Spaces:
Sleeping
Sleeping
trying to add tabs to separate the validation and forecasts (also adding the actual forecasts)
Browse files
app.py
CHANGED
@@ -34,7 +34,7 @@ def load_data(file):
|
|
34 |
return None, f"Error loading data: {str(e)}"
|
35 |
|
36 |
# Function to generate and return a plot
|
37 |
-
def create_forecast_plot(forecast_df, original_df):
|
38 |
plt.figure(figsize=(10, 6))
|
39 |
unique_ids = forecast_df['unique_id'].unique()
|
40 |
forecast_cols = [col for col in forecast_df.columns if col not in ['unique_id', 'ds', 'cutoff']]
|
@@ -47,7 +47,32 @@ def create_forecast_plot(forecast_df, original_df):
|
|
47 |
if col in forecast_data.columns:
|
48 |
plt.plot(forecast_data['ds'], forecast_data[col], label=col)
|
49 |
|
50 |
-
plt.title(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
plt.xlabel('Date')
|
52 |
plt.ylabel('Value')
|
53 |
plt.legend()
|
@@ -72,11 +97,12 @@ def run_forecast(
|
|
72 |
use_seasonal_window_avg,
|
73 |
seasonal_window_size,
|
74 |
use_autoets,
|
75 |
-
use_autoarima
|
|
|
76 |
):
|
77 |
df, message = load_data(file)
|
78 |
if df is None:
|
79 |
-
return None, None, None, message
|
80 |
|
81 |
models = []
|
82 |
model_aliases = []
|
@@ -104,27 +130,33 @@ def run_forecast(
|
|
104 |
model_aliases.append('autoarima')
|
105 |
|
106 |
if not models:
|
107 |
-
return None, None, None, "Please select at least one forecasting model"
|
108 |
|
109 |
sf = StatsForecast(models=models, freq=frequency, n_jobs=-1)
|
110 |
|
111 |
try:
|
|
|
112 |
if eval_strategy == "Cross Validation":
|
113 |
cv_results = sf.cross_validation(df=df, h=horizon, step_size=step_size, n_windows=num_windows)
|
114 |
evaluation = evaluate(df=cv_results, metrics=[bias, mae, rmse, mape], models=model_aliases)
|
115 |
eval_df = pd.DataFrame(evaluation).reset_index()
|
116 |
-
|
117 |
-
return eval_df, cv_results, fig_forecast, "Cross validation completed successfully!"
|
118 |
-
|
119 |
else: # Fixed window
|
120 |
-
cv_results = sf.cross_validation(df=df, h=horizon, step_size=10, n_windows=1)
|
121 |
evaluation = evaluate(df=cv_results, metrics=[bias, mae, rmse, mape], models=model_aliases)
|
122 |
eval_df = pd.DataFrame(evaluation).reset_index()
|
123 |
-
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
|
126 |
except Exception as e:
|
127 |
-
return None, None, None, f"Error during forecasting: {str(e)}"
|
128 |
|
129 |
# Sample CSV file generation
|
130 |
def download_sample():
|
@@ -163,32 +195,49 @@ with gr.Blocks(title="StatsForecast Demo") as app:
|
|
163 |
download_output = gr.File(label="Click to download", visible=True)
|
164 |
download_btn.click(fn=download_sample, outputs=download_output)
|
165 |
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
gr.
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
186 |
|
187 |
with gr.Column(scale=3):
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
192 |
|
193 |
submit_btn.click(
|
194 |
fn=run_forecast,
|
@@ -196,9 +245,9 @@ with gr.Blocks(title="StatsForecast Demo") as app:
|
|
196 |
file_input, frequency, eval_strategy, horizon, step_size, num_windows,
|
197 |
use_historical_avg, use_naive, use_seasonal_naive, seasonality,
|
198 |
use_window_avg, window_size, use_seasonal_window_avg, seasonal_window_size,
|
199 |
-
use_autoets, use_autoarima
|
200 |
],
|
201 |
-
outputs=[eval_output, forecast_output,
|
202 |
)
|
203 |
|
204 |
if __name__ == "__main__":
|
|
|
34 |
return None, f"Error loading data: {str(e)}"
|
35 |
|
36 |
# Function to generate and return a plot
|
37 |
+
def create_forecast_plot(forecast_df, original_df, title="Forecasting Results"):
|
38 |
plt.figure(figsize=(10, 6))
|
39 |
unique_ids = forecast_df['unique_id'].unique()
|
40 |
forecast_cols = [col for col in forecast_df.columns if col not in ['unique_id', 'ds', 'cutoff']]
|
|
|
47 |
if col in forecast_data.columns:
|
48 |
plt.plot(forecast_data['ds'], forecast_data[col], label=col)
|
49 |
|
50 |
+
plt.title(title)
|
51 |
+
plt.xlabel('Date')
|
52 |
+
plt.ylabel('Value')
|
53 |
+
plt.legend()
|
54 |
+
plt.grid(True)
|
55 |
+
fig = plt.gcf()
|
56 |
+
return fig
|
57 |
+
|
58 |
+
# Function to create a plot for future forecasts
|
59 |
+
def create_future_forecast_plot(forecast_df, original_df):
|
60 |
+
plt.figure(figsize=(10, 6))
|
61 |
+
unique_ids = forecast_df['unique_id'].unique()
|
62 |
+
forecast_cols = [col for col in forecast_df.columns if col not in ['unique_id', 'ds']]
|
63 |
+
|
64 |
+
for unique_id in unique_ids:
|
65 |
+
# Plot historical data
|
66 |
+
original_data = original_df[original_df['unique_id'] == unique_id]
|
67 |
+
plt.plot(original_data['ds'], original_data['y'], 'k-', label='Historical')
|
68 |
+
|
69 |
+
# Plot forecast data
|
70 |
+
forecast_data = forecast_df[forecast_df['unique_id'] == unique_id]
|
71 |
+
for col in forecast_cols:
|
72 |
+
if col in forecast_data.columns:
|
73 |
+
plt.plot(forecast_data['ds'], forecast_data[col], label=col)
|
74 |
+
|
75 |
+
plt.title('Future Forecast')
|
76 |
plt.xlabel('Date')
|
77 |
plt.ylabel('Value')
|
78 |
plt.legend()
|
|
|
97 |
use_seasonal_window_avg,
|
98 |
seasonal_window_size,
|
99 |
use_autoets,
|
100 |
+
use_autoarima,
|
101 |
+
future_horizon
|
102 |
):
|
103 |
df, message = load_data(file)
|
104 |
if df is None:
|
105 |
+
return None, None, None, None, None, message
|
106 |
|
107 |
models = []
|
108 |
model_aliases = []
|
|
|
130 |
model_aliases.append('autoarima')
|
131 |
|
132 |
if not models:
|
133 |
+
return None, None, None, None, None, "Please select at least one forecasting model"
|
134 |
|
135 |
sf = StatsForecast(models=models, freq=frequency, n_jobs=-1)
|
136 |
|
137 |
try:
|
138 |
+
# Run cross-validation
|
139 |
if eval_strategy == "Cross Validation":
|
140 |
cv_results = sf.cross_validation(df=df, h=horizon, step_size=step_size, n_windows=num_windows)
|
141 |
evaluation = evaluate(df=cv_results, metrics=[bias, mae, rmse, mape], models=model_aliases)
|
142 |
eval_df = pd.DataFrame(evaluation).reset_index()
|
143 |
+
fig_validation = create_forecast_plot(cv_results, df, "Cross Validation Results")
|
|
|
|
|
144 |
else: # Fixed window
|
145 |
+
cv_results = sf.cross_validation(df=df, h=horizon, step_size=10, n_windows=1) # any step size for 1 window
|
146 |
evaluation = evaluate(df=cv_results, metrics=[bias, mae, rmse, mape], models=model_aliases)
|
147 |
eval_df = pd.DataFrame(evaluation).reset_index()
|
148 |
+
fig_validation = create_forecast_plot(cv_results, df, "Fixed Window Validation Results")
|
149 |
+
|
150 |
+
# Generate future forecasts
|
151 |
+
fitted_sf = StatsForecast(models=models, freq=frequency, n_jobs=-1)
|
152 |
+
fitted_sf.fit(df)
|
153 |
+
future_forecasts = fitted_sf.forecast(h=future_horizon)
|
154 |
+
fig_future = create_future_forecast_plot(future_forecasts, df)
|
155 |
+
|
156 |
+
return eval_df, cv_results, fig_validation, future_forecasts, fig_future, "Analysis completed successfully!"
|
157 |
|
158 |
except Exception as e:
|
159 |
+
return None, None, None, None, None, f"Error during forecasting: {str(e)}"
|
160 |
|
161 |
# Sample CSV file generation
|
162 |
def download_sample():
|
|
|
195 |
download_output = gr.File(label="Click to download", visible=True)
|
196 |
download_btn.click(fn=download_sample, outputs=download_output)
|
197 |
|
198 |
+
with gr.Accordion("Data & Validation Settings", open=True):
|
199 |
+
frequency = gr.Dropdown(choices=["H", "D", "WS", "MS", "QS", "YS"], label="Frequency", value="D")
|
200 |
+
eval_strategy = gr.Radio(choices=["Fixed Window", "Cross Validation"], label="Evaluation Strategy", value="Cross Validation")
|
201 |
+
horizon = gr.Slider(1, 100, value=10, step=1, label="Validation Horizon")
|
202 |
+
step_size = gr.Slider(1, 50, value=10, step=1, label="Step Size")
|
203 |
+
num_windows = gr.Slider(1, 20, value=3, step=1, label="Number of Windows")
|
204 |
+
|
205 |
+
with gr.Accordion("Forecast Settings", open=True):
|
206 |
+
future_horizon = gr.Slider(1, 100, value=20, step=1, label="Future Forecast Horizon")
|
207 |
+
|
208 |
+
with gr.Accordion("Model Configuration", open=True):
|
209 |
+
use_historical_avg = gr.Checkbox(label="Use Historical Average", value=True)
|
210 |
+
use_naive = gr.Checkbox(label="Use Naive", value=True)
|
211 |
+
|
212 |
+
with gr.Row():
|
213 |
+
use_seasonal_naive = gr.Checkbox(label="Use Seasonal Naive")
|
214 |
+
seasonality = gr.Number(label="Seasonality", value=10)
|
215 |
+
|
216 |
+
with gr.Row():
|
217 |
+
use_window_avg = gr.Checkbox(label="Use Window Average")
|
218 |
+
window_size = gr.Number(label="Window Size", value=3)
|
219 |
+
|
220 |
+
with gr.Row():
|
221 |
+
use_seasonal_window_avg = gr.Checkbox(label="Use Seasonal Window Average")
|
222 |
+
seasonal_window_size = gr.Number(label="Seasonal Window Size", value=2)
|
223 |
+
|
224 |
+
use_autoets = gr.Checkbox(label="Use AutoETS")
|
225 |
+
use_autoarima = gr.Checkbox(label="Use AutoARIMA")
|
226 |
+
|
227 |
+
submit_btn = gr.Button("Run Forecast", variant="primary")
|
228 |
|
229 |
with gr.Column(scale=3):
|
230 |
+
message_output = gr.Textbox(label="Status Message")
|
231 |
+
|
232 |
+
with gr.Tabs() as tabs:
|
233 |
+
with gr.TabItem("Validation Results"):
|
234 |
+
eval_output = gr.Dataframe(label="Evaluation Metrics")
|
235 |
+
validation_output = gr.Dataframe(label="Validation Data")
|
236 |
+
validation_plot = gr.Plot(label="Validation Plot")
|
237 |
+
|
238 |
+
with gr.TabItem("Future Forecast"):
|
239 |
+
forecast_output = gr.Dataframe(label="Future Forecast Data")
|
240 |
+
forecast_plot = gr.Plot(label="Future Forecast Plot")
|
241 |
|
242 |
submit_btn.click(
|
243 |
fn=run_forecast,
|
|
|
245 |
file_input, frequency, eval_strategy, horizon, step_size, num_windows,
|
246 |
use_historical_avg, use_naive, use_seasonal_naive, seasonality,
|
247 |
use_window_avg, window_size, use_seasonal_window_avg, seasonal_window_size,
|
248 |
+
use_autoets, use_autoarima, future_horizon
|
249 |
],
|
250 |
+
outputs=[eval_output, validation_output, validation_plot, forecast_output, forecast_plot, message_output]
|
251 |
)
|
252 |
|
253 |
if __name__ == "__main__":
|