elizabethmyn's picture
Add demo for Sale forcasting
84548c1
import os
import matplotlib.pyplot as plt
import matplotlib.ticker as ticker
import numpy as np
import pandas as pd
import seaborn as sns
from matplotlib.dates import DateFormatter
# Set up plotting style
plt.style.use("seaborn-v0_8-whitegrid")
sns.set_palette("deep")
plt.rcParams["figure.figsize"] = (14, 8)
plt.rcParams["font.size"] = 12
def visualize_predictions_by_store_item(test_results, output_dir="visualizations"):
"""
Create visualizations of actual vs predicted values for each store-item combination.
Args:
test_results: DataFrame containing test results with columns:
'date', 'store_name', 'item_name', 'sales', 'prediction'
output_dir: Directory to save the visualizations
"""
# Create output directory if it doesn't exist
os.makedirs(output_dir, exist_ok=True)
# Create a time series plot for each store-item combination
store_items = test_results.groupby(["store_name", "item_name"])
# Get total number of combinations for progress tracking
total_combinations = len(store_items)
print(
f"Creating visualizations for {total_combinations} store-item combinations..."
)
# Counter for progress tracking
counter = 0
# For each store-item combination, create a plot
for (store, item), group in store_items:
# Sort by date to ensure proper time series order
group = group.sort_values("date")
# Convert date to datetime if it's not already
if not pd.api.types.is_datetime64_any_dtype(group["date"]):
group["date"] = pd.to_datetime(group["date"])
# Create the plot
fig, ax = plt.subplots(figsize=(14, 6))
# Plot actual and predicted values
ax.plot(
group["date"], group["sales"], "o-", label="Actual", alpha=0.7, linewidth=2
)
ax.plot(
group["date"],
group["prediction"],
"s--",
label="Predicted",
alpha=0.7,
linewidth=2,
)
# Calculate error metrics for this store-item
mae = np.mean(np.abs(group["sales"] - group["prediction"]))
mape = (
np.mean(np.abs((group["sales"] - group["prediction"]) / group["sales"]))
* 100
)
# Add title and labels
ax.set_title(f"Store: {store}, Item: {item}\nMAE: {mae:.2f}, MAPE: {mape:.2f}%")
ax.set_xlabel("Date")
ax.set_ylabel("Sales")
# Format x-axis dates
date_formatter = DateFormatter("%Y-%m-%d")
ax.xaxis.set_major_formatter(date_formatter)
# Rotate date labels for better readability
plt.xticks(rotation=45)
# Add grid for easier reading
ax.grid(True, linestyle="--", alpha=0.7)
# Add legend
ax.legend()
# Adjust layout
plt.tight_layout()
# Save the figure
safe_store = store.replace(" ", "_").replace("/", "_")
safe_item = item.replace(" ", "_").replace("/", "_")
filename = f"{safe_store}_{safe_item}.png"
plt.savefig(os.path.join(output_dir, filename))
# Close the figure to free memory
plt.close(fig)
# Update progress
counter += 1
if counter % 10 == 0:
print(f"Processed {counter}/{total_combinations} combinations")
print(f"All visualizations saved to {output_dir}/")
def visualize_aggregated_predictions(test_results, output_dir="visualizations"):
"""
Create aggregated visualizations of actual vs predicted values by store, item, and date.
Args:
test_results: DataFrame containing test results
output_dir: Directory to save the visualizations
"""
# Create output directory if it doesn't exist
os.makedirs(output_dir, exist_ok=True)
# Ensure date is in datetime format
if not pd.api.types.is_datetime64_any_dtype(test_results["date"]):
test_results["date"] = pd.to_datetime(test_results["date"])
# 1. Aggregate by date
daily_results = (
test_results.groupby("date")
.agg({"sales": "sum", "prediction": "sum"})
.reset_index()
)
# Plot daily aggregated results
fig, ax = plt.subplots(figsize=(14, 6))
ax.plot(
daily_results["date"],
daily_results["sales"],
"o-",
label="Actual",
alpha=0.7,
linewidth=2,
)
ax.plot(
daily_results["date"],
daily_results["prediction"],
"s--",
label="Predicted",
alpha=0.7,
linewidth=2,
)
# Add title and labels
ax.set_title("Total Daily Sales: Actual vs Predicted")
ax.set_xlabel("Date")
ax.set_ylabel("Total Sales")
# Format x-axis dates
date_formatter = DateFormatter("%Y-%m-%d")
ax.xaxis.set_major_formatter(date_formatter)
plt.xticks(rotation=45)
# Add grid and legend
ax.grid(True, linestyle="--", alpha=0.7)
ax.legend()
# Adjust layout and save
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "total_daily_sales.png"))
plt.close(fig)
# 2. Aggregate by store
store_results = (
test_results.groupby(["store_name", "date"])
.agg({"sales": "sum", "prediction": "sum"})
.reset_index()
)
# Plot for each store
stores = store_results["store_name"].unique()
for store in stores:
store_data = store_results[store_results["store_name"] == store]
fig, ax = plt.subplots(figsize=(14, 6))
ax.plot(
store_data["date"],
store_data["sales"],
"o-",
label="Actual",
alpha=0.7,
linewidth=2,
)
ax.plot(
store_data["date"],
store_data["prediction"],
"s--",
label="Predicted",
alpha=0.7,
linewidth=2,
)
# Add title and labels
ax.set_title(f"Store: {store} - Total Daily Sales")
ax.set_xlabel("Date")
ax.set_ylabel("Total Sales")
# Format x-axis dates
ax.xaxis.set_major_formatter(date_formatter)
plt.xticks(rotation=45)
# Add grid and legend
ax.grid(True, linestyle="--", alpha=0.7)
ax.legend()
# Adjust layout and save
plt.tight_layout()
safe_store = store.replace(" ", "_").replace("/", "_")
plt.savefig(os.path.join(output_dir, f"store_{safe_store}_total.png"))
plt.close(fig)
# 3. Aggregate by item
item_results = (
test_results.groupby(["item_name", "date"])
.agg({"sales": "sum", "prediction": "sum"})
.reset_index()
)
# Plot for each item
items = item_results["item_name"].unique()
for item in items:
item_data = item_results[item_results["item_name"] == item]
fig, ax = plt.subplots(figsize=(14, 6))
ax.plot(
item_data["date"],
item_data["sales"],
"o-",
label="Actual",
alpha=0.7,
linewidth=2,
)
ax.plot(
item_data["date"],
item_data["prediction"],
"s--",
label="Predicted",
alpha=0.7,
linewidth=2,
)
# Add title and labels
ax.set_title(f"Item: {item} - Total Daily Sales")
ax.set_xlabel("Date")
ax.set_ylabel("Total Sales")
# Format x-axis dates
ax.xaxis.set_major_formatter(date_formatter)
plt.xticks(rotation=45)
# Add grid and legend
ax.grid(True, linestyle="--", alpha=0.7)
ax.legend()
# Adjust layout and save
plt.tight_layout()
safe_item = item.replace(" ", "_").replace("/", "_")
plt.savefig(os.path.join(output_dir, f"item_{safe_item}_total.png"))
plt.close(fig)
print(f"Aggregated visualizations saved to {output_dir}/")
def create_interactive_dashboard(test_results, output_dir="visualizations"):
"""
Create an interactive HTML dashboard with plots for all store-item combinations.
Requires Plotly and Dash libraries.
Args:
test_results: DataFrame containing test results
output_dir: Directory to save the dashboard
"""
try:
import plotly.express as px
import plotly.graph_objects as go
from plotly.subplots import make_subplots
print("Creating interactive dashboard...")
# Create output directory if it doesn't exist
os.makedirs(output_dir, exist_ok=True)
# Ensure date is in datetime format
if not pd.api.types.is_datetime64_any_dtype(test_results["date"]):
test_results["date"] = pd.to_datetime(test_results["date"])
# Create overall performance figure
daily_results = (
test_results.groupby("date")
.agg({"sales": "sum", "prediction": "sum"})
.reset_index()
)
fig = go.Figure()
fig.add_trace(
go.Scatter(
x=daily_results["date"],
y=daily_results["sales"],
mode="lines+markers",
name="Actual",
line=dict(color="blue"),
)
)
fig.add_trace(
go.Scatter(
x=daily_results["date"],
y=daily_results["prediction"],
mode="lines+markers",
name="Predicted",
line=dict(color="red", dash="dash"),
)
)
fig.update_layout(
title="Overall Sales Performance: Actual vs Predicted",
xaxis_title="Date",
yaxis_title="Total Sales",
legend_title="Series",
height=600,
)
# Save the overall chart as HTML
fig.write_html(os.path.join(output_dir, "overall_performance.html"))
# Create an error heatmap
store_item_error = (
test_results.groupby(["store_name", "item_name"])
.apply(
lambda x: np.mean(np.abs((x["sales"] - x["prediction"]) / x["sales"]))
* 100
)
.reset_index()
)
store_item_error.columns = ["store_name", "item_name", "mape"]
# Pivot the data for the heatmap
heatmap_data = store_item_error.pivot(
index="store_name", columns="item_name", values="mape"
)
# Create heatmap figure
heatmap_fig = px.imshow(
heatmap_data,
labels=dict(x="Item", y="Store", color="MAPE (%)"),
x=heatmap_data.columns,
y=heatmap_data.index,
color_continuous_scale="RdBu_r",
title="Mean Absolute Percentage Error by Store and Item",
)
heatmap_fig.update_layout(height=800, width=1200)
# Save the heatmap as HTML
heatmap_fig.write_html(os.path.join(output_dir, "error_heatmap.html"))
print(f"Interactive dashboard elements saved to {output_dir}/")
except ImportError:
print("Could not create interactive dashboard. Plotly library is required.")
print("Install it with: pip install plotly dash")
def visualize_error_distribution(test_results, output_dir="visualizations"):
"""
Visualize the distribution and patterns of prediction errors.
Args:
test_results: DataFrame containing test results
output_dir: Directory to save the visualizations
"""
# Create output directory if it doesn't exist
os.makedirs(output_dir, exist_ok=True)
# Calculate errors
test_results["error"] = test_results["sales"] - test_results["prediction"]
test_results["abs_error"] = np.abs(test_results["error"])
test_results["pct_error"] = (test_results["error"] / test_results["sales"]) * 100
# 1. Error distribution histogram
plt.figure(figsize=(12, 6))
sns.histplot(test_results["error"], kde=True, bins=50)
plt.axvline(x=0, color="red", linestyle="--")
plt.title("Distribution of Prediction Errors")
plt.xlabel("Error (Actual - Predicted)")
plt.ylabel("Frequency")
plt.grid(True, linestyle="--", alpha=0.7)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "error_distribution.png"))
plt.close()
# 2. Error vs Actual Sales
plt.figure(figsize=(12, 6))
plt.scatter(test_results["sales"], test_results["error"], alpha=0.5)
plt.axhline(y=0, color="red", linestyle="--")
plt.title("Prediction Error vs Actual Sales")
plt.xlabel("Actual Sales")
plt.ylabel("Error (Actual - Predicted)")
plt.grid(True, linestyle="--", alpha=0.7)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "error_vs_sales.png"))
plt.close()
# 3. Error over time
plt.figure(figsize=(14, 6))
# Ensure date is in datetime format
if not pd.api.types.is_datetime64_any_dtype(test_results["date"]):
test_results["date"] = pd.to_datetime(test_results["date"])
# Group by date to see overall error trend
daily_error = test_results.groupby("date")["error"].mean().reset_index()
plt.plot(daily_error["date"], daily_error["error"], "o-")
plt.axhline(y=0, color="red", linestyle="--")
plt.title("Mean Prediction Error Over Time")
plt.xlabel("Date")
plt.ylabel("Mean Error")
date_formatter = DateFormatter("%Y-%m-%d")
plt.gca().xaxis.set_major_formatter(date_formatter)
plt.xticks(rotation=45)
plt.grid(True, linestyle="--", alpha=0.7)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "error_over_time.png"))
plt.close()
# 4. Error by day of week
test_results["day_of_week"] = test_results["date"].dt.dayofweek
test_results["day_name"] = test_results["date"].dt.day_name()
plt.figure(figsize=(12, 6))
day_error = (
test_results.groupby("day_name")["pct_error"]
.mean()
.reindex(
[
"Monday",
"Tuesday",
"Wednesday",
"Thursday",
"Friday",
"Saturday",
"Sunday",
]
)
)
sns.barplot(x=day_error.index, y=day_error.values)
plt.title("Mean Percentage Error by Day of Week")
plt.xlabel("Day of Week")
plt.ylabel("Mean Percentage Error (%)")
plt.axhline(y=0, color="red", linestyle="--")
plt.grid(True, linestyle="--", alpha=0.7)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "error_by_day_of_week.png"))
plt.close()
# 5. Error by category - only if 'category' column exists
if "category" in test_results.columns:
plt.figure(figsize=(12, 6))
cat_error = test_results.groupby("category")["pct_error"].mean().sort_values()
sns.barplot(x=cat_error.index, y=cat_error.values)
plt.title("Mean Percentage Error by Category")
plt.xlabel("Category")
plt.ylabel("Mean Percentage Error (%)")
plt.axhline(y=0, color="red", linestyle="--")
plt.xticks(rotation=45)
plt.grid(True, linestyle="--", alpha=0.7)
plt.tight_layout()
plt.savefig(os.path.join(output_dir, "error_by_category.png"))
plt.close()
print(f"Error analysis visualizations saved to {output_dir}/")
def create_forecast_dashboard(
model, X_test, y_test, test_results, data, output_dir="visualizations"
):
"""
Create a comprehensive dashboard of forecast visualizations.
Args:
model: Trained model
X_test: Test features
y_test: Test target values
test_results: DataFrame with test results
data: Original data with date, store, item info
output_dir: Directory to save visualizations
"""
# Create all visualizations
print("Creating forecast visualizations...")
# 1. Individual store-item visualizations (limited to avoid too many plots)
# Get the top 20 store-item combinations by sales volume
store_item_sales = (
test_results.groupby(["store_name", "item_name"])["sales"].sum().reset_index()
)
top_combinations = store_item_sales.sort_values("sales", ascending=False).head(20)
# Filter test_results to include only these top combinations
top_results = pd.merge(
test_results,
top_combinations[["store_name", "item_name"]],
on=["store_name", "item_name"],
)
# Create visualizations for top combinations
visualize_predictions_by_store_item(top_results, output_dir)
# 2. Aggregated visualizations
visualize_aggregated_predictions(test_results, output_dir)
# 3. Error distribution and patterns
visualize_error_distribution(test_results, output_dir)
# 4. Try to create interactive dashboard if plotly is available
create_interactive_dashboard(test_results, output_dir)
print("Forecast visualization dashboard created successfully!")