import streamlit as st import pandas as pd from datetime import timedelta def get_frequency(datasets, columns, load_options): freq_options = _frequency_options(datasets["data"], columns) frequency = st.selectbox("", freq_options, label_visibility="collapsed", index=1) #frequency settings if frequency == "Daily": freq = "D" freq_label = "days" elif frequency == "Hourly": freq = "H" freq_label = "hours" elif frequency == "Weekly": freq = "W" freq_label = "weeks" load_options["frequency"] = frequency load_options["freq"] = freq load_options["freq_label"] = freq_label return load_options def tmp_renaming_columns(datasets, columns, load_options): #renaming columns datasets["data"].rename(columns={columns["date_column"]: 'date'}, inplace=True) datasets["data"] = datasets["data"].sort_values(by='date', ascending=True) if load_options["two_files"]: datasets["data"]['food_amount'] = datasets["data"][columns["quantity_column"]] * datasets["data"][columns["quantity_per_item_column"]] else: datasets["data"].rename(columns={columns["quantity_column"]: 'food_amount'}, inplace=True) # returns frequency options available for forecast def _frequency_options(data, columns): contains_hours = (data[columns['date_column']].dt.hour != 0).any() contains_minutes = (data[columns['date_column']].dt.minute != 0).any() contains_seconds = (data[columns['date_column']].dt.second != 0).any() if contains_hours or contains_minutes or contains_seconds: return ["Hourly", "Daily", "Weekly"] elif columns["time_column"] != "": return ["Hourly", "Daily", "Weekly"] return ["Daily", "Weekly"] def get_dates(food_df): dates = dict() # Calculate initial_split_date as a Python datetime object initial_split_date = (food_df['date'].iloc[-1] - pd.DateOffset(weeks=4)).to_pydatetime() # Convert the Timestamp objects to Python datetime objects min_date = food_df['date'].min().to_pydatetime() max_date = food_df['date'].max().to_pydatetime() # Use Python datetime objects in the st.slider training_slider = st.slider( "Select the training data range:", value=(food_df['date'].iloc[0].to_pydatetime(), initial_split_date), min_value=min_date, max_value=max_date - timedelta(days=1) ) dates["train_start"], dates["train_end"] = training_slider # How many points in the future to predict test_intitial = dates["train_end"] + timedelta(days=1) testing_slider = st.slider( "Select the testing data range:", value=(test_intitial, max_date), min_value=test_intitial, max_value=max_date ) dates["test_start"], dates["test_end"] = testing_slider return dates def split_dataset(food_df, dates): # splitting data train_df = food_df[(food_df["date"] >= dates["train_start"]) & (food_df["date"] <= dates["train_end"])].copy() test_df = food_df[(food_df["date"] >= dates["test_start"]) & (food_df["date"] <= dates["test_end"])].copy() _display_dates(train_df, test_df) return train_df, test_df # display number of days in each split def _display_dates(train_df, test_df): training_duration = train_df['date'].max() - train_df['date'].min() training_days = training_duration.days testing_duration = test_df['date'].max() - test_df['date'].min() testing_days = testing_duration.days st.success(f''' Training Period: {training_days} days Testing Period: {testing_days} days ''') def get_forecast_dates(food_df, load_options): duration = st.number_input(f"Forecast length in {load_options['freq_label']}", min_value=1, max_value= 100, value=5) load_options["duration"] = duration min_date = food_df['date'].min().to_pydatetime() max_date = food_df['date'].max().to_pydatetime() forecast_slider = st.slider( "Select forecast training data range:", value=(food_df['date'].iloc[0].to_pydatetime(), max_date), min_value=min_date, max_value=max_date ) train_start, train_end = forecast_slider forecast_df = food_df[(food_df["date"] >= train_start) & (food_df["date"] <= train_end)].copy() return forecast_df, load_options