):
        return (row==value).sum()
    
    # Apply the custom function to each row of the DataFrame
    csv_file['Total Applicable C'] = csv_file.apply(lambda row: count_occurrences(row, applicability_symbol), axis=1)
    
    # Step 2 -- Delete all rows with the maximum value in the Total Applicable column
    drop_row_condition = csv_file['Total Applicable C'] == csv_file['Total Applicable C'].max()
    csv_file = csv_file[~drop_row_condition]
    
    # Step 3 -- Total of Applicable Products per Part -- Additional Row

    # Transpose dataframe
    csv_file = csv_file.transpose()
    
    # Add another column -- Total Applicable R
    def count_occurrences(row, value):
        return (row==value).sum()
    
    # Apply the custom function to each row of the DataFrame
    csv_file['Total Applicable R'] = csv_file.apply(lambda row: count_occurrences(row, applicability_symbol), axis=1)
    
    # Transpose dataframe
    csv_file = csv_file.transpose()
    
    # Step 4 - Print the column name of the row with highest value of Total Applicable R
    mother_product = csv_file.iloc[-1].idxmax()
    
    # Step 5 -- Drop unnecessary columns
    # Multiselect box to choose columns to drop
    columns_to_drop = st.multiselect("Select columns to drop:", csv_file.columns)
    
    branch_column = st.selectbox("Select Branch Column:", csv_file.columns)

    # Drop selected columns
    csv_file = csv_file.drop(columns=columns_to_drop)
    
    # Step 6 -- Final, Final
    # Initialize a list to store the deleted values in the 'Part' column and the chosen column name for each iteration
    deleted_values_per_iteration = []

    # Continue the process until the DataFrame is empty
    while not csv_file.empty:
        # Identify the column with the highest count of non-NaN values
        column_with_max_count = csv_file.iloc[:, 1:].count().idxmax()
        
        # Remove rows with non-NaN values in the identified column
        deleted_values = csv_file.loc[~csv_file[column_with_max_count].isna(), branch_column].tolist()
        
        # Append the chosen column name to the list
        deleted_values_per_iteration.append({'Column': column_with_max_count, 'DeletedValues': deleted_values})
        
        # Update the DataFrame by removing rows with non-NaN values in the identified column
        csv_file = csv_file[csv_file[column_with_max_count].isna()]

    # Display the deleted values and the chosen column for each iteration
    st.title("Products Needed to Form the Merge Master Sample")
    for i, values_dict in enumerate(deleted_values_per_iteration, start=1):
        column_name = values_dict['Column']
        deleted_values = values_dict['DeletedValues']
        
        # Highlight the first iteration as the "Mother Product"
        if i == 1:
            st.subheader(f'{i}: Mother Product: {column_name} \n Branches: {deleted_values}')
        else:
            st.subheader(f'{i}: Additional Product: {column_name} \n Branches: {deleted_values}')
2

Hi @APQPxSystems,

Thanks for sharing this question!

There are several reasons as to why your app might be slow. The main one is if you’re uploading a very large file, reading and processing can be resource-intensive and slow especially if it is not cached.

Using apply() across rows and columns in pandas can be slow especially if the df is large.

You should also consider using st.cache_data() to cache the results of the expensive computations that don’t need to be run on every interaction.