import pandas as pd import numpy as np from sklearn.linear_model import LinearRegression # for model from sklearn.preprocessing import MinMaxScaler #sklearn better for regression, statsmodels better for time series import statsmodels.api as sm from flask import Flask, render_template, request # webb app import warnings warnings.filterwarnings('ignore') app = Flask(__name__) print("models under construction") # build clone of modles for dashboard (not pullling models from testing) # ========================================== # load & train lcoal model (Linear Regression) # ========================================== df_local = pd.read_excel("delta_coffee_model_data~v1.xlsx", sheet_name=0) # grab only the columns we need df_local = df_local[['date.y', 'Median Price', 'milk', 'sugar', 'coco']] # drop all missing data df_local = df_local.dropna() # Calculate Deltas df_local_delta = df_local.copy() df_local_delta['Delta_Milk'] = df_local_delta['milk'].diff() df_local_delta['Delta_Sugar'] = df_local_delta['sugar'].diff() df_local_delta['Delta_Cocoa'] = df_local_delta['coco'].diff() df_local_delta['Delta_Coffee'] = df_local_delta['Median Price'].diff() # Drop the single NaN row created by the .diff() function df_local_delta = df_local_delta.dropna() # Train on 100% of data for production X_local = df_local_delta[['Delta_Milk', 'Delta_Sugar', 'Delta_Cocoa']] y_local = df_local_delta[['Delta_Coffee']] ##### scale using minmax: which means the highest price of sugar is a 1, and lowest is a 0. then every other price gets % inside ### where i found how to use minmaxscaler(): https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html scaler_X = MinMaxScaler() scaler_y = MinMaxScaler() X_scaled = scaler_X.fit_transform(X_local) y_scaled = scaler_y.fit_transform(y_local) #print(pd.DataFrame(X_scaled, columns=['Delta_Milk', 'Delta_Sugar', 'Delta_Cocoa'])) # print to veridy scaled correctly # lm_model.coef_ # show coefs lm_model = LinearRegression() lm_model.fit(X_scaled, y_scaled) # learning historical data, so when user imput comes, model can explain print("LM Micro model listo") # Grab baseline data for dashboard local_hist_dates = df_local['date.y'].astype(str).tolist() local_hist_prices = df_local['Median Price'].tolist() recent_local_price = local_hist_prices[-1] recent_milk = df_local['milk'].iloc[-1] #grab last row in the data frame recent_sugar = df_local['sugar'].iloc[-1] recent_cocoa = df_local['coco'].iloc[-1] # ========================================== # load and train (arima) # ========================================== df_global = pd.read_csv("master~global.csv").dropna() df_global.columns = df_global.columns.str.strip() df_global['Global_Arabica_Price'] = df_global['Global_Arabica_Price'] / 100 # Cents to Dollars # Calculate Deltas df_global_delta = df_global.copy() df_global_delta['Delta_Sugar'] = df_global_delta['Global_Sugar_Price'].diff() df_global_delta['Delta_Cocoa'] = df_global_delta['Global_Cocoa_Price'].diff() df_global_delta['Delta_Arabica'] = df_global_delta['Global_Arabica_Price'].diff() df_global_delta = df_global_delta.dropna() X_global = df_global_delta[['Delta_Sugar', 'Delta_Cocoa']] y_global = df_global_delta['Delta_Arabica'] # didnt scale arima data, because arima captures real-world price behavior. # arimax_model.params # show arima coefs # Train arima on all data arimax_model = sm.tsa.ARIMA(endog=y_global, exog=X_global, order=(1, 0, 0)).fit()#(1,0,0) only looking at prev 1 month print("Global Mac listo") global_hist_dates = df_global['observation_date'].astype(str).tolist() global_hist_prices = df_global['Global_Arabica_Price'].tolist() recent_global_price = global_hist_prices[-1] recent_global_sugar = df_global['Global_Sugar_Price'].iloc[-1] recent_global_cocoa = df_global['Global_Cocoa_Price'].iloc[-1] # ========================================== # web route, how live.html and app.py talk to eachother # ========================================== @app.route('/', methods=['GET', 'POST']) def index(): scroll_to = None # code from Dr.Mcvey # Set defaults for sliders milk_pct, local_sugar_pct, local_cocoa_pct = 0, 0, 0 global_sugar_pct, global_cocoa_pct = 0, 0 local_projections, global_projections = None, None if request.method == 'POST': # Grab from LOCAL sliders scroll_to = 'target-section' # code from Dr.Mcvey milk_pct = float(request.form.get('milk_shock', 0)) local_sugar_pct = float(request.form.get('local_sugar_shock', 0)) local_cocoa_pct = float(request.form.get('local_cocoa_shock', 0)) # Grab from GLOBAL sliders global_sugar_pct = float(request.form.get('global_sugar_shock', 0)) global_cocoa_pct = float(request.form.get('global_cocoa_shock', 0)) # local predict local_milk_delta = recent_milk * (milk_pct / 100) local_sugar_delta = recent_sugar * (local_sugar_pct / 100) local_cocoa_delta = recent_cocoa * (local_cocoa_pct / 100) user_input_scaled = scaler_X.transform([[local_milk_delta, local_sugar_delta, local_cocoa_delta]]) scaled_prediction = lm_model.predict(user_input_scaled) local_pred_change = scaler_y.inverse_transform(scaled_prediction)[0][0] #inverse_transformed is used to unscales and transforms to real usd # 3 month functions p3, p2, p1 = local_hist_prices[-3], local_hist_prices[-2], local_hist_prices[-1] pred_1 = np.median([p3, p2, p1]) + local_pred_change # median smoothing instead of true projecting ( 3 points) pred_2 = np.median([p2, p1, pred_1]) + local_pred_change pred_3 = np.median([p1, pred_1, pred_2]) + local_pred_change local_projections = [pred_1, pred_2, pred_3] # global predict global_sugar_delta = recent_global_sugar * (global_sugar_pct / 100) global_cocoa_delta = recent_global_cocoa * (global_cocoa_pct / 100) # Build exog data for future 3 points exog_future = pd.DataFrame({ 'Delta_Sugar': [global_sugar_delta, global_sugar_delta, global_sugar_delta], 'Delta_Cocoa': [global_cocoa_delta, global_cocoa_delta, global_cocoa_delta] }) # Forecast 3 steps out forecast = arimax_model.get_forecast(steps=3, exog=exog_future).predicted_mean.values # Add changes to the base price global_projections = [ recent_global_price + forecast[0], recent_global_price + forecast[0] + forecast[1], recent_global_price + forecast[0] + forecast[1] + forecast[2] ] # get coeffs for Local Model coef_milk, coef_sugar, coef_cocoa = lm_model.coef_[0] # going to pass the recent march 2026 data points to live.html front end so user knows what they are shocking... # get Coefficients for global # g the multipliers for the exog variables #g_coef_sugar = arimax_model.params['Delta_Sugar'] #g_coef_cocoa = arimax_model.params['Delta_Cocoa'] return render_template('live.html', scroll=scroll_to, # scroll method from Dr.Mcvey milk_pct=milk_pct, local_sugar_pct=local_sugar_pct, local_cocoa_pct=local_cocoa_pct, global_sugar_pct=global_sugar_pct, global_cocoa_pct=global_cocoa_pct, local_projections=local_projections, local_dates=local_hist_dates, local_prices=local_hist_prices, global_projections=global_projections, global_dates=global_hist_dates, global_prices=global_hist_prices, recent_milk=recent_milk, recent_sugar=recent_sugar, recent_cocoa=recent_cocoa, recent_global_sugar=recent_global_sugar, recent_global_cocoa=recent_global_cocoa, coef_milk=coef_milk, coef_sugar=coef_sugar, coef_cocoa=coef_cocoa, g_coef_sugar=g_coef_sugar, g_coef_cocoa=g_coef_cocoa) if __name__ == '__main__': app.run(host='0.0.0.0', port=3923, debug=True) #5029 for local, 3923 for compsci