get_forecast.py

import pandas as pd
from dateutil.relativedelta import relativedelta
import matplotlib.pyplot as plt
from prophet.serialize import model_from_json

def get_forecast(serie: str, periods, percent_change: int):
    
    if serie == 'SOM':
        model_file, data_file = 'som_model.json', 'som_data.csv'    
    elif serie == 'Volumen':
        model_file, data_file = 'volumen_model.json', 'volumen_data.csv'
    else:
        raise ValueError('Input not valid')
    
    # load model files
    with open(model_file, 'r') as fin:
        model = model_from_json(fin.read())  
    history = pd.read_csv(data_file, encoding = 'utf-8-sig', sep = ';')
    history['ds'] = pd.to_datetime(history['ds'])
        
    # make future dataframe
    future = model.make_future_dataframe(periods = periods, freq = 'MS')
    future = future.tail(periods)

    # filter prices last year
    last_year_dates = future.ds.apply(lambda x: x - relativedelta(years=1))
    past_prices = history[history['ds'].isin(last_year_dates)]['precio_hl'].values

    # generate new_prices
    
    percent_change = percent_change / 100
    
    new_prices = past_prices * (1 + percent_change)

    future['precio_hl'] = new_prices

    forecast = model.predict(future)

    future_values = forecast[['ds', 'yhat']]

    last_obs = history.iloc[-1:][['ds', 'y']].rename(columns = {'y': 'yhat'})
    future_aux = pd.concat([last_obs, future_values])

    fig = plt.figure()
    #plt.plot(history['ds'], history['y'], label = 'Historic data', marker = '.', color = 'C0')
    plt.plot(future_aux['ds'], future_aux['yhat'], label = 'Forecast', marker = '.', color = 'C1')
    plt.plot(history['ds'], history['y'], label = 'Historic data', marker = '.', color = 'C0')
    plt.xticks(rotation = 45)
    #plt.xlabel('Date')
    plt.ylabel(serie)
    plt.tight_layout()
    plt.legend()
    
    return fig