app.py

import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt
import pytorch_lightning as pl
from neuralforecast.core import NeuralForecast
from neuralforecast.models import NHITS, TimesNet, LSTM, TFT
from neuralforecast.losses.pytorch import HuberMQLoss
from neuralforecast.utils import AirPassengersDF
import time
from st_aggrid import AgGrid
from nixtla import NixtlaClient
import os

st.set_page_config(layout='wide')
    
@st.cache_resource
def load_model(path, freq):
    nf = NeuralForecast.load(path=path)
    return nf

@st.cache_resource
def load_all_models():
    nhits_paths = {
        'D': './M4/NHITS/daily',
        'M': './M4/NHITS/monthly',
        'H': './M4/NHITS/hourly',
        'W': './M4/NHITS/weekly',
        'Y': './M4/NHITS/yearly'
    }
    
    timesnet_paths = {
        'D': './M4/TimesNet/daily',
        'M': './M4/TimesNet/monthly',
        'H': './M4/TimesNet/hourly',
        'W': './M4/TimesNet/weekly',
        'Y': './M4/TimesNet/yearly'
    }
    
    lstm_paths = {
        'D': './M4/LSTM/daily',
        'M': './M4/LSTM/monthly',
        'H': './M4/LSTM/hourly',
        'W': './M4/LSTM/weekly',
        'Y': './M4/LSTM/yearly'
    }
    
    tft_paths = {
        'D': './M4/TFT/daily',
        'M': './M4/TFT/monthly',
        'H': './M4/TFT/hourly',
        'W': './M4/TFT/weekly',
        'Y': './M4/TFT/yearly'
    }
    nhits_models = {freq: load_model(path, freq) for freq, path in nhits_paths.items()}
    timesnet_models = {freq: load_model(path, freq) for freq, path in timesnet_paths.items()}
    lstm_models = {freq: load_model(path, freq) for freq, path in lstm_paths.items()}
    tft_models = {freq: load_model(path, freq) for freq, path in tft_paths.items()}

    return nhits_models, timesnet_models, lstm_models, tft_models

def generate_forecast(model, df,tag=False):
    if tag == 'retrain':
        forecast_df = model.predict()
    else:
        forecast_df = model.predict(df=df)
    return forecast_df

def determine_frequency(df):
    df['ds'] = pd.to_datetime(df['ds'])
    df = df.drop_duplicates(subset='ds')
    df = df.set_index('ds')
    
    # # Create a complete date range
    # full_range = pd.date_range(start=df.index.min(), end=df.index.max(),freq=freq)
    
    # # Reindex the DataFrame to this full date range
    # df_full = df.reindex(full_range)
    
    # Infer the frequency
    # freq = pd.infer_freq(df_full.index)

    freq = pd.infer_freq(df.index)
    if not freq:
        st.warning('The forecast will use default Daily forecast due to date inconsistency. Please check your data.',icon="⚠️")
        freq = 'D'
        
    return freq

def plot_forecasts(forecast_df, train_df, title):
    fig, ax = plt.subplots(1, 1, figsize=(20, 7))
    plot_df = pd.concat([train_df, forecast_df]).set_index('ds')
    historical_col = 'y'
    forecast_col = next((col for col in plot_df.columns if 'median' in col), None)
    lo_col = next((col for col in plot_df.columns if 'lo-90' in col), None)
    hi_col = next((col for col in plot_df.columns if 'hi-90' in col), None)
    if forecast_col is None:
        raise KeyError("No forecast column found in the data.")
    plot_df[[historical_col, forecast_col]].plot(ax=ax, linewidth=2, label=['Historical', 'Forecast'])
    if lo_col and hi_col:
        ax.fill_between(
            plot_df.index,
            plot_df[lo_col],
            plot_df[hi_col],
            color='blue',
            alpha=0.3,
            label='90% Confidence Interval'
        )
    ax.set_title(title, fontsize=22)
    ax.set_ylabel('Value', fontsize=20)
    ax.set_xlabel('Timestamp [t]', fontsize=20)
    ax.legend(prop={'size': 15})
    ax.grid()
    st.pyplot(fig)

def select_model_based_on_frequency(freq, nhits_models, timesnet_models, lstm_models, tft_models):
    if freq == 'D':
        return nhits_models['D'], timesnet_models['D'], lstm_models['D'], tft_models['D']
    elif freq == 'ME':
        return nhits_models['M'], timesnet_models['M'], lstm_models['M'], tft_models['M']
    elif freq == 'H':
        return nhits_models['H'], timesnet_models['H'], lstm_models['H'], tft_models['H']
    elif freq in ['W', 'W-SUN']:
        return nhits_models['W'], timesnet_models['W'], lstm_models['W'], tft_models['W']
    elif freq in ['Y', 'Y-DEC']:
        return nhits_models['Y'], timesnet_models['Y'], lstm_models['Y'], tft_models['Y']
    else:
        raise ValueError(f"Unsupported frequency: {freq}")

def select_model(horizon, model_type, max_steps=50):
    if model_type == 'NHITS':
        return NHITS(input_size=5 * horizon,
                     h=horizon,
                     max_steps=max_steps,
                     stack_types=3*['identity'],
                     n_blocks=3*[1],
                     mlp_units=[[256, 256] for _ in range(3)],
                     n_pool_kernel_size=3*[1],
                     batch_size=32,
                     scaler_type='standard',
                     n_freq_downsample=[12, 4, 1],
                     loss=HuberMQLoss(level=[90]))
    elif model_type == 'TimesNet':
        return TimesNet(h=horizon,
                        input_size=horizon * 5,
                        hidden_size=16,
                        conv_hidden_size=32,
                        loss=HuberMQLoss(level=[90]),
                        scaler_type='standard',
                        learning_rate=1e-3,
                        max_steps=max_steps,
                        val_check_steps=200,
                        valid_batch_size=64,
                        windows_batch_size=128,
                        inference_windows_batch_size=512)
    elif model_type == 'LSTM':
        return LSTM(h=horizon,
                    input_size=horizon * 5,
                    loss=HuberMQLoss(level=[90]),
                    scaler_type='standard',
                    encoder_n_layers=2,
                    encoder_hidden_size=64,
                    context_size=10,
                    decoder_hidden_size=64,
                    decoder_layers=2,
                    max_steps=max_steps)
    elif model_type == 'TFT':
        return TFT(h=horizon,
                   input_size=horizon,
                   hidden_size=16,
                   loss=HuberMQLoss(level=[90]),
                   learning_rate=0.005,
                   scaler_type='standard',
                   windows_batch_size=128,
                   max_steps=max_steps,
                   val_check_steps=200,
                   valid_batch_size=64,
                   enable_progress_bar=True)
    else:
        raise ValueError(f"Unsupported model type: {model_type}")

def model_train(df,model, freq):
    nf = NeuralForecast(models=[model], freq=freq)
    df['ds'] = pd.to_datetime(df['ds'])
    nf.fit(df)
    return nf

def forecast_time_series(df, model_type, horizon, max_steps,y_col):
    start_time = time.time()  # Start timing
    freq = determine_frequency(df)
    st.sidebar.write(f"Data frequency: {freq}")
    
    selected_model = select_model(horizon, model_type, max_steps)
    st.spinner(f"Training {model_type} model...")
    model = model_train(df, selected_model,freq)
    
    forecast_results = {}
    forecast_results[model_type] = generate_forecast(model, df, tag='retrain')

    for model_name, forecast_df in forecast_results.items():
        plot_forecasts(forecast_df, df, f'{model_name} Forecast for {y_col}')
        
    end_time = time.time()  # End timing
    time_taken = end_time - start_time
    st.success(f"Time taken for {model_type} forecast: {time_taken:.2f} seconds")

@st.cache_data
def load_default():
    df = AirPassengersDF.copy()
    return df

def transfer_learning_forecasting():
    st.title("Zero-shot Forecasting")
    st.markdown("""
    Instant time series forecasting and visualization by using various pre-trained deep neural network-based model trained on M4 data.
    """)

    nhits_models, timesnet_models, lstm_models, tft_models = load_all_models()
    
    with st.sidebar.expander("Upload and Configure Dataset", expanded=True):
        if 'uploaded_file' not in st.session_state:
            uploaded_file = st.file_uploader("Upload your time series data (CSV)", type=["csv"])
            if uploaded_file:
                df = pd.read_csv(uploaded_file)
                st.session_state.df = df
                st.session_state.uploaded_file = uploaded_file
            else:
                df = load_default()
                st.session_state.df = df
        else:
            if st.checkbox("Upload a new file (CSV)"):
                uploaded_file = st.file_uploader("Upload your time series data (CSV)", type=["csv"])
                if uploaded_file:
                    df = pd.read_csv(uploaded_file)
                    st.session_state.df = df
                    st.session_state.uploaded_file = uploaded_file
                else:
                    df = st.session_state.df
            else:
                df = st.session_state.df
            
            
        columns = df.columns.tolist()  # Convert Index to list
        opt = []
        ds_col = st.selectbox("Select Date/Time column", options=columns, index=columns.index('ds') if 'ds' in columns else 0)
        if 'ds' in columns and 'unique_id' in columns:
            columns.pop(columns.index('ds'))
            columns.pop(columns.index('unique_id'))
        opt = columns
        if 'ds' in opt:
            opt.remove('ds')
        y_col = st.selectbox("Select Target column", options=opt, index=0)

    st.session_state.ds_col = ds_col
    st.session_state.y_col = y_col

    # Model selection and forecasting
    st.sidebar.subheader("Model Selection and Forecasting")
    model_choice = st.sidebar.selectbox("Select model", ["NHITS", "TimesNet", "LSTM", "TFT"])
    horizon = st.sidebar.number_input("Forecast horizon", value=12)

    df = df.rename(columns={ds_col: 'ds', y_col: 'y'})
    df['unique_id']=1
    df = df[['unique_id','ds','y']]
    st.session_state.df = df

    # Determine frequency of data
    frequency = determine_frequency(df)
    st.sidebar.write(f"Detected frequency: {frequency}")


    nhits_model, timesnet_model, lstm_model, tft_model = select_model_based_on_frequency(frequency, nhits_models, timesnet_models, lstm_models, tft_models)
    forecast_results = {}

    

    if st.sidebar.button("Submit"):
        start_time = time.time()  # Start timing
        if model_choice == "NHITS":
            forecast_results['NHITS'] = generate_forecast(nhits_model, df)
        elif model_choice == "TimesNet":
            forecast_results['TimesNet'] = generate_forecast(timesnet_model, df)
        elif model_choice == "LSTM":
            forecast_results['LSTM'] = generate_forecast(lstm_model, df)
        elif model_choice == "TFT":
            forecast_results['TFT'] = generate_forecast(tft_model, df)

        st.session_state.forecast_results = forecast_results
        for model_name, forecast_df in forecast_results.items():
            plot_forecasts(forecast_df.iloc[:horizon,:], df, f'{model_name} Forecast for {y_col}')

        end_time = time.time()  # End timing
        time_taken = end_time - start_time
        st.success(f"Time taken for {model_choice} forecast: {time_taken:.2f} seconds")

    if 'forecast_results' in st.session_state:
        forecast_results = st.session_state.forecast_results

        st.markdown('You can download Input and Forecast Data below')
        tab_insample, tab_forecast  = st.tabs(
                        ["Input data", "Forecast"]
                    )
            
        with tab_insample:
            df_grid = df.drop(columns="unique_id")
            st.write(df_grid)
            # grid_table = AgGrid(
            #                 df_grid,
            #                 theme="alpine",
            #             )
    
        with tab_forecast:
            if model_choice in forecast_results:
                df_grid = forecast_results[model_choice]
                st.write(df_grid)
                # grid_table = AgGrid(
                #                 df_grid,
                #                 theme="alpine",
                #             )


def dynamic_forecasting():
    st.title("Dynamic Forecasting")
    st.markdown("""
    Train time series forecasting model from scratch and provide forecasts/visualization by using various deep neural network-based model trained on user data.
    
    Forecasting speed depends on CPU/GPU availabilty.
    """)
    
    with st.sidebar.expander("Upload and Configure Dataset", expanded=True):
        uploaded_file = st.file_uploader("Upload your time series data (CSV)", type=["csv"])
        if uploaded_file:
            df = pd.read_csv(uploaded_file)
            st.session_state.df = df
        else:
            df = load_default()
            st.session_state.df = df
        
        columns = df.columns.tolist()  # Convert Index to list
        opt = []
        ds_col = st.selectbox("Select Date/Time column", options=columns, index=columns.index('ds') if 'ds' in columns else 0)
        if 'ds' in columns and 'unique_id' in columns:
            columns.pop(columns.index('ds'))
            columns.pop(columns.index('unique_id'))
        opt = columns
        y_col = st.selectbox("Select Target column", options=opt, index=0)

    df = df.rename(columns={ds_col: 'ds', y_col: 'y'})
    
    df['unique_id']=1
    df = df[['unique_id','ds','y']]
    st.session_state.df = df
    
    st.session_state.ds_col = ds_col
    st.session_state.y_col = y_col

    # Dynamic forecasting
    st.sidebar.subheader("Dynamic Model Selection and Forecasting")
    dynamic_model_choice = st.sidebar.selectbox("Select model for dynamic forecasting", ["NHITS", "TimesNet", "LSTM", "TFT"], key="dynamic_model_choice")
    dynamic_horizon = st.sidebar.number_input("Forecast horizon", value=12)
    dynamic_max_steps = st.sidebar.number_input('Max steps', value=20)

    if st.sidebar.button("Submit"):
        forecast_time_series(df, dynamic_model_choice, dynamic_horizon, dynamic_max_steps,y_col)

def timegpt_fcst():
    nixtla_token = os.environ.get("NIXTLA_API_KEY")
    nixtla_client = NixtlaClient(
    api_key = nixtla_token
    )

    
    st.title("TimeGPT Forecasting")
    st.markdown("""
    Instant time series forecasting and visualization by using the TimeGPT API provided by Nixtla.
    """)
    with st.sidebar.expander("Upload and Configure Dataset", expanded=True):
        uploaded_file = st.file_uploader("Upload your time series data (CSV)", type=["csv"])
        if uploaded_file:
            df = pd.read_csv(uploaded_file)
            st.session_state.df = df
        else:
            df = load_default()
            st.session_state.df = df
        
        columns = df.columns.tolist()  # Convert Index to list
        opt = []
        ds_col = st.selectbox("Select Date/Time column", options=columns, index=columns.index('ds') if 'ds' in columns else 0)
        if 'ds' in columns and 'unique_id' in columns:
            columns.pop(columns.index('ds'))
            columns.pop(columns.index('unique_id'))
        opt = columns
        y_col = st.selectbox("Select Target column", options=opt, index=0)
        h = st.number_input("Forecast horizon", value=14)

        df = df.rename(columns={ds_col: 'ds', y_col: 'y'})

        
        id_col = 'ts_test'
        df['unique_id']=id_col
        df = df[['unique_id','ds','y']]
        st.session_state.df = df
        
        st.session_state.ds_col = ds_col
        st.session_state.y_col = y_col
        

        freq = determine_frequency(df)

        df = df.drop_duplicates(subset=['ds']).reset_index(drop=True)
        
        # st.write(df)
        if st.sidebar.button("Submit"):
            forecast_df = nixtla_client.forecast(
                df=df,
                h=h,
                freq=freq,
                level=[90]
            )
            st.session_state.forecast_df = forecast_df

    if 'forecast_df' in st.session_state:
        forecast_df = st.session_state.forecast_df
        st.pyplot(nixtla_client.plot(df, forecast_df, level=[90]))



def timegpt_anom():
    nixtla_token = os.environ.get("NIXTLA_API_KEY")
    nixtla_client = NixtlaClient(
    api_key = nixtla_token
    )

    
    st.title("TimeGPT Anomaly Detection")
    st.markdown("""
    Instant time series anomaly detection and visualization by using the TimeGPT API provided by Nixtla.
    """)
    with st.sidebar.expander("Upload and Configure Dataset", expanded=True):
        uploaded_file = st.file_uploader("Upload your time series data (CSV)", type=["csv"])
        if uploaded_file:
            df = pd.read_csv(uploaded_file)
            st.session_state.df = df
        else:
            df = load_default()
            st.session_state.df = df
        
        columns = df.columns.tolist()  # Convert Index to list
        opt = []
        ds_col = st.selectbox("Select Date/Time column", options=columns, index=columns.index('ds') if 'ds' in columns else 0)
        if 'ds' in columns and 'unique_id' in columns:
            columns.pop(columns.index('ds'))
            columns.pop(columns.index('unique_id'))
        opt = columns
        y_col = st.selectbox("Select Target column", options=opt, index=0)

        df = df.rename(columns={ds_col: 'ds', y_col: 'y'})

        id_col = 'ts_test'
        df['unique_id']=id_col
        df = df[['unique_id','ds','y']]
        st.session_state.df = df
        
        st.session_state.ds_col = ds_col
        st.session_state.y_col = y_col

        freq = determine_frequency(df)

        df = df.drop_duplicates(subset=['ds']).reset_index(drop=True)
        if st.sidebar.button("Submit"):
            anom_df = nixtla_client.detect_anomalies(
                df=df,
                freq=freq,
                level=90
            )
            st.session_state.anom_df = anom_df

    if 'anom_df' in st.session_state:
        anom_df = st.session_state.anom_df
        st.pyplot(nixtla_client.plot(df, anom_df))
        
    
    
    

pg = st.navigation({
    "Neuralforecast": [
        # Load pages from functions
        st.Page(transfer_learning_forecasting, title="Zero-shot Forecasting", default=True, icon=":material/query_stats:"),
        st.Page(dynamic_forecasting, title="Dynamic Forecasting", icon=":material/monitoring:"),
    ],
        "TimeGPT": [
        # Load pages from functions
        st.Page(timegpt_fcst, title="TimeGPT Forecast", icon=":material/smart_toy:"),
        st.Page(timegpt_anom, title="TimeGPT Anomalies Detection", icon=":material/detector_offline:")
        ]
})

pg.run()