Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
import json
|
3 |
import torch
|
@@ -49,6 +60,18 @@ device = None
|
|
49 |
|
50 |
@asynccontextmanager
|
51 |
async def lifespan(app: FastAPI):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
# Load resources on startup
|
53 |
global model, explainer, processor, device
|
54 |
|
@@ -125,6 +148,12 @@ app.add_middleware(
|
|
125 |
|
126 |
@app.get("/")
|
127 |
async def root():
|
|
|
|
|
|
|
|
|
|
|
|
|
128 |
return {
|
129 |
"message": "Kickstarter Success Prediction API",
|
130 |
"description": "Send a POST request to /predict with campaign data to get a prediction"
|
@@ -132,6 +161,24 @@ async def root():
|
|
132 |
|
133 |
@app.post("/predict")
|
134 |
async def predict(request: Request):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
try:
|
136 |
# Parse the incoming JSON data
|
137 |
logger.info("Received prediction request")
|
@@ -193,28 +240,114 @@ async def predict(request: Request):
|
|
193 |
raise HTTPException(status_code=500, detail=f"Prediction error: {str(e)}")
|
194 |
|
195 |
def preprocess_raw_data(campaign_data):
|
196 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
197 |
try:
|
198 |
# Process the single campaign
|
199 |
logger.info("Processing campaign with CampaignProcessor...")
|
200 |
-
processed_data = processor.process_campaign(campaign_data, idx=0)
|
201 |
|
202 |
-
#
|
203 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
204 |
if field in campaign_data:
|
205 |
processed_data[field] = campaign_data[field]
|
206 |
logger.info(f"Using provided value for {field}: {campaign_data[field]}")
|
207 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
208 |
return processed_data
|
209 |
|
210 |
except Exception as e:
|
211 |
logger.error(f"Error preprocessing raw data: {str(e)}", exc_info=True)
|
212 |
raise Exception(f"Error preprocessing raw data: {str(e)}")
|
213 |
|
214 |
-
# Debugging endpoint to check the environment and loaded resources
|
215 |
@app.get("/debug")
|
216 |
async def debug():
|
217 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
218 |
global model, explainer, processor, device
|
219 |
|
220 |
# Check internet connectivity
|
|
|
1 |
+
"""
|
2 |
+
Kickstarter Success Prediction API
|
3 |
+
|
4 |
+
This module serves as the main FastAPI application for the Kickstarter Success Prediction service.
|
5 |
+
It provides endpoints for predicting the success probability of Kickstarter campaigns and
|
6 |
+
includes the Longformer embedding in the response for further analysis.
|
7 |
+
|
8 |
+
Author: Angus Fung
|
9 |
+
Date: April 2025
|
10 |
+
"""
|
11 |
+
|
12 |
import os
|
13 |
import json
|
14 |
import torch
|
|
|
60 |
|
61 |
@asynccontextmanager
|
62 |
async def lifespan(app: FastAPI):
|
63 |
+
"""
|
64 |
+
Lifecycle manager for the FastAPI application.
|
65 |
+
|
66 |
+
This function handles the startup and shutdown of the application,
|
67 |
+
managing resources like model loading and caching directories.
|
68 |
+
|
69 |
+
Args:
|
70 |
+
app: The FastAPI application instance
|
71 |
+
|
72 |
+
Yields:
|
73 |
+
None: Control is yielded back to the application while it's running
|
74 |
+
"""
|
75 |
# Load resources on startup
|
76 |
global model, explainer, processor, device
|
77 |
|
|
|
148 |
|
149 |
@app.get("/")
|
150 |
async def root():
|
151 |
+
"""
|
152 |
+
Root endpoint providing API information.
|
153 |
+
|
154 |
+
Returns:
|
155 |
+
dict: Basic API information and usage instructions
|
156 |
+
"""
|
157 |
return {
|
158 |
"message": "Kickstarter Success Prediction API",
|
159 |
"description": "Send a POST request to /predict with campaign data to get a prediction"
|
|
|
161 |
|
162 |
@app.post("/predict")
|
163 |
async def predict(request: Request):
|
164 |
+
"""
|
165 |
+
Prediction endpoint for Kickstarter campaign success.
|
166 |
+
|
167 |
+
This endpoint processes campaign data and returns:
|
168 |
+
- Success probability
|
169 |
+
- Predicted outcome (Success/Failure)
|
170 |
+
- SHAP values for feature importance explanation
|
171 |
+
- Longformer embedding of the campaign description
|
172 |
+
|
173 |
+
Args:
|
174 |
+
request: FastAPI request object containing campaign data as JSON
|
175 |
+
|
176 |
+
Returns:
|
177 |
+
JSONResponse: Prediction results and explanations
|
178 |
+
|
179 |
+
Raises:
|
180 |
+
HTTPException: If an error occurs during prediction
|
181 |
+
"""
|
182 |
try:
|
183 |
# Parse the incoming JSON data
|
184 |
logger.info("Received prediction request")
|
|
|
240 |
raise HTTPException(status_code=500, detail=f"Prediction error: {str(e)}")
|
241 |
|
242 |
def preprocess_raw_data(campaign_data):
|
243 |
+
"""
|
244 |
+
Preprocess raw campaign data using CampaignProcessor.
|
245 |
+
|
246 |
+
This function transforms raw text and numerical campaign data into
|
247 |
+
the format required by the prediction model, including:
|
248 |
+
- Text embeddings generation for description, blurb, and risks
|
249 |
+
- Logarithmic transformation of monetary values (funding goals, pledged amounts)
|
250 |
+
- Country name standardization (conversion to ISO alpha-2 codes)
|
251 |
+
- Category and country encoding
|
252 |
+
- Extraction and normalization of numerical features
|
253 |
+
|
254 |
+
Args:
|
255 |
+
campaign_data (dict): Raw campaign data with text and numerical features
|
256 |
+
|
257 |
+
Returns:
|
258 |
+
dict: Processed data with embeddings and normalized numerical features
|
259 |
+
|
260 |
+
Raises:
|
261 |
+
Exception: If preprocessing fails
|
262 |
+
"""
|
263 |
try:
|
264 |
# Process the single campaign
|
265 |
logger.info("Processing campaign with CampaignProcessor...")
|
|
|
266 |
|
267 |
+
# Log country conversion if present
|
268 |
+
if 'raw_country' in campaign_data:
|
269 |
+
country_name = campaign_data.get('raw_country', '')
|
270 |
+
if country_name:
|
271 |
+
logger.info(f"Found country in input data: '{country_name}' (will be converted to ISO alpha-2 code)")
|
272 |
+
|
273 |
+
# Map field names to the expected structure for the processor
|
274 |
+
# Make a deep copy to avoid modifying the original
|
275 |
+
import copy
|
276 |
+
prepared_data = copy.deepcopy(campaign_data)
|
277 |
+
|
278 |
+
# Log input values for debugging
|
279 |
+
logger.info(f"Input previous_projects_count: {prepared_data.get('previous_projects_count', 'N/A')}")
|
280 |
+
logger.info(f"Input previous_success_rate: {prepared_data.get('previous_success_rate', 'N/A')}")
|
281 |
+
logger.info(f"Input previous_pledged: {prepared_data.get('previous_pledged', 'N/A')}")
|
282 |
+
logger.info(f"Input previous_funding_goal: {prepared_data.get('previous_funding_goal', 'N/A')}")
|
283 |
+
|
284 |
+
# Special handling for success rate calculation
|
285 |
+
if 'previous_success_rate' in campaign_data and 'previous_projects_count' in campaign_data:
|
286 |
+
success_rate = float(campaign_data['previous_success_rate'])
|
287 |
+
projects_count = int(campaign_data['previous_projects_count'])
|
288 |
+
# Calculate successful projects from rate and count
|
289 |
+
if projects_count > 0:
|
290 |
+
prepared_data['previous_successful_projects'] = round(success_rate * projects_count)
|
291 |
+
logger.info(f"Calculated previous_successful_projects: {prepared_data['previous_successful_projects']} " +
|
292 |
+
f"from success rate: {success_rate} and count: {projects_count}")
|
293 |
+
|
294 |
+
# Now process the prepared data
|
295 |
+
processed_data = processor.process_campaign(prepared_data, idx=0)
|
296 |
+
|
297 |
+
# SELECTIVE OVERRIDE: Only override non-transformed numeric fields
|
298 |
+
# Fields that should NOT undergo logarithmic transformation
|
299 |
+
non_transformed_fields = [
|
300 |
+
'description_length', 'image_count', 'video_count',
|
301 |
+
'campaign_duration', 'previous_projects_count', 'previous_success_rate'
|
302 |
+
]
|
303 |
+
|
304 |
+
# Fields that SHOULD undergo logarithmic transformation
|
305 |
+
transformed_fields = [
|
306 |
+
'funding_goal', 'previous_funding_goal', 'previous_pledged'
|
307 |
+
]
|
308 |
+
|
309 |
+
# Override only the non-transformed fields if they exist in input
|
310 |
+
for field in non_transformed_fields:
|
311 |
if field in campaign_data:
|
312 |
processed_data[field] = campaign_data[field]
|
313 |
logger.info(f"Using provided value for {field}: {campaign_data[field]}")
|
314 |
|
315 |
+
# For transformed fields, check if the user explicitly wants to bypass transformation
|
316 |
+
for field in transformed_fields:
|
317 |
+
if field in campaign_data and campaign_data.get('bypass_transformation', False):
|
318 |
+
processed_data[field] = campaign_data[field]
|
319 |
+
logger.warning(
|
320 |
+
f"Bypassing logarithmic transformation for {field} as requested. "
|
321 |
+
"This may affect model performance."
|
322 |
+
)
|
323 |
+
elif field in campaign_data:
|
324 |
+
# Log that we're keeping the transformed value
|
325 |
+
logger.info(f"Using logarithmically transformed {field} value for better model performance.")
|
326 |
+
|
327 |
+
# Verify that the previous metrics are set correctly
|
328 |
+
logger.info(f"Final previous_projects_count: {processed_data.get('previous_projects_count', 'N/A')}")
|
329 |
+
logger.info(f"Final previous_success_rate: {processed_data.get('previous_success_rate', 'N/A')}")
|
330 |
+
logger.info(f"Final previous_pledged: {processed_data.get('previous_pledged', 'N/A')}")
|
331 |
+
logger.info(f"Final previous_funding_goal: {processed_data.get('previous_funding_goal', 'N/A')}")
|
332 |
+
|
333 |
+
logger.info("Preprocessing complete with numerical transformations applied")
|
334 |
return processed_data
|
335 |
|
336 |
except Exception as e:
|
337 |
logger.error(f"Error preprocessing raw data: {str(e)}", exc_info=True)
|
338 |
raise Exception(f"Error preprocessing raw data: {str(e)}")
|
339 |
|
|
|
340 |
@app.get("/debug")
|
341 |
async def debug():
|
342 |
+
"""
|
343 |
+
Debug endpoint for checking API status and component health.
|
344 |
+
|
345 |
+
This endpoint provides diagnostic information about the API's status,
|
346 |
+
model loading, connectivity, disk space, and other components.
|
347 |
+
|
348 |
+
Returns:
|
349 |
+
JSONResponse: Comprehensive diagnostic information
|
350 |
+
"""
|
351 |
global model, explainer, processor, device
|
352 |
|
353 |
# Check internet connectivity
|