Update app.py
Browse filesadd som to app.py
app.py
CHANGED
|
@@ -5,7 +5,7 @@ import torch
|
|
| 5 |
from fastapi import FastAPI, UploadFile, File, HTTPException, BackgroundTasks
|
| 6 |
from fastapi.middleware.cors import CORSMiddleware
|
| 7 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 8 |
-
from typing import List
|
| 9 |
from collections import Counter
|
| 10 |
from fastapi.staticfiles import StaticFiles
|
| 11 |
from huggingface_hub import HfApi,hf_hub_download
|
|
@@ -13,8 +13,19 @@ import datetime
|
|
| 13 |
import csv, sys
|
| 14 |
import io
|
| 15 |
import numpy as np
|
| 16 |
-
|
| 17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
CURRENT_TAG = os.getenv("MODEL_TAG", "main")
|
| 19 |
# ====== FastAPI 初始化 ======
|
| 20 |
app = FastAPI(
|
|
@@ -67,6 +78,36 @@ model.to(DEVICE).eval()
|
|
| 67 |
print("✅ Model loaded successfully")
|
| 68 |
|
| 69 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
# ----------------------------------------------------------
|
| 71 |
# ROUTES
|
| 72 |
# ----------------------------------------------------------
|
|
@@ -74,13 +115,16 @@ print("✅ Model loaded successfully")
|
|
| 74 |
@app.get("/")
|
| 75 |
def home():
|
| 76 |
return {
|
| 77 |
-
"message": "Malware Classifier API (Space A)",
|
| 78 |
"model": MODEL_REPO,
|
| 79 |
"space_b_url": SPACE_B_URL,
|
|
|
|
| 80 |
"endpoints": {
|
| 81 |
-
"/predict": "POST - Upload TXT files for prediction",
|
|
|
|
| 82 |
"/model-info": "GET - Get model information",
|
| 83 |
-
"/
|
|
|
|
| 84 |
}
|
| 85 |
}
|
| 86 |
|
|
@@ -103,7 +147,9 @@ def get_model_info():
|
|
| 103 |
"labels": list(model.config.id2label.values())
|
| 104 |
if hasattr(model.config, 'id2label') else None,
|
| 105 |
"device": str(DEVICE),
|
| 106 |
-
"id2label_sample": id2label_sample
|
|
|
|
|
|
|
| 107 |
}
|
| 108 |
|
| 109 |
|
|
@@ -154,6 +200,125 @@ def upload_embeddings_to_hub(local_path: str):
|
|
| 154 |
return False
|
| 155 |
|
| 156 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 157 |
@app.post("/predict")
|
| 158 |
async def predict_malware(files: List[UploadFile] = File(...)):
|
| 159 |
"""
|
|
@@ -337,6 +502,26 @@ async def predict_malware(files: List[UploadFile] = File(...)):
|
|
| 337 |
except Exception as e:
|
| 338 |
print(f"⚠️ Dataset upload failed: {e}")
|
| 339 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 340 |
return result
|
| 341 |
|
| 342 |
except Exception as e:
|
|
@@ -344,7 +529,23 @@ async def predict_malware(files: List[UploadFile] = File(...)):
|
|
| 344 |
import traceback
|
| 345 |
traceback.print_exc()
|
| 346 |
raise HTTPException(status_code=500, detail=f"Prediction failed: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 347 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
csv.field_size_limit(sys.maxsize)
|
| 349 |
|
| 350 |
JUMP_MNEMONICS = {"jmp", "je", "jne", "jl", "jg", "jle", "jge", "call", "ret"}
|
|
|
|
| 5 |
from fastapi import FastAPI, UploadFile, File, HTTPException, BackgroundTasks
|
| 6 |
from fastapi.middleware.cors import CORSMiddleware
|
| 7 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 8 |
+
from typing import List, Optional, Dict
|
| 9 |
from collections import Counter
|
| 10 |
from fastapi.staticfiles import StaticFiles
|
| 11 |
from huggingface_hub import HfApi,hf_hub_download
|
|
|
|
| 13 |
import csv, sys
|
| 14 |
import io
|
| 15 |
import numpy as np
|
| 16 |
+
from datasets import load_dataset
|
| 17 |
+
|
| 18 |
+
# ===== 新增:SOM 相關 import =====
|
| 19 |
+
from datasets import load_dataset
|
| 20 |
+
|
| 21 |
+
# ⚠️ 注意: som_analyzer.py 必須存在於同一目錄
|
| 22 |
+
try:
|
| 23 |
+
from som_analyzer import SOMAnalyzer
|
| 24 |
+
SOM_FILE_AVAILABLE = True
|
| 25 |
+
except ImportError:
|
| 26 |
+
print("⚠️ WARNING: 'som_analyzer.py' not found. SOM features will be disabled.")
|
| 27 |
+
SOM_FILE_AVAILABLE = False
|
| 28 |
+
|
| 29 |
CURRENT_TAG = os.getenv("MODEL_TAG", "main")
|
| 30 |
# ====== FastAPI 初始化 ======
|
| 31 |
app = FastAPI(
|
|
|
|
| 78 |
print("✅ Model loaded successfully")
|
| 79 |
|
| 80 |
|
| 81 |
+
# ===== 新增:載入 SOM 模型 (from app.py) =====
|
| 82 |
+
print("🧠 Initializing SOM Analyzer...")
|
| 83 |
+
SOM_REPO = "malvec/codebert_Malvec" # (app.py 的 SOM repo)
|
| 84 |
+
SOM_AVAILABLE = False
|
| 85 |
+
|
| 86 |
+
if SOM_FILE_AVAILABLE: # 檢查 'som_analyzer.py' 是否 import 成功
|
| 87 |
+
try:
|
| 88 |
+
som_weights_path = hf_hub_download(
|
| 89 |
+
repo_id=SOM_REPO,
|
| 90 |
+
filename="som_weights.npy",
|
| 91 |
+
token=HF_TOKEN
|
| 92 |
+
)
|
| 93 |
+
background_data_path = hf_hub_download(
|
| 94 |
+
repo_id=SOM_REPO,
|
| 95 |
+
filename="som_background_data.pkl",
|
| 96 |
+
token=HF_TOKEN
|
| 97 |
+
)
|
| 98 |
+
|
| 99 |
+
som_analyzer = SOMAnalyzer(
|
| 100 |
+
som_weights_path=som_weights_path,
|
| 101 |
+
background_data_path=background_data_path
|
| 102 |
+
)
|
| 103 |
+
print("✅ SOM Analyzer initialized successfully!")
|
| 104 |
+
SOM_AVAILABLE = True
|
| 105 |
+
except Exception as e:
|
| 106 |
+
print(f"⚠️ SOM Analyzer initialization failed: {e}")
|
| 107 |
+
print(" SOM analysis will be disabled")
|
| 108 |
+
else:
|
| 109 |
+
print(" SOM analysis disabled because 'som_analyzer.py' is missing.")
|
| 110 |
+
|
| 111 |
# ----------------------------------------------------------
|
| 112 |
# ROUTES
|
| 113 |
# ----------------------------------------------------------
|
|
|
|
| 115 |
@app.get("/")
|
| 116 |
def home():
|
| 117 |
return {
|
| 118 |
+
"message": "Malware Classifier API (Space A) with SOM", # ✨
|
| 119 |
"model": MODEL_REPO,
|
| 120 |
"space_b_url": SPACE_B_URL,
|
| 121 |
+
"som_status": "Available" if SOM_AVAILABLE else "Disabled", # ✨
|
| 122 |
"endpoints": {
|
| 123 |
+
"/predict": "POST - Upload TXT files for prediction (with SOM)",
|
| 124 |
+
"/predict-batch": "POST - Simplified batch prediction (from app.py)",
|
| 125 |
"/model-info": "GET - Get model information",
|
| 126 |
+
"/attention-heatmap": "POST - Generate attention heatmap (with CLIP)",
|
| 127 |
+
"/check-new-samples": "POST - Trigger Space B Active Learning"
|
| 128 |
}
|
| 129 |
}
|
| 130 |
|
|
|
|
| 147 |
"labels": list(model.config.id2label.values())
|
| 148 |
if hasattr(model.config, 'id2label') else None,
|
| 149 |
"device": str(DEVICE),
|
| 150 |
+
"id2label_sample": id2label_sample,
|
| 151 |
+
"som_model_repo": SOM_REPO, # ✨
|
| 152 |
+
"som_available": SOM_AVAILABLE # ✨
|
| 153 |
}
|
| 154 |
|
| 155 |
|
|
|
|
| 200 |
return False
|
| 201 |
|
| 202 |
|
| 203 |
+
# ----------------------------------------------------------
|
| 204 |
+
# 新增:SOM 輔助函數 (from app.py)
|
| 205 |
+
# ----------------------------------------------------------
|
| 206 |
+
|
| 207 |
+
async def get_embeddings_from_hf_dataset(filename: str) -> Optional[List[np.ndarray]]:
|
| 208 |
+
""" (from app.py) """
|
| 209 |
+
try:
|
| 210 |
+
print(f"\n📦 Fetching embeddings from HF dataset for: {filename}")
|
| 211 |
+
# 注意: 這裡使用的是 app.py 中定義的 dataset (lyi029/malvec-embeddings)
|
| 212 |
+
dataset = load_dataset("lyi029/malvec-embeddings", split="train", streaming=True)
|
| 213 |
+
embeddings = []
|
| 214 |
+
count = 0
|
| 215 |
+
for sample in dataset:
|
| 216 |
+
sample_filename = sample.get('source', '') or sample.get('filename', '')
|
| 217 |
+
if filename.lower() in sample_filename.lower():
|
| 218 |
+
emb = sample.get('embedding') or sample.get('emb')
|
| 219 |
+
if emb is not None:
|
| 220 |
+
embeddings.append(np.array(emb, dtype=np.float32))
|
| 221 |
+
count += 1
|
| 222 |
+
if count >= 1000: break
|
| 223 |
+
if embeddings:
|
| 224 |
+
print(f"✅ Found {len(embeddings)} embeddings for {filename}")
|
| 225 |
+
return embeddings
|
| 226 |
+
else:
|
| 227 |
+
print(f"⚠️ No embeddings found for {filename} in dataset")
|
| 228 |
+
return None
|
| 229 |
+
except Exception as e:
|
| 230 |
+
print(f"❌ Error fetching from HF dataset: {e}")
|
| 231 |
+
return None
|
| 232 |
+
|
| 233 |
+
def analyze_som_region_probability(
|
| 234 |
+
som_analyzer: SOMAnalyzer,
|
| 235 |
+
center_row: int,
|
| 236 |
+
center_col: int,
|
| 237 |
+
embeddings: List[np.ndarray],
|
| 238 |
+
radius: int = 2
|
| 239 |
+
) -> Dict[str, float]:
|
| 240 |
+
""" (from app.py) """
|
| 241 |
+
print(f" 🔍 Analyzing region around ({center_row}, {center_col}) with radius={radius}")
|
| 242 |
+
region_samples = []
|
| 243 |
+
for emb in embeddings:
|
| 244 |
+
sample_row, sample_col = som_analyzer.som.winner(emb) #
|
| 245 |
+
if abs(sample_row - center_row) <= radius and abs(sample_col - center_col) <= radius:
|
| 246 |
+
region_samples.append(emb)
|
| 247 |
+
print(f" Found {len(region_samples)} samples in region")
|
| 248 |
+
if len(region_samples) == 0:
|
| 249 |
+
return {"dropper": 0.0, "APT30": 0.0}
|
| 250 |
+
|
| 251 |
+
feature_counts = {"dropper": [], "spreader": [], "APT30": [], "Codoso_Gh0st_1": []}
|
| 252 |
+
if hasattr(som_analyzer, 'background_embeddings') and hasattr(som_analyzer, 'background_labels'):
|
| 253 |
+
for i, bg_emb in enumerate(som_analyzer.background_embeddings):
|
| 254 |
+
bg_row, bg_col = som_analyzer.som.winner(bg_emb) #
|
| 255 |
+
if abs(bg_row - center_row) <= radius and abs(bg_col - center_col) <= radius:
|
| 256 |
+
for feature_name in feature_counts.keys():
|
| 257 |
+
if feature_name in som_analyzer.background_labels:
|
| 258 |
+
feature_counts[feature_name].append(
|
| 259 |
+
som_analyzer.background_labels[feature_name][i]
|
| 260 |
+
)
|
| 261 |
+
probabilities = {}
|
| 262 |
+
for feature_name, values in feature_counts.items():
|
| 263 |
+
if len(values) > 0:
|
| 264 |
+
prob = sum(values) / len(values)
|
| 265 |
+
probabilities[feature_name] = float(prob)
|
| 266 |
+
else:
|
| 267 |
+
probabilities[feature_name] = 0.0
|
| 268 |
+
return probabilities
|
| 269 |
+
|
| 270 |
+
|
| 271 |
+
async def analyze_with_som(embedding, predicted_family, filename):
|
| 272 |
+
""" (from app.py) """
|
| 273 |
+
if not SOM_AVAILABLE:
|
| 274 |
+
print("⚠️ SOM analysis disabled")
|
| 275 |
+
return None
|
| 276 |
+
try:
|
| 277 |
+
print("\n🗺️ Performing SOM analysis...")
|
| 278 |
+
if isinstance(embedding, list):
|
| 279 |
+
embedding_array = np.array(embedding, dtype=np.float32)
|
| 280 |
+
else:
|
| 281 |
+
embedding_array = embedding
|
| 282 |
+
|
| 283 |
+
winner_row, winner_col = som_analyzer.find_winner_neuron(embedding_array)
|
| 284 |
+
print(f" 📍 Winner neuron: ({winner_row}, {winner_col})")
|
| 285 |
+
|
| 286 |
+
hf_embeddings = await get_embeddings_from_hf_dataset(filename)
|
| 287 |
+
|
| 288 |
+
if hf_embeddings and len(hf_embeddings) > 0:
|
| 289 |
+
print(f"✅ Using {len(hf_embeddings)} embeddings from HF dataset")
|
| 290 |
+
features_prob = analyze_som_region_probability( #
|
| 291 |
+
som_analyzer, winner_row, winner_col, hf_embeddings, radius=2
|
| 292 |
+
)
|
| 293 |
+
data_source = "hf_dataset"
|
| 294 |
+
else:
|
| 295 |
+
print(f"⚠️ No HF data found, using background data")
|
| 296 |
+
features_dict = som_analyzer.predict_features_from_position(
|
| 297 |
+
winner_row, winner_col, radius=1
|
| 298 |
+
)
|
| 299 |
+
features_prob = {k: float(v) for k, v in features_dict.items()}
|
| 300 |
+
data_source = "background_data"
|
| 301 |
+
|
| 302 |
+
features_binary = {
|
| 303 |
+
feature: 1 if prob > 0.2 else 0
|
| 304 |
+
for feature, prob in features_prob.items()
|
| 305 |
+
}
|
| 306 |
+
print(f" ✅ Features detected: {features_binary}")
|
| 307 |
+
|
| 308 |
+
result = {
|
| 309 |
+
# 🐛 (from app.py) Bug Fix: col_winner -> winner_col
|
| 310 |
+
"winner_position": {"row": int(winner_row), "col": int(winner_col)},
|
| 311 |
+
"features": features_binary,
|
| 312 |
+
"features_probability": features_prob,
|
| 313 |
+
"data_source": data_source
|
| 314 |
+
}
|
| 315 |
+
print(f"✅ SOM analysis complete!")
|
| 316 |
+
return result
|
| 317 |
+
except Exception as e:
|
| 318 |
+
print(f"❌ SOM analysis error: {e}")
|
| 319 |
+
return None
|
| 320 |
+
|
| 321 |
+
|
| 322 |
@app.post("/predict")
|
| 323 |
async def predict_malware(files: List[UploadFile] = File(...)):
|
| 324 |
"""
|
|
|
|
| 502 |
except Exception as e:
|
| 503 |
print(f"⚠️ Dataset upload failed: {e}")
|
| 504 |
|
| 505 |
+
# ===== 新增:執行 SOM 分析 (from app.py) =====
|
| 506 |
+
som_analysis = None
|
| 507 |
+
final_embedding_values = result.get("embedding", {}).get("values")
|
| 508 |
+
|
| 509 |
+
if SOM_AVAILABLE and final_embedding_values:
|
| 510 |
+
print("\n🗺️ Performing SOM analysis...")
|
| 511 |
+
try:
|
| 512 |
+
# 假設 SOM 分析需要第一個檔案的名稱作為參考
|
| 513 |
+
som_analysis = await analyze_with_som(
|
| 514 |
+
embedding=final_embedding_values,
|
| 515 |
+
predicted_family=result.get("final_label", "Unknown"),
|
| 516 |
+
filename=files[0].filename
|
| 517 |
+
)
|
| 518 |
+
except Exception as e:
|
| 519 |
+
print(f"❌ SOM analysis error: {e}")
|
| 520 |
+
|
| 521 |
+
# 將 SOM 結果附加到最終回傳
|
| 522 |
+
result["som_analysis"] = som_analysis
|
| 523 |
+
print(f"✅ SOM analysis complete.")
|
| 524 |
+
|
| 525 |
return result
|
| 526 |
|
| 527 |
except Exception as e:
|
|
|
|
| 529 |
import traceback
|
| 530 |
traceback.print_exc()
|
| 531 |
raise HTTPException(status_code=500, detail=f"Prediction failed: {str(e)}")
|
| 532 |
+
|
| 533 |
+
# ===== 新增:/predict-batch 路由 (from app.py) =====
|
| 534 |
+
@app.post("/predict-batch")
|
| 535 |
+
async def predict_batch(files: List[UploadFile] = File(...)):
|
| 536 |
+
"""
|
| 537 |
+
(from app.py)
|
| 538 |
+
簡化版: 只返回 label 和 embedding,不包含詳細資訊
|
| 539 |
+
"""
|
| 540 |
+
# 這將調用上面被修改過的 predict_malware 函數
|
| 541 |
+
result = await predict_malware(files)
|
| 542 |
|
| 543 |
+
return {
|
| 544 |
+
"final_label": result["final_label"],
|
| 545 |
+
"embedding": result["embedding"]["values"]
|
| 546 |
+
# SOM results are not included here, matching app.py's behavior
|
| 547 |
+
}
|
| 548 |
+
|
| 549 |
csv.field_size_limit(sys.maxsize)
|
| 550 |
|
| 551 |
JUMP_MNEMONICS = {"jmp", "je", "jne", "jl", "jg", "jle", "jge", "call", "ret"}
|