pages/4_Fracture_Prediction.py

import numpy as np
import lightgbm as lgb

import pickle, os
import bz2file as bz2

import altair as alt

import streamlit as st

from ui import *
from mLogsFunctions import *
from mLogsFunctions.fx import *


#ignore version warnings
import warnings
warnings.filterwarnings("ignore")

#Global variables
THRESHOLD_GLOBAL = 0.5

#All wells to be trained
wells_name = ["01-97-HXS-1X", "15-1-SN-1X", "15-1-SN-2X", "15-1-SN-3XST", "15-1-SN-4X", "15-1-SNN-1P",
              "15-1-SNN-2P", "15-1-SNN-3P", "15-1-SNN-4P", "15-1-SNS-1P", "15-1-SNS-2P", "15-1-SNS-4P"]
#Obtain data and label of wells
name_features = ["GR", "LLD", "LLS", "NPHI", "RHOB", "DTC", "DTS"]
label = "FRACTURE_ZONE"
#-----------------------------------------------------------------
# processing pipeline
def remove_negative_val(df, col):
    return df.drop(index=df[df[col] < 0].index)
def rel_depth(df):
    dfx = []
    for well in df.WELL.unique():
        df_ = df[df.WELL==well].sort_values(by="DEPTH", ascending=True)
        dfx.append(df_.assign(rel_depth=df_.DEPTH / df_.DEPTH.values[0]))
    return pd.concat(dfx).reset_index(drop=True)

def tweak_data_S(df):
    return (
        df.assign(
                # FRACTURE_ZONE=df.FRACTURE_ZONE.replace({-9999: 0, np.nan: 0}).astype('int8'),
                GR=df.GR.replace({-9999.:0.}).astype('float32'),
                DCALI_FINAL=df.DCALI_FINAL.replace({-9999.:0.}).astype('float32'),
                LLD=df.LLD.replace({-9999.:0.}).astype('float32'),
                LLS=df.LLS.replace({-9999.:0.}).astype('float32'),
                NPHI=df.NPHI.replace({-9999.:0.}).astype('float32'),
                RHOB=df.RHOB.replace({-9999.:0.}).astype('float32'),
                DTC=df.DTC.replace({-9999.:0.}).astype('float32'),
                DTS=df.DTS.replace({-9999.:0.}).astype('float32'),
                DEPTH=df.DEPTH.astype('float32')
                )
                .pipe(remove_negative_val, "RHOB")
                .pipe(remove_negative_val, "DTC")
                .pipe(remove_negative_val, "DTS")
                .pipe(remove_negative_val, "GR")
                .pipe(remove_negative_val, "LLD")
            ).pipe(rel_depth)
# Calculate the confusion matrix of applying model on dataframe (including features and label) df with threshold
def calculate_confusion_matrix (model = None, df= None, threshold=None):
    model_prediction = [model]
    # Apply model on dataframe
    proba = Prediction_LGBM(trained_models=model_prediction, data = df, feature_names=name_features)
    proba_well = proba.loc[:, "model_0"]
    # Apply threshold
    if threshold==None: threshold = 0.5
    # Get label from dataframe df
    well_proba = proba_well.apply(lambda x: 1 if x >= threshold else 0)
    return well_proba

#------------------------------------------------------------------
# Load any compressed pickle file
file = "/work/2022_VPIMLogs_WebApp/models/LightGBM_0.45.pbz2"
def decompress_pickle(file):
    data = bz2.BZ2File(file, 'rb')
    data = pickle.load(data)
    return data
# model_best = decompress_pickle(file)

# Loading Modeling
# model_best = lgb.Booster(model_file="/work/2022_VPIMLogs_WebApp/models/LGBM_20221125.json")

#Loading data from browser:----------------------------------------
#Streamlit Dashboard------------------------------------------------------------------------------------------
pagetile = """<center><h1>PREDICTION SITE</h1></center>"""
st.markdown(pagetile, unsafe_allow_html=True)
# set_page_config(page='custom')

hide_menu_button()
condense_layout()

logo_site, info_site = st.columns([1.5, 8.5])
with logo_site:
    st.image("https://i.ibb.co/Yd42K98/LogoVPI.png", use_column_width='auto')
with info_site:
    # Option 1: CSV File Loading
    st.write('You can load your csv file using the file upload or selection from LAS Exploration option below.')
    st.subheader("1. CSV File Loading")
    st.caption('## 1.1. CSV from Uploader')
    df = csv_uploader()

    # Option 2: CSV from LAS Exploration
    st.caption('## 1.2. CSV from LAS Exploration')
    dir_path = 'data/merged/'
    csv_files = [filename for filename in os.listdir(dir_path) if filename.endswith('.csv')]
    selected_csv_file= st.multiselect('Select a CSV file', csv_files, key = 'st.session_state.selected_well_multi')
    if selected_csv_file: # Nếu người dùng đã chọn file CSV
        # Đọc file csv được chọn vào DataFrame
        file_path = 'data/merged/'
        wells_df_predict = pd.concat([pd.read_csv(file_path + f) for f in selected_csv_file])
        # wells_df_predict = tweak_data_S(wells_df_predict)
    else: # Nếu người dùng không chọn file CSV
        wells_df_predict = df
        # wells_df_predict = tweak_data_S(wells_df_predict)
    
    st.write('You can load your json file using the file upload or selection from TRAINING SECTION below.')

    st.subheader("2. JSON File Loading")
    st.caption('## 2.1. JSON from Uploader')
    model_best_uploader = None
    uploaded_file = st.file_uploader("Choose a JSON file", type="json")
    # Kiểm tra xem có file được upload hay không
    if uploaded_file is not None:
        # Lưu file JSON tạm thời
        with open("temp.json", "w") as f:
            f.write(uploaded_file.read().decode("utf-8"))
        # Đường dẫn tới file JSON tạm thời
        temp_file_path = os.path.abspath("temp.json")
        # Tạo Booster từ file JSON
        model_best_uploader = lgb.Booster(model_file=temp_file_path)

        # Xóa file tạm sau khi sử dụng
        os.remove(temp_file_path)
    # Option 2: JSON from TRAINING SECTION
    st.caption('## 2.2. JSON from TRANING SECTION')
    dir_path = 'models/'
    json_files = [filename for filename in os.listdir(dir_path) if filename.endswith('.json')]
    selected_json_file= st.multiselect('Select a JSON file', json_files, key = 'st.session_state.selected_well_multi_JSON')
    if selected_json_file: # Nếu người dùng đã chọn file json
        # Đọc file json được chọn vào Booster
        file_path = 'models/'
        model_files = "/models/05_13_2023_11_50_38_model_LGBM.json"
        model_best = lgb.Booster(model_file=model_files)
    else: # Nếu người dùng không chọn file json
        model_best = model_best_uploader

if wells_df_predict is not None:
    wells_df_predict = tweak_data_S(wells_df_predict)
    wells_df_predict = wells_df_predict.replace({-9999: np.nan}).dropna(how='any', subset = "FRACTURE_ZONE")
    st.write("Data Input:")
    st.dataframe(wells_df_predict.sort_index(), width=1400, height=300)
    st.write('---')
    st.write("Selected Prediction Model:")
    st.write(model_best)
#------------------------------------------------------------------
    feature_names = [col for col in wells_df_predict.columns if col not in ["WELL", "DEPTH","FRACTURE_ZONE"]]    
    # Full data for export data
    st.session_state.pred = st.button("Predict Fracture Zone")
    if st.session_state.pred:
        threshold = 0.5
        #Make label Prediction 
        predictions = (model_best.predict(wells_df_predict[feature_names])> threshold).astype(int)
        wells_df_predict['FRACTURE_ZONE_PRED'] = predictions
        st.dataframe(wells_df_predict, width=1400, height=300)
#Plot Data------------------------------------------------------------------
        plotting_curves = [c for c in wells_df_predict.columns.unique() if c not in ["DEPTH", "WELL", "TVD", "FRACTURE_ZONE", "FRACTURE_ZONE_PRED", "DCALI_FINAL", "INCL", "AZIM_TN", "rel_depth"]]
        plotting_curves.sort()
        if "FRACTURE_ZONE_PRED" in wells_df_predict.columns.unique():
            plotting_curves.append("FRACTURE_ZONE_PRED")
        for well in wells_df_predict.WELL.unique():
            st.write('---')
            st.write(f"{well} Logs: \n")
            well_plot = wells_df_predict[wells_df_predict.WELL == well]
            charts_dict={}
            for i, c in enumerate(plotting_curves):
                charts_dict[i] = curve_plot(data=well_plot,filted_data=None, x_column=c)
    #Show Curve-----------------------------------------------------------------------
            st.write(alt.concat(*charts_dict.values(), columns = 12).configure(autosize='fit'))
    # Download --------------------------------------------------------------
        st.write('---')
        st.write("Download final result to csv file")

        st.download_button(label='Download All Wells',
                        data = wells_df_predict.to_csv(),
                        file_name='FracturePredictionALL.csv',
                        mime='text/csv')
        
hide_menu_button()
condense_layout()