app.py

# FIX v1

# Module Imports
import pandas as pd
import numpy as np
import streamlit as st
from pycaret import regression as reg
from pycaret import classification as clf
from sklearn.metrics import mean_absolute_error, max_error, r2_score, mean_squared_error, confusion_matrix, ConfusionMatrixDisplay, accuracy_score
import matplotlib.pyplot as plt
import streamlit.components.v1 as components
import mpld3
import time
# ----------------------------------------------------------------------------------------------------------------------  # 
# Settings:
st.set_option('deprecation.showPyplotGlobalUse', False)

# ----------------------------------------------------------------------------------------------------------------------  # 
# Collecting User Input 
## Preamble & Formatting

st.markdown(f'<h1 style="color:#0096FF;font-size:54px;">{"No Code ML"}</h1>', unsafe_allow_html=True)
col1, mid, col2 = st.columns([10,1,20])
with col1:
    st.image('https://images.pexels.com/photos/2599244/pexels-photo-2599244.jpeg?auto=compress&cs=tinysrgb&w=1260&h=750&dpr=1')
with col2:
    st.markdown("""This tool prepares a machine learning model using your tabular data. The tool can be used in 2 ways:""", unsafe_allow_html=True)
    st.markdown("""1) Benchmark different algorithms for your dataset to find the best algorithm and then tune that model to determine best hyperparameters.""", unsafe_allow_html=True)
st.markdown("""2) In the case of experimental science, the best obtained model can be used to make predictions for various combinations of the provided data to try to obtain a combination that achieves a desired target value (if possible).""", unsafe_allow_html=True)
st.markdown("""**The tool is currently under active development. Please direct any bug reports or inquiries to the <a href="http://cleanenergy.utoronto.ca/">clean energy lab at UofT.</a>**""", unsafe_allow_html=True)
st.markdown("""---""")

st.markdown(f"**To use this tool**, fill out all the requested fields from top to bottom.")
st.markdown(f"**Note:** If an error is obtained refresh the page and start over.")

## Column Name
st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"1) Provide name of the column you want to predict with model."}</h3>', unsafe_allow_html=True)
target_col = st.text_input("Enter the exact name of the column with your target variable. This field is case sensitive. (i.e., capital letters must match.)")

## Task Type: Regression or Classification
st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"2) Select type of model you would like to build"}</h3>', unsafe_allow_html=True)
mod_type = st.selectbox("What type of model would you like to train? Pick regression model for continous values or classifier for categorical values.", ('regression', 'classifier'))

## Mode of Use
st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"3) Select mode of use"}</h3>', unsafe_allow_html=True)
mode_type = st.selectbox("What would you like to use the tool for?", ('Benchmarking (finding the best algorithm for your problem)', 'Parameter Search (find combination of parameters to get a desired value)'))
if mode_type == 'Parameter Search (find combination of parameters to get a desired value)': 
     ## Desired Target Value 
    if mod_type == 'classifier':
        st.write('Parameter search is not currently supported with classifier type models.')
        st.write('Please refresh page and try again with the supported tasks.')
        exit()
   
    if mod_type == 'regression':
        st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"4) Type of parameter search"}</h3>', unsafe_allow_html=True)
        opt_type = st.selectbox("What do you want to do with the output?", ('Maximize it', 'Minimize it', 'Obtain a desired value'))
    if opt_type == 'Obtain a desired value':
        desired_value = float(st.number_input("Enter the desired value for the target variable."))

    ## Ask for Dataset
    st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"5) Upload CSV file "}</h3>', unsafe_allow_html=True)
    uploaded_file = st.file_uploader("Upload a CSV file", type="csv")

else:
    ## Ask for Dataset
    st.markdown(f'<h3 style="color:#000000;font-size:20px;">{"4) Upload CSV file "}</h3>', unsafe_allow_html=True)
    uploaded_file = st.file_uploader("Upload a CSV file", type="csv")

# ----------------------------------------------------------------------------------------------------------------------  # 
if uploaded_file:
    # Read CSV File and Provide Preview of Data and Statistical Summary:
    data = pd.read_csv(uploaded_file)
    data_size = len(data)

    if target_col not in list(data.columns):
        st.error("ERROR: Provided name of the target column is not in the CSV file. Please make sure you provide the exact match (case sensitive).Please provide the correct label and try again.")
        exit()

    st.subheader("Data preview:")
    st.write(data.head())
    st.subheader("Statistical Summary of the Provided Data:")
    st.write(data.describe())

    # Prepare Train/Test Split:
    fraction_check = 10_000/data_size      # Cap Training Dataset to 10k to allow tuning to occur in a timely manner
    if fraction_check < 0.8:
        train_frac = fraction_check
        train_data = data.sample(frac=train_frac, random_state=0)
        test_data = data.drop(train_data.index)
        if len(test_data) > 5_000:
            test_data = test_data[0:5000]
    else:
        train_frac = 0.8
        train_data = data.sample(frac=train_frac, random_state=0)
        test_data = data.drop(train_data.index)

    # Figure out Column Data Types 
    object_columns = data.select_dtypes(include="object").columns.tolist()

    # Create a list of Tree Models:
    tree_mods_list = ['Extra Trees Regressor', 'Extra Trees Classifier', 'Random Forest Regressor', 'Random Forest Classifier', 'Decision Tree Regressor', 'Decision Tree Classifier', 'CatBoost Regressor', 'Light Gradient Boosting Machine']

# ----------------------------------------------------------------------------------------------------------------------  # 
    # Build Regression Model
    if mod_type == "regression":
        # Setup Regressor Problem
        if object_columns:
            if data_size > 20:
                s = reg.setup(train_data, target = target_col, normalize=True, categorical_features=object_columns, fold=5, silent= True)
            else:
                s = reg.setup(data, target = target_col, normalize=True, categorical_features=object_columns, silent= True)        
        else:        
            if data_size > 20:
                s = reg.setup(train_data, target = target_col, normalize=True, silent= True, fold=5)
            else:
                s = reg.setup(data, target = target_col, normalize=True, silent= True)        

        # Find the best algorithm to build Model:
        st.subheader("Algorithm Selection")
        start_algo = time.time()
        with st.spinner(text="Finding the best algorithm for your dataset..."):
            best_mod = reg.compare_models()
            regression_results = reg.pull()
            best_mod_name = regression_results.Model[0]
            st.write(regression_results)
        end_algo = time.time()
        st.write('Time taken to select algorithm:', end_algo - start_algo, 'seconds')

        # Tune the hyperparameters for the best algorithm: 
        st.subheader("Tuning the Model")
        start_tune = time.time()
        with st.spinner(text="Tuning the algorithm..."):
            tuned_mod = reg.tune_model(best_mod, optimize = 'RMSE', n_iter=5)
        end_tune = time.time()
        st.write('Time taken to select hyperparameters:', end_tune - start_tune, 'seconds')

        # Finalize the model (Train on the entire train dataset):
        with st.spinner("Finalizing the model..."):
            final_mod = reg.finalize_model(tuned_mod)

        st.success('Model successfully trained! Here are your results:')
        st.write('Best algorithm: ', best_mod_name)
        st.write('Best hyperparameters: ', final_mod.get_params())

        # Print a SHAP Analysis Summary Plot:
        if best_mod_name in tree_mods_list:
            st.subheader("SHAP Analysis Summary Plot")        
            st.pyplot(reg.interpret_model(final_mod))

        if len(data) > 20:
            # Predict on the test set if it was created:
            st.subheader("Evaluating model on the test/hold out data:")
            predictions = reg.predict_model(final_mod, data=test_data)
            st.success('Here are your results:')
            st.write(predictions)
            st.caption('"Label" is the value predicted by the model.')

            # Accuracy of predictions:
            MAE_val = mean_absolute_error(predictions[target_col], predictions['Label'])
            RMSE_err = mean_squared_error(predictions[target_col], predictions['Label'], squared=False)
            Max_err = max_error(predictions[target_col], predictions['Label'])
            r2_val = r2_score(predictions[target_col], predictions['Label'])
            err_dict = {'Mean Absolute Error': MAE_val, 'Root Mean Squared Error': RMSE_err, 'Maximum Error': Max_err}
            df_err = pd.DataFrame(err_dict, index=[0])
            st.write(df_err)

            # Create an true vs. predicted plot 
            fig = plt.figure(figsize=(8,8))
            plt.grid(b=None)
            plt.scatter(x=predictions[target_col], y=predictions['Label'])
            plt.xlabel("True Value", fontsize=18)
            plt.ylabel("Predicted Value", fontsize=18)
            fig_html = mpld3.fig_to_html(fig)
            components.html(fig_html, height=1000)

# ----------------------------------------------------------------------------------------------------------------------  #         
        # Use Trained Model to Explore Parameter Space
        if mode_type == 'Parameter Search (find combination of parameters to get a desired value)':
            if object_columns:
                st.write("Optimization with string data types not currently supported.")
            else:
                with st.spinner("Generating parameter combinations for search"):
                    # Creating Variables for Data Generation Used in the Optimization Segment
                    list_of_cols = list(data.columns[0:-1])
                    
                    # Find min and max value for the input features in the training dataset
                    max_list = [data[i].max() for i in list_of_cols]
                    min_list = [data[i].min() for i in list_of_cols]

                    # Generate DF from New Parameters
                    generated_data = np.array([np.random.randint(low=min_list[i], high=max_list[i], size=50_000) for i in range(0,len(max_list))]).T
                    generated_data_df = pd.DataFrame(generated_data, columns = list_of_cols)

                    # Make Predictions with Trained Model
                    generated_predictions = reg.predict_model(final_mod, data = generated_data_df)

                if opt_type == 'Obtain a desired value':
                    st.subheader("Using the trained model to obtain the desired target value:")

                    # Filter results to get the places closed to desired value
                    ## Determine +/- window for search 
                    data_spread = data[target_col].std()/3   
                    dv_min = desired_value - data_spread
                    dv_max = desired_value + data_spread

                    ## Apply +/- window to determine lower and upper bound to filter DF ('Generated_predictions)
                    lower_bound = generated_predictions["Label"] >=dv_min
                    upper_bound = generated_predictions["Label"] <=dv_max

                    ## Filter DF using upper and lower bounds - sort values based on absolute distance to desired value provided by user.
                    proposed_values_to_try = generated_predictions[lower_bound & upper_bound]
                    proposed_values_to_try['distance_to_desired_value'] = np.abs(generated_predictions['Label'] - desired_value)
                    proposed_values_to_try.sort_values('distance_to_desired_value', inplace=True)
                    proposed_values_to_try.reset_index(drop=True, inplace=True)
                    
                    ## Display top 10 rows
                    final_proposed_parameters = proposed_values_to_try[0:10]

                if opt_type == 'Maximize it':
                    st.subheader("Using the trained model to maximize target value:")
                    generated_preds = generated_predictions.copy()
                    
                    # Sort results in descending order based on predicted values
                    generated_preds.sort_values('Label', ascending=False, inplace=True)
                    generated_preds.reset_index(drop=True, inplace=True)
                    
                    ## Display top 10 rows
                    final_proposed_parameters = generated_preds[0:10]

                if opt_type == 'Minimize it':
                    st.subheader("Using the trained model to minimize target value:")
                    generated_preds = generated_predictions.copy()
                    
                    # Sort results in descending order based on predicted values
                    generated_preds.sort_values('Label', inplace=True)
                    generated_preds.reset_index(drop=True, inplace=True)
                    
                    ## Display top 10 rows
                    final_proposed_parameters = generated_preds[0:10]

                if len(final_proposed_parameters) == 0:
                    st.write("No parameters could be found for the desired value based on current model. Try collecting additional data or provide a different target value.")
                else:
                    st.write(final_proposed_parameters)
                    st.download_button(label="Download the Proposed Parameters to Try", data = final_proposed_parameters.to_csv(index=False), file_name='Final_proposed_parameters.csv')


# ----------------------------------------------------------------------------------------------------------------------  # 
    # Build Classifier Model
    if mod_type == "classifier":
        # Setup Classifier Problem
        if data_size > 20:
            s = clf.setup(train_data, target = target_col, normalize=True, silent= True, fold=5)
        else:
            s = clf.setup(data, target = target_col, normalize=True, silent= True)        

        # Find the best algorithm to build Model:
        st.subheader("Algorithm Selection")
        start_algo = time.time()
        with st.spinner(text="Finding the best algorithm for your dataset..."):
            best_mod = clf.compare_models()
            classifier_results = clf.pull()
            best_mod_name = classifier_results.Model[0]
            st.write(classifier_results)
        end_algo = time.time()
        st.write('Time taken to select algorithm:', end_algo - start_algo, 'seconds')

        # Tune the hyperparameters for the best algorithm: 
        st.subheader("Tuning the Model")
        start_tune = time.time()
        with st.spinner(text="Tuning the algorithm..."):
            tuned_mod = clf.tune_model(best_mod, optimize = 'AUC', n_iter=5)
        end_tune = time.time()
        st.write('Time taken to select hyperparameters:', end_tune - start_tune, 'seconds')

        # Finalize the model (Train on the entire train dataset):
        with st.spinner("Finalizing the model..."):
            final_mod = clf.finalize_model(tuned_mod)

        st.success('Model successfully trained! Here are your results:')
        st.write('Best algorithm: ', best_mod_name)
        st.write('Best hyperparameters: ', final_mod.get_params())

        # Print a Feature Importance Plot:
        if best_mod_name in tree_mods_list:
            st.subheader("Feature Importance Plot")        
            st.pyplot(clf.plot_model(final_mod, plot='feature'))

        if len(data) > 20:
            # Predict on the test set if it was created:
            st.subheader("Evaluating model on the test/hold out data:")
            predictions = clf.predict_model(final_mod, data=test_data)
            st.success('Here are your results:')
            st.write(predictions)
            st.caption('"Label" is the value predicted by the model.')
            st.write('---')

            # Provide Accuracy:
            mod_accuracy = accuracy_score(predictions[target_col], predictions['Label'])
            st.write('**Model accuracy on test set :**', f'{(mod_accuracy):.2f}')

            # Create a confusion matrix:
            st.subheader("Confusion Matrix for test set:")
            cm = confusion_matrix(predictions[target_col], predictions['Label'])
            disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=predictions[target_col].unique())
            disp.plot()
            plt.grid(b=None)
            st.pyplot()

# Visitor Badge
st.markdown("![visitor badge](https://visitor-badge.glitch.me/badge?page_id=singhk28_nocodeml)")