RFI / pages /9_Scenario_Planner.py
Manoj
latest
fde220d
# Importing necessary libraries
import streamlit as st
st.set_page_config(
page_title="Scenario Planner",
page_icon="⚖️",
layout="wide",
initial_sidebar_state="collapsed",
)
import os
import math
import pickle
import sqlite3
import numpy as np
from classes import numerize
import plotly.graph_objects as go
from collections import OrderedDict
from scipy.optimize import minimize
from utilities import project_selection, initialize_data, set_header, load_local_css
from utilities import (
get_panels_names,
get_metrics_names,
name_formating,
load_json_files,
load_pickle_files,
generate_rcs_data,
generate_scenario_data,
)
# Initialize ROI threshold
if "roi_threshold" not in st.session_state:
st.session_state.roi_threshold = 1
# Initialize message display holder
if "message_display" not in st.session_state:
st.session_state.message_display = {"type": "success", "message": None, "icon": ""}
# Function to reset modified_scenario_data
def reset_scenario(metrics_selected=None, panel_selected=None):
# Clear message_display
st.session_state.message_display = {"type": "success", "message": None, "icon": ""}
# Use default values from session state if not provided
if metrics_selected is None:
metrics_selected = st.session_state["response_metrics_selectbox_sp"]
if panel_selected is None:
panel_selected = st.session_state["response_panel_selectbox_sp"]
# Define the path to the pickle files
original_pickle_file_path = os.path.join(
st.session_state["project_path"], "scenario_data_original.pkl"
)
modified_pickle_file_path = os.path.join(
st.session_state["project_path"], "scenario_data_modified.pkl"
)
# Reset the modified_scenario_data back to the original_scenario_data
try:
# Open and load original scenario data
with open(original_pickle_file_path, "rb") as original_pickle_file:
original_data = pickle.load(original_pickle_file)
original_scenario_data = original_data[metrics_selected][panel_selected]
# Open and load modified scenario data
with open(modified_pickle_file_path, "rb+") as modified_pickle_file:
data = pickle.load(modified_pickle_file)
# Update the specific section with the original scenario data
data[metrics_selected][panel_selected] = original_scenario_data
# Go to the beginning of the file to overwrite it
modified_pickle_file.seek(0)
pickle.dump(data, modified_pickle_file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
return
# Function to build s curve
def s_curve(x, power, K, b, a, x0):
return K / (1 + b * np.exp(-a * ((x / 10**power) - x0)))
# Function to retrieve S-curve parameters for a given metric, panel, and channel
def get_s_curve_params(
metrics_selected,
panel_selected,
channel_selected,
original_json_data,
modified_json_data,
modified_pickle_file_path,
):
# Retrieve 'power' parameter from the original data for the specific metric, panel, and channel
power = original_json_data[metrics_selected][panel_selected][channel_selected][
"power"
]
# Get the S-curve parameters from the modified data for the same metric, panel, and channel
s_curve_param = modified_json_data[metrics_selected][panel_selected][
channel_selected
]
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update modified S-curve parameters
data[metrics_selected][panel_selected]["channels"][channel_selected][
"response_curve_params"
] = s_curve_param
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update the 'power' parameter in the modified S-curve parameters with the original 'power' value
s_curve_param["power"] = power
# Return the updated S-curve parameters
return s_curve_param
# Function to calculate total contribution
def get_total_contribution(
spends, channels, s_curve_params, channels_proportion, modified_scenario_data
):
total_contribution = 0
for i in range(len(channels)):
channel_name = channels[i]
channel_s_curve_params = s_curve_params[channel_name]
spend_proportion = spends[i] * channels_proportion[channel_name]
total_contribution += sum(
s_curve(
spend_proportion,
channel_s_curve_params["power"],
channel_s_curve_params["K"],
channel_s_curve_params["b"],
channel_s_curve_params["a"],
channel_s_curve_params["x0"],
)
)
return total_contribution + sum(modified_scenario_data["constant"])
# Function to calculate total spends
def get_total_spends(spends, channels_conversion_ratio):
return np.sum(spends * np.array(list(channels_conversion_ratio.values())))
# Function to optimizes spends for all channels given bounds and a total spend target
def optimizer(
optimization_goal,
s_curve_params,
channels_spends,
channels_proportion,
channels_conversion_ratio,
total_target,
bounds_dict,
modified_scenario_data,
):
# Extract channel names and corresponding actual spends
channels = list(channels_spends.keys())
actual_spends = np.array(list(channels_spends.values()))
num_channels = len(actual_spends)
# Define the objective function based on the optimization goal
def objective_fun(spends):
if optimization_goal == "Spends":
# Minimize negative total contribution to maximize the total contribution
return -get_total_contribution(
spends,
channels,
s_curve_params,
channels_proportion,
modified_scenario_data,
)
else:
# Minimize total spends
return get_total_spends(spends, channels_conversion_ratio)
def constraint_fun(spends):
if optimization_goal == "Spends":
# Ensure the total spends equals the total spend target
return get_total_spends(spends, channels_conversion_ratio)
else:
# Ensure the total contribution equals the total contribution target
return get_total_contribution(
spends,
channels,
s_curve_params,
channels_proportion,
modified_scenario_data,
)
# Equality constraint
constraints = {
"type": "eq",
"fun": lambda spends: constraint_fun(spends) - total_target,
} # Sum of all channel spends/metrics should equal the total spend/metrics target
# Bounds for each channel's spend based
bounds = [
(
actual_spends[i] * (1 + bounds_dict[channels[i]][0] / 100),
actual_spends[i] * (1 + bounds_dict[channels[i]][1] / 100),
)
for i in range(num_channels)
]
# Initial guess for the optimization
initial_guess = np.array(actual_spends)
# Calculate xtol as 0.1% of the minimum of spends
xtol = max(10, 0.001 * np.min(actual_spends))
# Perform the optimization using 'trust-constr' method
result = minimize(
objective_fun,
initial_guess,
method="trust-constr",
constraints=constraints,
bounds=bounds,
options={
"disp": True, # Display the optimization process
"xtol": xtol, # Dynamic step size tolerance
"maxiter": 1e5, # Maximum number of iterations
},
)
# Print the optimization result
print(result)
# Extract the optimized spends from the result
optimized_spends_array = result.x
# Convert optimized spends back to a dictionary with channel names
optimized_spends = {
channels[i]: optimized_spends_array[i] for i in range(num_channels)
}
return optimized_spends, result.success
# Function to calculate achievable targets at lower and upper spend bounds
@st.cache_data(show_spinner=False)
def max_target_achievable(
channels_spends,
s_curve_params,
channels_proportion,
modified_scenario_data,
bounds_dict,
):
# Extract channel names and corresponding actual spends
channels = list(channels_spends.keys())
actual_spends = np.array(list(channels_spends.values()))
num_channels = len(actual_spends)
# Bounds for each channel's spend
lower_spends, upper_spends = [], []
for i in range(num_channels):
lower_spends.append(actual_spends[i] * (1 + bounds_dict[channels[i]][0] / 100))
upper_spends.append(actual_spends[i] * (1 + bounds_dict[channels[i]][1] / 100))
# Calculate achievable targets at lower and upper spend bounds
lower_achievable_target = get_total_contribution(
lower_spends,
channels,
s_curve_params,
channels_proportion,
modified_scenario_data,
)
upper_achievable_target = get_total_contribution(
upper_spends,
channels,
s_curve_params,
channels_proportion,
modified_scenario_data,
)
# Return achievable targets with ±0.1% safety margin
return max(0, 1.001 * lower_achievable_target), 0.999 * upper_achievable_target
# Function to check if number is in valid format
def is_valid_number_format(number_str):
# Check for None
if number_str is None:
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
# Define the valid suffixes
valid_suffixes = {"K", "M", "B", "T"}
# Check for negative numbers
if number_str[0] == "-":
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
# Check if the string ends with a digit
if number_str[-1].isdigit():
try:
# Attempt to convert the entire string to float
number = float(number_str)
# Ensure the number is non-negative
if number >= 0:
return True
else:
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
except ValueError:
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
# Check if the string ends with a valid suffix
suffix = number_str[-1].upper()
if suffix in valid_suffixes:
num_part = number_str[:-1] # Extract the numerical part
try:
# Attempt to convert the numerical part to float
number = float(num_part)
# Ensure the number part is non-negative
if number >= 0:
return True
else:
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
except ValueError:
# Store the message details in session state for invalid input
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
# If neither condition is met, return False
st.session_state.message_display = {
"type": "warning",
"message": "Invalid input: Please enter a valid number.",
"icon": "⚠️",
}
return False
# Function to converts a string with number suffixes (K, M, B, T) to a float
def convert_to_float(number_str):
# Dictionary mapping suffixes to their multipliers
multipliers = {
"K": 1e3, # Thousand
"M": 1e6, # Million
"B": 1e9, # Billion
"T": 1e12, # Trillion
}
# If there's no suffix, directly convert to float
if number_str[-1].isdigit():
return float(number_str)
# Extract the suffix (last character) and the numerical part
suffix = number_str[-1].upper()
num_part = number_str[:-1]
# Convert the numerical part to float and multiply by the corresponding multiplier
return float(num_part) * multipliers[suffix]
# Function to update absolute_channel_spends change
def absolute_channel_spends_change(
channel_key,
channel_spends_actual,
channel,
metrics_selected,
panel_selected,
modified_pickle_file_path,
):
# Do not update if the number is in an invalid format
if not is_valid_number_format(st.session_state[f"{channel_key}_abs_spends_key"]):
return
# Get updated absolute spends from session state
new_absolute_spends = convert_to_float(
st.session_state[f"{channel_key}_abs_spends_key"]
)
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Total channel spends
total_channel_spends = 0
for current_channel in list(
data[metrics_selected][panel_selected]["channels"].keys()
):
# Channel key
channel_key = f"{metrics_selected}_{panel_selected}_{current_channel}"
total_channel_spends += convert_to_float(
st.session_state[f"{channel_key}_abs_spends_key"]
)
# Check if total channel spends are within the allowed range (±50% of the original total spends)
if (
total_channel_spends
< 1.5 * data[metrics_selected][panel_selected]["actual_total_spends"]
and total_channel_spends
> 0.5 * data[metrics_selected][panel_selected]["actual_total_spends"]
):
# Update the modified_total_spends for the specified channel
data[metrics_selected][panel_selected]["channels"][channel][
"modified_total_spends"
] = new_absolute_spends / float(
data[metrics_selected][panel_selected]["channels"][channel][
"conversion_rate"
]
)
# Update total spends
data[metrics_selected][panel_selected][
"modified_total_spends"
] = total_channel_spends
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
else:
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Keep total spending within ±50% of the original value.",
"icon": "⚠️",
}
# Function to update percentage_channel_spends change
def percentage_channel_spends_change(
channel_key,
channel_spends_actual,
channel,
metrics_selected,
panel_selected,
modified_pickle_file_path,
):
# Retrieve the percentage spend change from session state
percentage_channel_spends = round(
st.session_state[f"{channel_key}_per_spends_key"], 0
)
# Calculate the new absolute spends
new_absolute_spends = channel_spends_actual * (1 + percentage_channel_spends / 100)
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Total channel spends
total_channel_spends = 0
for current_channel in list(
data[metrics_selected][panel_selected]["channels"].keys()
):
# Channel key
channel_key = f"{metrics_selected}_{panel_selected}_{current_channel}"
# Current channel spends actual
current_channel_spends_actual = data[metrics_selected][panel_selected][
"channels"
][current_channel]["actual_total_spends"]
# Current channel conversion rate
current_channel_conversion_rate = data[metrics_selected][panel_selected][
"channels"
][current_channel]["conversion_rate"]
# Calculate the current channel absolute spends
current_channel_absolute_spends = (
current_channel_spends_actual
* current_channel_conversion_rate
* (1 + st.session_state[f"{channel_key}_per_spends_key"] / 100)
)
total_channel_spends += current_channel_absolute_spends
# Check if total channel spends are within the allowed range (±50% of the original total spends)
if (
total_channel_spends
< 1.5 * data[metrics_selected][panel_selected]["actual_total_spends"]
and total_channel_spends
> 0.5 * data[metrics_selected][panel_selected]["actual_total_spends"]
):
# Update the modified_total_spends for the specified channel
data[metrics_selected][panel_selected]["channels"][channel][
"modified_total_spends"
] = float(new_absolute_spends) / float(
data[metrics_selected][panel_selected]["channels"][channel][
"conversion_rate"
]
)
# # Update total spends
# data[metrics_selected][panel_selected][
# "modified_total_spends"
# ] = total_channel_spends
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Function to update total input change
def total_input_change(modified_pickle_file_path, per_change):
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Get the list of all channels in the specified panel and metric
channel_list = list(data[metrics_selected][panel_selected]["channels"].keys())
# # Calculate the total actual spends excluding constant values
# total_actual_spends = data[metrics_selected][panel_selected]["actual_total_spends"]
# Iterate over each channel to update their modified spends
for channel in channel_list:
# Retrieve the actual spends for the channel
channel_actual_spends = data[metrics_selected][panel_selected]["channels"][
channel
]["actual_total_spends"]
# Calculate the modified spends for the channel based on the percent change
modified_channel_metrics = channel_actual_spends * ((100 + per_change) / 100)
# Update the channel's modified total spends in the data
data[metrics_selected][panel_selected]["channels"][channel][
"modified_total_spends"
] = modified_channel_metrics
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Function to update total_absolute_main_key change
def total_absolute_main_key_change(
metrics_selected, panel_selected, modified_pickle_file_path, optimization_goal
):
# Do not update if the number is in an invalid format
if not is_valid_number_format(st.session_state["total_absolute_main_key"]):
return
# Get updated absolute from session state
new_absolute = convert_to_float(st.session_state["total_absolute_main_key"])
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
if optimization_goal == "Spends":
# Retrieve the old absolute spends
old_absolute = data[metrics_selected][panel_selected]["actual_total_spends"]
else:
# Retrieve the old absolute metrics
old_absolute = data[metrics_selected][panel_selected]["actual_total_sales"]
# Calculate the allowable range for new spends
lower_bound = old_absolute * 0.5
upper_bound = old_absolute * 1.5
# Ensure the new spends are within ±50% of the old value
if new_absolute < lower_bound or new_absolute > upper_bound:
new_absolute = old_absolute
if optimization_goal == "Spends":
# Update the modified_total_spends with the constrained value
data[metrics_selected][panel_selected]["modified_total_spends"] = new_absolute
else:
# Update the modified_total_sales with the constrained value
data[metrics_selected][panel_selected]["modified_total_sales"] = new_absolute
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update total input change
if optimization_goal == "Spends":
per_change = ((new_absolute - old_absolute) / old_absolute) * 100
total_input_change(modified_pickle_file_path, per_change)
# Function to update total_absolute_key change
def total_absolute_key_change(
metrics_selected, panel_selected, modified_pickle_file_path, optimization_goal
):
# Get updated absolute from session state
new_absolute = convert_to_float(st.session_state["total_absolute_key"])
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
if optimization_goal == "Spends":
# Update the modified_total_spends for the specified channel
data[metrics_selected][panel_selected]["modified_total_spends"] = new_absolute
old_absolute = data[metrics_selected][panel_selected]["actual_total_spends"]
else:
# Update the modified_total_sales for the specified channel
data[metrics_selected][panel_selected]["modified_total_sales"] = new_absolute
old_absolute = data[metrics_selected][panel_selected]["actual_total_sales"]
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update total input change
if optimization_goal == "Spends":
per_change = ((new_absolute - old_absolute) / old_absolute) * 100
total_input_change(modified_pickle_file_path, per_change)
# Function to update total_absolute_key change
def total_percentage_key_change(
metrics_selected,
panel_selected,
modified_pickle_file_path,
absolute_value,
optimization_goal,
):
# Get updated absolute from session state
new_absolute = absolute_value * (1 + st.session_state["total_percentage_key"] / 100)
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
if optimization_goal == "Spends":
# Update the modified_total_spends for the specified channel
data[metrics_selected][panel_selected]["modified_total_spends"] = new_absolute
old_absolute = data[metrics_selected][panel_selected]["actual_total_spends"]
else:
# Update the modified_total_sales for the specified channel
data[metrics_selected][panel_selected]["modified_total_sales"] = new_absolute
old_absolute = data[metrics_selected][panel_selected]["actual_total_sales"]
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update total input change
if optimization_goal == "Spends":
per_change = ((new_absolute - old_absolute) / old_absolute) * 100
total_input_change(modified_pickle_file_path, per_change)
# Function to update bound change
def bound_change(
metrics_selected, panel_selected, modified_pickle_file_path, channel_key, channel
):
# Get updated bounds from session state
new_lower_bound = st.session_state[f"{channel_key}_lower_key"]
new_upper_bound = st.session_state[f"{channel_key}_upper_key"]
if new_lower_bound > new_upper_bound:
new_bounds = [-10, 10]
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Lower bound cannot be greater than Upper bound.",
"icon": "⚠️",
}
else:
new_bounds = [new_lower_bound, new_upper_bound]
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update the bounds for the specified channel
data[metrics_selected][panel_selected]["channels"][channel]["bounds"] = new_bounds
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Function to update freeze change
def freeze_change(
metrics_selected,
panel_selected,
modified_pickle_file_path,
channel_key,
channel,
):
if st.session_state[f"{channel_key}_allow_optimize_key"]:
# Updated bounds from session state
new_lower_bound, new_upper_bound = 0, 0
new_bounds = [new_lower_bound, new_upper_bound]
new_freeze = True
else:
# Updated bounds from session state
new_lower_bound, new_upper_bound = -10, 10
new_bounds = [new_lower_bound, new_upper_bound]
new_freeze = False
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update the bounds for the specified channel
data[metrics_selected][panel_selected]["channels"][channel]["bounds"] = new_bounds
data[metrics_selected][panel_selected]["channels"][channel]["freeze"] = new_freeze
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Function to calculate y, ROI and MROI for given point
def get_point_parms(
x_val, current_s_curve_params, current_channel_proportion, current_conversion_rate
):
# Calculate y value for the given spend point
y_val = sum(
s_curve(
(x_val * current_channel_proportion),
current_s_curve_params["power"],
current_s_curve_params["K"],
current_s_curve_params["b"],
current_s_curve_params["a"],
current_s_curve_params["x0"],
)
)
# Calculate MROI using a small nudge for actual spends
nudge = 1e-3
x1 = float(x_val * current_conversion_rate)
y1 = float(y_val)
x2 = x1 + nudge
y2 = sum(
s_curve(
((x2 / current_conversion_rate) * current_channel_proportion),
current_s_curve_params["power"],
current_s_curve_params["K"],
current_s_curve_params["b"],
current_s_curve_params["a"],
current_s_curve_params["x0"],
)
)
mroi_val = (float(y2) - y1) / (x2 - x1) if x2 != x1 else 0
# Calculate ROI
roi_val = y_val / (x_val * current_conversion_rate)
return roi_val, mroi_val, y_val
# Function to find segment value
def find_segment_value(x, roi, mroi, roi_threshold=1, mroi_threshold=0.05):
# Initialize the start and end values of the x array
start_value = x[0]
end_value = x[-1]
# Define the condition for the "green region" where both ROI and MROI exceed their respective thresholds
green_condition = (roi > roi_threshold) & (mroi > mroi_threshold)
# Find indices where ROI exceeds the ROI threshold
left_indices = np.where(roi > roi_threshold)[0]
# Find indices where both ROI and MROI exceed their thresholds (green condition)
right_indices = np.where(green_condition)[0]
# Determine the left value based on the first index where ROI exceeds the threshold
left_value = x[left_indices[0]] if left_indices.size > 0 else x[0]
# Determine the right value based on the last index where both ROI and MROI exceed their thresholds
right_value = x[right_indices[-1]] if right_indices.size > 0 else x[0]
# Ensure the left value does not exceed the right value, adjust if necessary
if left_value > right_value:
left_value = right_value
return start_value, end_value, left_value, right_value
# Function to generate response curve plots
@st.cache_data(show_spinner=False)
def generate_response_curve_plots(
channel_list, s_curve_params, channels_proportion, original_scenario_data
):
figures, channel_roi_mroi, region_start_end = [], {}, {}
for channel in channel_list:
spends_actual = original_scenario_data["channels"][channel][
"actual_total_spends"
]
conversion_rate = original_scenario_data["channels"][channel]["conversion_rate"]
x_actual = np.linspace(0, 5 * spends_actual, 100)
x_plot = x_actual * conversion_rate
# Calculate y values for the S-curve
y_plot = [
sum(
s_curve(
(x * channels_proportion[channel]),
s_curve_params[channel]["power"],
s_curve_params[channel]["K"],
s_curve_params[channel]["b"],
s_curve_params[channel]["a"],
s_curve_params[channel]["x0"],
)
)
for x in x_actual
]
# Calculate ROI and ensure they are scalar values
roi = [float(y) / float(x) if x != 0 else 0 for x, y in zip(x_plot, y_plot)]
# Calculate MROI using a small nudge
nudge = 1e-3
mroi = []
for i in range(len(x_plot)):
x1 = float(x_plot[i])
y1 = float(y_plot[i])
x2 = x1 + nudge
y2 = sum(
s_curve(
((x2 / conversion_rate) * channels_proportion[channel]),
s_curve_params[channel]["power"],
s_curve_params[channel]["K"],
s_curve_params[channel]["b"],
s_curve_params[channel]["a"],
s_curve_params[channel]["x0"],
)
)
mroi_value = (float(y2) - y1) / (x2 - x1) if x2 != x1 else 0
mroi.append(mroi_value)
# Calculate y, ROI and MROI for the actual spend point
roi_actual, mroi_actual, y_actual = get_point_parms(
spends_actual,
s_curve_params[channel],
channels_proportion[channel],
conversion_rate,
)
# Create the plotly figure
fig = go.Figure()
# Add S-curve line
fig.add_trace(
go.Scatter(
x=x_plot,
y=y_plot,
mode="lines",
name="Metrics",
hoverinfo="text",
text=[
f"Spends: {numerize(x)}<br>{metrics_selected_formatted}: {numerize(y)}<br>ROI: {r:.2f}<br>MROI: {m:.2f}"
for x, y, r, m in zip(x_plot, y_plot, roi, mroi)
],
)
)
# Add current spend point
fig.add_trace(
go.Scatter(
x=[spends_actual * conversion_rate],
y=[y_actual],
mode="markers",
marker=dict(color="cyan", size=10, symbol="circle"),
name="Actual Spend",
hoverinfo="text",
text=[
f"Actual Spend: {numerize(spends_actual * conversion_rate)}<br>{metrics_selected_formatted}: {numerize(y_actual)}<br>ROI: {roi_actual:.2f}<br>MROI: {mroi_actual:.2f}"
],
showlegend=True,
)
)
# ROI Threshold
roi_threshold = st.session_state.roi_threshold
# Scale x and y values
x, y = np.array(x_plot), np.array(y_plot)
x_scaled, y_scaled = x / max(x), y / max(y)
# Calculate MROI scaled starting from the first point
mroi_scaled = np.zeros_like(x_scaled)
for j in range(1, len(x_scaled)):
x1, y1 = x_scaled[j - 1], y_scaled[j - 1]
x2, y2 = x_scaled[j], y_scaled[j]
mroi_scaled[j] = (y2 - y1) / (x2 - x1) if (x2 - x1) != 0 else 0
# Get the start_value, end_value, left_value, right_value for segments
start_value, end_value, left_value, right_value = find_segment_value(
x_plot, np.array(roi), mroi_scaled, roi_threshold, 0.05
)
# Store region start and end points
region_start_end[channel] = {
"start_value": start_value,
"end_value": end_value,
"left_value": left_value,
"right_value": right_value,
}
# Adding background colors
y_max = max(y_plot) * 1.3 # 30% extra space above the max
# Yellow region
fig.add_shape(
type="rect",
x0=start_value,
y0=0,
x1=left_value,
y1=y_max,
line=dict(width=0),
fillcolor="rgba(255, 255, 0, 0.3)",
layer="below",
)
# Green region
fig.add_shape(
type="rect",
x0=left_value,
y0=0,
x1=right_value,
y1=y_max,
line=dict(width=0),
fillcolor="rgba(0, 255, 0, 0.3)",
layer="below",
)
# Red region
fig.add_shape(
type="rect",
x0=right_value,
y0=0,
x1=end_value,
y1=y_max,
line=dict(width=0),
fillcolor="rgba(255, 0, 0, 0.3)",
layer="below",
)
# Layout adjustments
fig.update_layout(
title=f"{name_formating(channel)}",
showlegend=False,
xaxis=dict(
showgrid=True,
showticklabels=True,
tickformat=".2s",
gridcolor="lightgrey",
gridwidth=0.5,
griddash="dot",
),
yaxis=dict(
showgrid=True,
showticklabels=True,
tickformat=".2s",
gridcolor="lightgrey",
gridwidth=0.5,
griddash="dot",
),
template="plotly_white",
margin=dict(l=20, r=20, t=30, b=20),
height=100 * math.ceil(len(channel_list) / 4),
)
figures.append(fig)
# Store data of each channel ROI and MROI
channel_roi_mroi[channel] = {
"actual_roi": roi_actual,
"actual_mroi": mroi_actual,
}
return figures, channel_roi_mroi, region_start_end
# Function to add modified spends/metrics point on plot
def modified_metrics_point(
fig, modified_spends, s_curve_params, channels_proportion, conversion_rate
):
# Calculate ROI, MROI, and y for the modified point
roi_modified, mroi_modified, y_modified = get_point_parms(
modified_spends, s_curve_params, channels_proportion, conversion_rate
)
# Add modified spend point
fig.add_trace(
go.Scatter(
x=[modified_spends * conversion_rate],
y=[y_modified],
mode="markers",
marker=dict(color="blueviolet", size=10, symbol="circle"),
name="Optimized Spend",
hoverinfo="text",
text=[
f"Modified Spend: {numerize(modified_spends * conversion_rate)}<br>{metrics_selected_formatted}: {numerize(y_modified)}<br>ROI: {roi_modified:.2f}<br>MROI: {mroi_modified:.2f}"
],
showlegend=True,
)
)
return roi_modified, mroi_modified, fig
# Function to update bound type change
def bound_type_change(modified_pickle_file_path):
# Get updated bound type from session state
new_bound_type = st.session_state["bound_type_key"]
# Open the pickle file and load the data
try:
with open(modified_pickle_file_path, "rb") as file:
data = pickle.load(file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Update the bound type
data[metrics_selected][panel_selected]["bound_type"] = new_bound_type
# Set bounds to default value if bound type is False (Default)
channel_list = list(data[metrics_selected][panel_selected]["channels"].keys())
if not new_bound_type:
for channel in channel_list:
data[metrics_selected][panel_selected]["channels"][channel]["bounds"] = [
-10,
10,
]
# Save the updated data back to the pickle file
try:
with open(modified_pickle_file_path, "wb") as file:
pickle.dump(data, file)
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Function to format the numbers with decimal
def format_value(input_value):
value = abs(input_value)
return f"{input_value:.4f}" if value < 1 else f"{numerize(input_value, 1)}"
# Function to format the numbers with decimal
def round_value(input_value):
value = abs(input_value)
return round(input_value, 4) if value < 1 else round(input_value, 1)
# Function to generate ROI and MROI plots for all channels
@st.cache_data(show_spinner=False)
def roi_mori_plot(channel_roi_mroi):
# Dictionary to store plots
channel_roi_mroi_plot = {}
for channel in channel_roi_mroi:
channel_roi_mroi_data = channel_roi_mroi[channel]
# Extract the data
actual_roi = channel_roi_mroi_data["actual_roi"]
optimized_roi = channel_roi_mroi_data["optimized_roi"]
actual_mroi = channel_roi_mroi_data["actual_mroi"]
optimized_mroi = channel_roi_mroi_data["optimized_mroi"]
# Plot ROI
fig_roi = go.Figure()
fig_roi.add_trace(
go.Bar(
x=["Actual ROI"],
y=[actual_roi],
name="Actual ROI",
marker_color="cyan",
width=1,
text=[format_value(actual_roi)],
textposition="auto",
textfont=dict(color="black", size=14),
)
)
fig_roi.add_trace(
go.Bar(
x=["Optimized ROI"],
y=[optimized_roi],
name="Optimized ROI",
marker_color="blueviolet",
width=1,
text=[format_value(optimized_roi)],
textposition="auto",
textfont=dict(color="black", size=14),
)
)
fig_roi.update_layout(
annotations=[
dict(
x=0.5,
y=1.3,
xref="paper",
yref="paper",
text="ROI",
showarrow=False,
font=dict(size=14),
)
],
barmode="group",
bargap=0,
showlegend=False,
width=110,
height=110,
xaxis=dict(
showticklabels=True,
showgrid=False,
tickangle=0,
ticktext=["Actual", "Optimized"],
tickvals=["Actual ROI", "Optimized ROI"],
),
yaxis=dict(showticklabels=False, showgrid=False),
margin=dict(t=20, b=20, r=0, l=0),
)
# Plot MROI
fig_mroi = go.Figure()
fig_mroi.add_trace(
go.Bar(
x=["Actual MROI"],
y=[actual_mroi],
name="Actual MROI",
marker_color="cyan",
width=1,
text=[format_value(actual_mroi)],
textposition="auto",
textfont=dict(color="black", size=14),
)
)
fig_mroi.add_trace(
go.Bar(
x=["Optimized MROI"],
y=[optimized_mroi],
name="Optimized MROI",
marker_color="blueviolet",
width=1,
text=[format_value(optimized_mroi)],
textposition="auto",
textfont=dict(color="black", size=14),
)
)
fig_mroi.update_layout(
annotations=[
dict(
x=0.5,
y=1.3,
xref="paper",
yref="paper",
text="MROI",
showarrow=False,
font=dict(size=14),
)
],
barmode="group",
bargap=0,
showlegend=False,
width=110,
height=110,
xaxis=dict(
showticklabels=True,
showgrid=False,
tickangle=0,
ticktext=["Actual", "Optimized"],
tickvals=["Actual MROI", "Optimized MROI"],
),
yaxis=dict(showticklabels=False, showgrid=False),
margin=dict(t=20, b=20, r=0, l=0),
)
# Store plots
channel_roi_mroi_plot[channel] = {"fig_roi": fig_roi, "fig_mroi": fig_mroi}
return channel_roi_mroi_plot
# Function to save the current scenario with the mentioned name
def save_scenario(
scenario_dict, metrics_selected, panel_selected, optimization_goal, channel_roi_mroi
):
# Remove extra space at start and ends
if st.session_state["scenario_name"] is not None:
st.session_state["scenario_name"] = st.session_state["scenario_name"].strip()
if (
st.session_state["scenario_name"] is None
or st.session_state["scenario_name"] == ""
):
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Please provide a name to save the scenario.",
"icon": "⚠️",
}
return
# Check if the dictionary is empty
if not scenario_dict:
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Nothing to save. The scenario data is empty.",
"icon": "⚠️",
}
return
# Add additional scenario details
scenario_dict["panel_selected"] = panel_selected
scenario_dict["metrics_selected"] = metrics_selected
scenario_dict["optimization"] = optimization_goal
scenario_dict["channel_roi_mroi"] = channel_roi_mroi
# Path to the saved scenarios file
saved_scenarios_dict_path = os.path.join(
st.session_state["project_path"], "saved_scenarios.pkl"
)
# Load existing scenarios if the file exists
try:
if os.path.exists(saved_scenarios_dict_path):
with open(saved_scenarios_dict_path, "rb") as f:
saved_scenarios_dict = pickle.load(f)
else:
saved_scenarios_dict = OrderedDict()
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Check if the name is already taken
if st.session_state["scenario_name"] in saved_scenarios_dict.keys():
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Name already exists. Please change the name or delete the existing scenario from the Saved Scenario page.",
"icon": "⚠️",
}
return
# Update the dictionary with the new scenario
saved_scenarios_dict[st.session_state["scenario_name"]] = scenario_dict
# Save the updated dictionary back to the file
try:
with open(saved_scenarios_dict_path, "wb") as f:
pickle.dump(saved_scenarios_dict, f)
# Store the message details in session state
st.session_state.message_display = {
"type": "success",
"message": f"Scenario '{st.session_state.scenario_name}' has been successfully saved!",
"icon": "💾",
}
except:
st.toast("Failed to Load/Update. Tool reset to default settings.", icon="⚠️")
reset_scenario()
return
# Clear the scenario name input
st.session_state["scenario_name"] = ""
# Function to calculate the RGBA color code based on the spends value and region boundaries
def calculate_rgba(spends_value, region_start_end):
# Get region start and end points
start_value = region_start_end["start_value"]
end_value = region_start_end["end_value"]
left_value = region_start_end["left_value"]
right_value = region_start_end["right_value"]
# Calculate alpha dynamically based on the position within the range
def calculate_alpha(position, start, end, min_alpha=0.1, max_alpha=0.4):
return min_alpha + (max_alpha - min_alpha) * (position - start) / (end - start)
if start_value <= spends_value <= left_value:
# Yellow range (0, 128, 0) - More transparent towards left, darker towards start
alpha = calculate_alpha(spends_value, left_value, start_value)
return (255, 255, 0, alpha) # RGB for yellow
elif left_value < spends_value <= right_value:
# Green range (0, 128, 0) - More transparent towards right, darker towards left
alpha = calculate_alpha(spends_value, right_value, left_value)
return (0, 128, 0, alpha) # RGB for green
elif right_value < spends_value <= end_value:
# Red range (255, 0, 0) - More transparent towards right, darker towards end
alpha = calculate_alpha(spends_value, right_value, end_value)
return (255, 0, 0, alpha) # RGB for red
# Function to format and display the channel name with a color and background color
def display_channel_name_with_background_color(
channel_name, background_color=(0, 128, 0, 0.1)
):
formatted_name = name_formating(channel_name)
# Unpack the RGBA values
r, g, b, a = background_color
# Create the HTML content with specified background color
html_content = f"""
<div style="
background-color: rgba({r}, {g}, {b}, {a});
padding: 10px;
display: inline-block;
border-radius: 5px;">
<strong>{formatted_name}</strong>
</div>
"""
return html_content
# Function to check optimization success
def check_optimization_success(
channel_list,
input_channels_spends,
output_channels_spends,
bounds_dict,
optimization_goal,
modified_total_metrics,
actual_total_metrics,
modified_total_spends,
actual_total_spends,
original_total_spends,
optimization_success,
):
for channel in channel_list:
input_channel_spends = input_channels_spends[channel]
output_channel_spends = output_channels_spends[channel]
lower_percent = bounds_dict[channel][0]
upper_percent = bounds_dict[channel][1]
lower_allowed_value = (
input_channel_spends * (100 + lower_percent - 1) / 100
) # 1% Tolerance
upper_allowed_value = (
input_channel_spends * (100 + upper_percent + 1) / 100
) # 1% Tolerance
# Check if output spends are within allowed bounds
if (
output_channel_spends > upper_allowed_value
or output_channel_spends < lower_allowed_value
):
error_message = "Optimization failed: strict bounds. Use flexible bounds."
return False, error_message, "❌"
# Check optimization goal and percent change
if optimization_goal == "Spends":
percent_change_happened = abs(
(modified_total_spends - actual_total_spends) / actual_total_spends
)
if percent_change_happened > 0.01: # Greater than 1% Tolerance
error_message = "Optimization failed: input and optimized spends differ. Use flexible bounds."
return False, error_message, "❌"
else:
percent_change_happened = abs(
(modified_total_metrics - actual_total_metrics) / actual_total_metrics
)
if percent_change_happened > 0.01: # Greater than 1% Tolerance
error_message = "Optimization failed: input and optimized metrics differ. Use flexible bounds."
return False, error_message, "❌"
# Define the allowable range for new spends
lower_limit = original_total_spends * 0.5
upper_limit = original_total_spends * 1.5
# Check if the new spends are within the allowed range
if modified_total_spends < lower_limit or modified_total_spends > upper_limit:
error_message = "New spends optimized are outside the allowed range of ±50%."
return False, error_message, "❌"
# Check if the optimization failed to converge
if not optimization_success:
error_message = "Optimization failed to converge."
return False, error_message, "❌"
return True, "Optimization successful.", "💸"
# Function to check if the optimization target is achievable within the given bounds
@st.cache_data(show_spinner=False)
def check_target_achievability(
optimize_allow,
fixed_target,
lower_achievable_target,
upper_achievable_target,
total_absolute_target,
):
# Format the messages with appropriate numerization and naming
minimum_achievable_message = f"Minimum achievable {fixed_target} with the given spends and bounds is {numerize(lower_achievable_target)}"
maximum_achievable_message = f"Maximum achievable {fixed_target} with the given spends and bounds is {numerize(upper_achievable_target)}"
# Check if the target is within achievable bounds
if (lower_achievable_target > total_absolute_target) or (
upper_achievable_target < total_absolute_target
):
if lower_achievable_target > total_absolute_target:
# Update session state with the minimum achievable error message
st.session_state.message_display = {
"type": "error",
"message": minimum_achievable_message,
"icon": "🔼",
}
else:
# Update session state with the maximum achievable error message
st.session_state.message_display = {
"type": "error",
"message": maximum_achievable_message,
"icon": "🔽",
}
optimize_allow = False
else:
# Reset message display if previous message matches the current scenario
if st.session_state.message_display["message"] in [
minimum_achievable_message,
maximum_achievable_message,
]:
st.session_state.message_display = {
"type": "success",
"message": None,
"icon": "",
}
return optimize_allow
# Function to display a message with the appropriate type and icon
def display_message():
# Retrieve the message details from the session state
message_type = st.session_state.message_display["type"]
message = st.session_state.message_display["message"]
icon = st.session_state.message_display["icon"]
# Display the message if it exists
if message is not None:
if message_type == "success":
st.success(message, icon=icon)
elif message_type == "warning":
st.warning(message, icon=icon)
elif message_type == "error":
st.error(message, icon=icon)
else:
st.info(message, icon=icon)
# Styling
load_local_css("styles.css")
set_header()
# Create project_dct
if "project_dct" not in st.session_state:
project_selection()
st.stop()
database_file = r"DB\User.db"
conn = sqlite3.connect(
database_file, check_same_thread=False
) # connection with sql db
c = conn.cursor()
# Display project info
col_project_data = st.columns([2, 1])
with col_project_data[0]:
st.markdown(f"**Welcome {st.session_state['username']}**")
with col_project_data[1]:
st.markdown(f"**Current Project: {st.session_state['project_name']}**")
# Page Title
st.title("Scenario Planner")
# Define the directory where the metrics data is located
directory = os.path.join(st.session_state["project_path"], "metrics_level_data")
# Retrieve the list of all metric names from the specified directory
metrics_list = get_metrics_names(directory)
# Check if there are any metrics available in the metrics list
if len(metrics_list) == 0:
# Display a warning message to the user if no metrics are found
st.warning(
"Please tune at least one model to generate response curves data.",
icon="⚠️",
)
# Stop further execution as there is no data to process
st.stop()
# Widget columns
metric_col, panel_col = st.columns(2)
# Metrics Selection
metrics_selected = metric_col.selectbox(
"Response Metrics",
sorted(metrics_list),
format_func=name_formating,
key="response_metrics_selectbox_sp",
index=0,
)
metrics_selected_formatted = name_formating(metrics_selected)
# Retrieve the list of all panel names for specified Metrics
file_selected = f"metrics_level_data/data_test_overview_panel@#{metrics_selected}.xlsx"
file_selected_path = os.path.join(st.session_state["project_path"], file_selected)
panel_list = get_panels_names(file_selected_path)
# Panel Selection
panel_selected = panel_col.selectbox(
"Panel",
sorted(panel_list),
key="panel_selected_selectbox_sp",
index=0,
)
panel_selected_formatted = name_formating(panel_selected)
# Define the path to the JSON files
original_json_file_path = os.path.join(
st.session_state["project_path"], "rcs_data_original.json"
)
modified_json_file_path = os.path.join(
st.session_state["project_path"], "rcs_data_modified.json"
)
# Check if the RCS JSON file does not exist
if not os.path.exists(original_json_file_path) or not os.path.exists(
modified_json_file_path
):
print(
f"RCS JSON file does not exist at {original_json_file_path}. Generating new RCS data..."
)
generate_rcs_data(original_json_file_path, modified_json_file_path)
else:
print(
f"RCS JSON file already exists at {original_json_file_path}. No need to generate new RCS data."
)
# Load JSON files if they exist
original_json_data, modified_json_data = load_json_files(
original_json_file_path, modified_json_file_path
)
# Define the path to the pickle files
original_pickle_file_path = os.path.join(
st.session_state["project_path"], "scenario_data_original.pkl"
)
modified_pickle_file_path = os.path.join(
st.session_state["project_path"], "scenario_data_modified.pkl"
)
# Check if the scenario pickle file does not exist
if not os.path.exists(original_pickle_file_path) or not os.path.exists(
modified_pickle_file_path
):
print(
f"Scenario file does not exist at {original_pickle_file_path}. Generating new senario file data..."
)
generate_scenario_data(original_pickle_file_path, modified_pickle_file_path)
else:
print(
f"Scenario file already exists at {original_pickle_file_path}. No need to generate new senario file data."
)
# Load pickle files if they exist
original_data, modified_data = load_pickle_files(
original_pickle_file_path, modified_pickle_file_path
)
# Extract original scenario data for the selected metric and panel
original_scenario_data = original_data[metrics_selected][panel_selected]
# Extract modified scenario data for the same metric and panel
modified_scenario_data = modified_data[metrics_selected][panel_selected]
# Display Actual Vs Optimized
st.divider()
(
actual_spends_col,
actual_metrics_col,
actual_CPA_col,
optimized_spends_col,
optimized_metrics_col,
optimized_CPA_col,
) = st.columns(6)
# Extracting and formatting values
actual_spends = numerize(original_scenario_data["actual_total_spends"])
actual_metric_value = numerize(original_scenario_data["actual_total_sales"])
optimized_spends = numerize(modified_scenario_data["modified_total_spends"])
optimized_metric_value = numerize(modified_scenario_data["modified_total_sales"])
# Calculate the deltas (differences)
spends_delta = numerize(
modified_scenario_data["modified_total_spends"]
- original_scenario_data["actual_total_spends"]
)
metrics_delta = numerize(
modified_scenario_data["modified_total_sales"]
- original_scenario_data["actual_total_sales"]
)
# Display current and optimized CPA
actual_CPA = (
original_scenario_data["actual_total_spends"]
/ original_scenario_data["actual_total_sales"]
)
optimized_CPA = (
modified_scenario_data["modified_total_spends"]
/ modified_scenario_data["modified_total_sales"]
)
CPA_delta = round_value(optimized_CPA - actual_CPA)
actual_CPA_col.metric("Actual CPA", round_value(actual_CPA))
optimized_spends_col.metric("Optimized Spends", optimized_spends, delta=spends_delta)
optimized_metrics_col.metric(
f"Optimized {metrics_selected_formatted}",
optimized_metric_value,
delta=metrics_delta,
)
optimized_CPA_col.metric(
"Optimized CPA",
round_value(optimized_CPA),
delta=CPA_delta,
delta_color="inverse",
)
# Displaying metrics in the columns
actual_spends_col.metric("Actual Spends", actual_spends)
actual_metrics_col.metric(f"Actual {metrics_selected_formatted}", actual_metric_value)
st.divider()
# Calculate ROI threshold
st.session_state.roi_threshold = (
original_scenario_data["actual_total_sales"]
/ original_scenario_data["actual_total_spends"]
)
# Retrieve the list of all channels names for specified Metrics and Panel
channel_list = list(original_scenario_data["channels"].keys())
# Create columns for optimization goal and buttons
optimization_goal_col, message_display_col, button_col = st.columns([3, 6, 6])
# Create columns for absolute text, slider, percentage number and bound type
absolute_text_col, absolute_slider_col, percentage_number_col, bound_type_col = (
st.columns([2, 4, 2, 2])
)
# Dropdown for selecting optimization goal
optimization_goal = optimization_goal_col.selectbox(
"Fix", ["Spends", metrics_selected_formatted]
)
# Button columns with padding for alignment
with button_col:
st.write("##") # Padding
optimize_button_col, reset_button_col = st.columns(2)
reset_button_col.button(
"Reset",
use_container_width=True,
on_click=reset_scenario,
args=(metrics_selected, panel_selected),
)
# Absolute value display
if optimization_goal == "Spends":
absolute_value = modified_scenario_data["actual_total_spends"]
st.session_state.total_absolute_main_key = numerize(
modified_scenario_data["modified_total_spends"]
)
else:
absolute_value = modified_scenario_data["actual_total_sales"]
st.session_state.total_absolute_main_key = numerize(
modified_scenario_data["modified_total_sales"]
)
total_absolute = absolute_text_col.text_input(
"Absolute",
key="total_absolute_main_key",
on_change=total_absolute_main_key_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
optimization_goal,
),
)
# Generate and process slider options
slider_options = list(
np.linspace(int(0.5 * absolute_value), int(1.5 * absolute_value), 50)
) # Generate range
slider_options.append(
modified_scenario_data["modified_total_spends"]
if optimization_goal == "Spends"
else modified_scenario_data["modified_total_sales"]
)
slider_options = sorted(slider_options) # Sort the list
numerized_slider_options = [
numerize(value) for value in slider_options
] # Numerize each value
# Slider for adjusting absolute value within a range
st.session_state.total_absolute_key = numerize(
modified_scenario_data["modified_total_spends"]
if optimization_goal == "Spends"
else modified_scenario_data["modified_total_sales"]
)
slider_value = absolute_slider_col.select_slider(
"Absolute",
numerized_slider_options,
key="total_absolute_key",
on_change=total_absolute_key_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
optimization_goal,
),
)
# Number input for percentage value
if optimization_goal == "Spends":
st.session_state.total_percentage_key = int(
round(
(
(
modified_scenario_data["modified_total_spends"]
- modified_scenario_data["actual_total_spends"]
)
/ modified_scenario_data["actual_total_spends"]
)
* 100,
0,
)
)
else:
st.session_state.total_percentage_key = int(
round(
(
(
modified_scenario_data["modified_total_sales"]
- modified_scenario_data["actual_total_sales"]
)
/ modified_scenario_data["actual_total_sales"]
)
* 100,
0,
)
)
percentage_target = percentage_number_col.number_input(
"Percentage",
min_value=-50,
max_value=50,
key="total_percentage_key",
on_change=total_percentage_key_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
absolute_value,
optimization_goal,
),
)
# Toggle input for bound type
st.session_state["bound_type_key"] = modified_scenario_data["bound_type"]
with bound_type_col:
st.write("##") # Padding
bound_type = st.toggle(
"Apply Custom Bounds",
on_change=bound_type_change,
args=(modified_pickle_file_path,),
key="bound_type_key",
)
# Collect inputs from the user interface
total_channel_spends, optimize_allow = 0, True
bounds_dict = {}
s_curve_params = {}
channels_spends = {}
channels_proportion = {}
channels_conversion_ratio = {}
channels_name_plot_placeholder = {}
# Optimization Inputs UI
with st.expander("Optimization Inputs", expanded=True):
for channel in channel_list:
st.divider()
# Channel key
channel_key = f"{metrics_selected}_{panel_selected}_{channel}"
# Create columns
if st.session_state["bound_type_key"]:
(
name_plot_col,
input_col,
spends_col,
metrics_col,
bounds_input_col,
bounds_display_col,
allow_col,
) = st.columns([2, 1, 1, 1, 1, 1, 1])
else:
(
name_plot_col,
input_col,
spends_col,
metrics_col,
bounds_display_col,
allow_col,
) = st.columns([2, 1, 1.5, 1.5, 1, 1])
bounds_input_col = st.empty()
# Display channel name and ROI/MROI plot
with name_plot_col:
# Placeholder for channel name
channel_name_placeholder = st.empty()
channel_name_placeholder.markdown(
display_channel_name_with_background_color(channel),
unsafe_allow_html=True,
)
# Placeholder for ROI and MROI plot
channel_plot_placeholder = st.container()
# Store placeholder for channel name and ROI/MROI plots
channels_name_plot_placeholder[channel] = {
"channel_name_placeholder": channel_name_placeholder,
"channel_plot_placeholder": channel_plot_placeholder,
}
# Channel spends and sales
channel_spends_actual = (
original_scenario_data["channels"][channel]["actual_total_spends"]
* original_scenario_data["channels"][channel]["conversion_rate"]
)
channel_metrics_actual = original_scenario_data["channels"][channel][
"modified_total_sales"
]
channel_spends_modified = (
modified_scenario_data["channels"][channel]["modified_total_spends"]
* original_scenario_data["channels"][channel]["conversion_rate"]
)
channel_metrics_modified = modified_scenario_data["channels"][channel][
"modified_total_sales"
]
# Channel spends input
with input_col:
# Absolute Spends Input
st.session_state[f"{channel_key}_abs_spends_key"] = numerize(
modified_scenario_data["channels"][channel]["modified_total_spends"]
* original_scenario_data["channels"][channel]["conversion_rate"]
)
absolute_channel_spends = st.text_input(
"Absolute Spends",
key=f"{channel_key}_abs_spends_key",
on_change=absolute_channel_spends_change,
args=(
channel_key,
channel_spends_actual,
channel,
metrics_selected,
panel_selected,
modified_pickle_file_path,
),
)
# Update Percentage Spends Input
st.session_state[f"{channel_key}_per_spends_key"] = int(
round(
(
(
convert_to_float(
st.session_state[f"{channel_key}_abs_spends_key"]
)
- float(channel_spends_actual)
)
/ channel_spends_actual
)
* 100,
0,
)
)
# Percentage Spends Input
percentage_channel_spends = st.number_input(
"Percentage Spends",
min_value=-1000,
max_value=1000,
key=f"{channel_key}_per_spends_key",
on_change=percentage_channel_spends_change,
args=(
channel_key,
channel_spends_actual,
channel,
metrics_selected,
panel_selected,
modified_pickle_file_path,
),
)
# Store channel spends, conversion ratio and proportion list
channels_spends[channel] = original_scenario_data["channels"][channel][
"actual_total_spends"
] * (1 + percentage_channel_spends / 100)
channels_conversion_ratio[channel] = original_scenario_data["channels"][
channel
]["conversion_rate"]
channels_proportion[channel] = original_scenario_data["channels"][channel][
"spends"
] / sum(original_scenario_data["channels"][channel]["spends"])
# Channel metrics display
with metrics_col:
# Absolute Metrics
st.metric(
f"Actual {name_formating(metrics_selected)}",
value=numerize(channel_metrics_actual),
)
# Optimized Metrics
st.metric(
f"Optimized {name_formating(metrics_selected)}",
value=numerize(channel_metrics_modified),
delta=numerize(channel_metrics_modified - channel_metrics_actual),
)
# Channel spends display
with spends_col:
# Absolute Spends
st.metric(
"Actual Spends",
value=numerize(channel_spends_actual),
)
# Optimized Spends
st.metric(
"Optimized Spends",
value=numerize(channel_spends_modified),
delta=numerize(channel_spends_modified - channel_spends_actual),
)
# Channel allows optimize
with allow_col:
# Allow Optimize (Freeze)
st.write("#") # Padding
st.session_state[f"{channel_key}_allow_optimize_key"] = (
modified_scenario_data["channels"][channel]["freeze"]
)
freeze = st.checkbox(
"Freeze",
key=f"{channel_key}_allow_optimize_key",
on_change=freeze_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
channel_key,
channel,
),
)
# If channel is frozen, set bounds to keep the spend unchanged
if freeze:
lower_bound, upper_bound = 0, 0 # Freeze the spend at current level
# Channel bounds input
if st.session_state["bound_type_key"]:
with bounds_input_col:
# Channel upper bound
st.session_state[f"{channel_key}_upper_key"] = (
modified_scenario_data["channels"][channel]["bounds"]
)[1]
upper_bound = st.number_input(
"Upper bound (%)",
min_value=-100,
max_value=100,
key=f"{channel_key}_upper_key",
disabled=st.session_state[f"{channel_key}_allow_optimize_key"],
on_change=bound_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
channel_key,
channel,
),
)
# Channel lower bound
st.session_state[f"{channel_key}_lower_key"] = (
modified_scenario_data["channels"][channel]["bounds"]
)[0]
lower_bound = st.number_input(
"Lower bound (%)",
min_value=-100,
max_value=100,
key=f"{channel_key}_lower_key",
disabled=st.session_state[f"{channel_key}_allow_optimize_key"],
on_change=bound_change,
args=(
metrics_selected,
panel_selected,
modified_pickle_file_path,
channel_key,
channel,
),
)
# Check if lower bound is greater than upper bound
if lower_bound > upper_bound:
lower_bound = -10 # Default lower bound
upper_bound = 10 # Default upper bound
# Store bounds
bounds_dict[channel] = [lower_bound, upper_bound]
else:
# If channel is frozen, set bounds to keep the spend unchanged
if freeze:
lower_bound, upper_bound = 0, 0 # Freeze the spend at current level
else:
lower_bound = -10 # Default lower bound
upper_bound = 10 # Default upper bound
# Store bounds
bounds_dict[channel] = modified_scenario_data["channels"][channel]["bounds"]
# Display the bounds for each channel's spend in the bounds_display_col
with bounds_display_col:
# Retrieve the actual spends for the channel from the original scenario data
actual_spends = (
modified_scenario_data["channels"][channel]["modified_total_spends"]
* modified_scenario_data["channels"][channel]["conversion_rate"]
)
# Calculate the limit for spends
upper_limit_spends = actual_spends * (1 + upper_bound / 100)
lower_limit_spends = actual_spends * (1 + lower_bound / 100)
# Display the upper limit spends
st.metric("Upper Bound", numerize(upper_limit_spends))
st.metric("Lower Bound", numerize(lower_limit_spends))
# Store S-curve parameters
s_curve_params[channel] = get_s_curve_params(
metrics_selected,
panel_selected,
channel,
original_json_data,
modified_json_data,
modified_pickle_file_path,
)
# Total channel spends
total_channel_spends += convert_to_float(
st.session_state[f"{channel_key}_abs_spends_key"]
)
# Check if total channel spends are within the allowed range (±50% of the original total spends)
if (
total_channel_spends > 1.5 * original_scenario_data["actual_total_spends"]
or total_channel_spends < 0.5 * original_scenario_data["actual_total_spends"]
):
# Store the message details in session state
st.session_state.message_display = {
"type": "warning",
"message": "Keep total spending within ±50% of the original value.",
"icon": "⚠️",
}
if optimization_goal == "Spends":
# Get maximum achievable spends
lower_achievable_target, upper_achievable_target = 0, 0
for channel in channel_list:
channel_spends_actual = (
channels_spends[channel] * channels_conversion_ratio[channel]
)
lower_achievable_target += channel_spends_actual * (
1 + bounds_dict[channel][0] / 100
)
upper_achievable_target += channel_spends_actual * (
1 + bounds_dict[channel][1] / 100
)
else:
# Get maximum achievable target metric
lower_achievable_target, upper_achievable_target = max_target_achievable(
channels_spends,
s_curve_params,
channels_proportion,
modified_scenario_data,
bounds_dict,
)
# Total target of selected metric
total_absolute_target = convert_to_float(total_absolute)
# Check if the target is achievable within the specified bounds
if optimize_allow:
optimize_allow = check_target_achievability(
optimize_allow,
name_formating(optimization_goal),
lower_achievable_target,
upper_achievable_target,
total_absolute_target,
)
# Perform the optimization
if optimize_button_col.button(
"Optimize", use_container_width=True, disabled=not optimize_allow
):
with message_display_col:
st.write("##") # Padding
with st.spinner("Optimizing..."):
# Call the optimizer function to get optimized spends
optimized_spends, optimization_success = optimizer(
optimization_goal,
s_curve_params,
channels_spends,
channels_proportion,
channels_conversion_ratio,
convert_to_float(total_absolute),
bounds_dict,
modified_scenario_data,
)
# Initialize dictionaries to store input and output channel spends
input_channels_spends, output_channels_spends = {}, {}
for channel in channel_list:
# Calculate input channel spends by converting spends using conversion ratio
input_channels_spends[channel] = (
channels_spends[channel] * channels_conversion_ratio[channel]
)
# Calculate output channel spends by converting optimized spends using conversion ratio
output_channels_spends[channel] = (
optimized_spends[channel] * channels_conversion_ratio[channel]
)
# Calculate total actual and modified spends
actual_total_spends = sum(list(input_channels_spends.values()))
modified_total_spends = sum(list(output_channels_spends.values()))
# Retrieve the actual total metrics from modified scenario data
actual_total_metrics = modified_scenario_data["modified_total_sales"]
modified_total_metrics = 0 # Initialize modified total metrics
modified_channels_metrics = {}
# Calculate modified metrics for each channel
for channel in optimized_spends.keys():
channel_s_curve_params = s_curve_params[channel]
spend_proportion = (
optimized_spends[channel] * channels_proportion[channel]
)
# Calculate the metrics using the S-curve function
modified_channels_metrics[channel] = sum(
s_curve(
spend_proportion,
channel_s_curve_params["power"],
channel_s_curve_params["K"],
channel_s_curve_params["b"],
channel_s_curve_params["a"],
channel_s_curve_params["x0"],
)
)
modified_total_metrics += modified_channels_metrics[
channel
] # Add channel metrics to total metrics
# Add the constant term to the modified total metrics
modified_total_metrics += sum(modified_scenario_data["constant"])
# Retrieve the original total spends from modified scenario data
original_total_spends = modified_scenario_data["actual_total_spends"]
# Check the success of the optimization process
success, message, icon = check_optimization_success(
channel_list,
input_channels_spends,
output_channels_spends,
bounds_dict,
optimization_goal,
modified_total_metrics,
actual_total_metrics,
modified_total_spends,
actual_total_spends,
original_total_spends,
optimization_success,
)
# Store the message details in session state
st.session_state.message_display = {
"type": "success" if success else "error",
"message": message,
"icon": icon,
}
# Update data only if the optimization is successful
if success:
# Update the modified spend and metrics for each channel in the scenario data
for channel in channel_list:
modified_scenario_data["channels"][channel][
"modified_total_spends"
] = optimized_spends[channel]
# Update the modified metrics for each channel in the scenario data
modified_scenario_data["channels"][channel][
"modified_total_sales"
] = modified_channels_metrics[channel]
# Update the total modified spends in the scenario data
modified_scenario_data["modified_total_spends"] = modified_total_spends
# Update the total modified metrics in the scenario data
modified_scenario_data["modified_total_sales"] = modified_total_metrics
# Save the updated modified_scenario_data back to the pickle file
try:
with open(modified_pickle_file_path, "rb+") as pickle_file:
# Load existing data to ensure we don't overwrite other data
data = pickle.load(pickle_file)
# Update the specific section with the modified scenario data
data[metrics_selected][panel_selected] = modified_scenario_data
# Go to the beginning of the file to overwrite it
pickle_file.seek(0)
pickle.dump(data, pickle_file)
except:
st.toast(
"Failed to Load/Update. Tool reset to default settings.",
icon="⚠️",
)
# Rerun to update values
st.rerun()
########################################## Response Curve ##########################################
# Generate plots
figures, channel_roi_mroi, region_start_end = generate_response_curve_plots(
channel_list, s_curve_params, channels_proportion, original_scenario_data
)
# Display Response Curve in Streamlit with 4 plots per row
st.subheader(f"Response Curve (X: Spends Vs Y: {metrics_selected_formatted})")
with st.expander("Response Curve", expanded=True):
cols = st.columns(4) # Create 4 columns for the first row
for i, fig in enumerate(figures):
col = cols[i % 4] # Rotate through the columns
with col:
# Get channel parameters
channel = channel_list[i]
modified_total_spends = modified_scenario_data["channels"][channel][
"modified_total_spends"
]
conversion_rate = modified_scenario_data["channels"][channel][
"conversion_rate"
]
# Updated figure with modified metrics point
roi_optimized, mroi_optimized, fig_updated = modified_metrics_point(
fig,
modified_total_spends,
s_curve_params[channel],
channels_proportion[channel],
conversion_rate,
)
# Store data of each channel ROI and MROI
channel_roi_mroi[channel]["optimized_roi"] = roi_optimized
channel_roi_mroi[channel]["optimized_mroi"] = mroi_optimized
st.plotly_chart(fig_updated, use_container_width=True)
# Start a new row after every 4 plots
if (i + 1) % 4 == 0 and i + 1 < len(figures):
cols = st.columns(4) # Create new row with 4 columns
# Generate the plots
channel_roi_mroi_plot = roi_mori_plot(channel_roi_mroi)
# Display the plots and name with background color
for channel in channel_list:
with channels_name_plot_placeholder[channel]["channel_plot_placeholder"]:
# Create subplots with 2 columns for ROI and MROI
roi_plot_col, mroi_plot_col = st.columns(2)
# Display ROI and MROI plots
roi_plot_col.plotly_chart(channel_roi_mroi_plot[channel]["fig_roi"])
mroi_plot_col.plotly_chart(channel_roi_mroi_plot[channel]["fig_mroi"])
# Placeholder for the channel name
channel_name_placeholder = channels_name_plot_placeholder[channel][
"channel_name_placeholder"
]
# Retrieve modified total spends and conversion rate for the channel
modified_total_spends = modified_scenario_data["channels"][channel][
"modified_total_spends"
]
conversion_rate = modified_scenario_data["channels"][channel]["conversion_rate"]
# Calculate the actual spend value for the channel
channel_spends_value = modified_total_spends * conversion_rate
# Calculate the RGBA color value for the channel based on its spend
channel_rgba_value = calculate_rgba(channel_spends_value, region_start_end[channel])
# Display the channel name with the calculated background color
channel_name_placeholder.markdown(
display_channel_name_with_background_color(channel, channel_rgba_value),
unsafe_allow_html=True,
)
# Input field for the scenario name
st.text_input("Scenario Name", key="scenario_name")
# Disable the "Save Scenario" button until a name is provided
if st.session_state["scenario_name"] is None or st.session_state["scenario_name"] == "":
save_scenario_button_disabled = True
else:
save_scenario_button_disabled = False
# Button to save the scenario
st.button(
"Save Scenario",
on_click=save_scenario,
args=(
modified_scenario_data,
metrics_selected,
panel_selected,
optimization_goal,
channel_roi_mroi,
),
disabled=save_scenario_button_disabled,
)
########################################## Display Message ##########################################
# Display all message
with message_display_col:
st.write("###") # Padding
display_message()
# Reset the message details in session state
st.session_state.message_display = {
"type": "success",
"message": None,
"icon": "",
}