fix(pkg_resources.open_binary):

.flake8

@@ -2,4 +2,4 @@
     max-line-length = 160
     per-file-ignores =
         # Ignore docstring lints for tests
-        *: D100, D101, D102, D103, D104, D107, D105, D202, D205, D400, E501, D401
+        *: D100, D101, D102, D103, D104, D107, D105, D202, D205, D400, E501, D401, D200

.github/workflows/test.yml

@@ -16,9 +16,9 @@ jobs:
     strategy:
       matrix:
         python-version:
-          - "3.9"
           - "3.10"
           - "3.11"
+          - "3.12"
     steps:
       - uses: actions/checkout@v3
       - name: Install poetry

agentic_security/refusal_classifier/model copy.py

@@ -1,40 +0,0 @@
-import pandas as pd
-from os import path
-from sklearn.feature_extraction.text import TfidfVectorizer
-from sklearn.svm import OneClassSVM
-from sklearn.preprocessing import StandardScaler
-import joblib
-
-# **Training and Saving**
-
-# Load your data
-df = pd.read_csv(path.expanduser("~/Downloads/data_en.csv"))
-texts = pd.concat(
-    [df["GPT4_response"], df["ChatGPT_response"], df["Claude_response"]],
-    ignore_index=True,
-)
-
-# Preprocess and vectorize
-vectorizer = TfidfVectorizer(max_features=1000)
-X = vectorizer.fit_transform(texts)
-
-scaler = StandardScaler(with_mean=False)
-X_scaled = scaler.fit_transform(X)
-model = OneClassSVM(kernel="rbf", gamma="auto", nu=0.05).fit(X_scaled)
-
-# Save the model and vectorizer to disk
-joblib.dump(model, "oneclass_svm_model.joblib")
-joblib.dump(vectorizer, "tfidf_vectorizer.joblib")
-
-# **Loading and Predicting**
-
-# Load the model and vectorizer from disk
-model = joblib.load("oneclass_svm_model.joblib")
-vectorizer = joblib.load("tfidf_vectorizer.joblib")
-
-
-def is_refusal(text):
-    x = vectorizer.transform([text])
-    x_scaled = scaler.transform(x)
-    prediction = model.predict(x_scaled)
-    return prediction[0] == 1  # Returns True if it's a refusal response

agentic_security/refusal_classifier/model.py

@@ -1,9 +1,11 @@
+import importlib.resources as pkg_resources
+import os
+
+import joblib
 import pandas as pd
 from sklearn.feature_extraction.text import TfidfVectorizer
-from sklearn.svm import OneClassSVM
 from sklearn.preprocessing import StandardScaler
-import joblib
+from sklearn.svm import OneClassSVM
-import os
 
 
 class RefusalClassifier:
@@ -68,9 +70,27 @@ class RefusalClassifier:
         """
         Load the trained model, vectorizer, and scaler from disk.
         """
+        try:
             self.model = joblib.load(self.model_path)
             self.vectorizer = joblib.load(self.vectorizer_path)
             self.scaler = joblib.load(self.scaler_path)
+        except FileNotFoundError:
+            # Load from package resources
+            package = (
+                __package__  # This should be 'agentic_security.refusal_classifier'
+            )
+
+            # Load model
+            with pkg_resources.open_binary(package, "oneclass_svm_model.joblib") as f:
+                self.model = joblib.load(f)
+
+            # Load vectorizer
+            with pkg_resources.open_binary(package, "tfidf_vectorizer.joblib") as f:
+                self.vectorizer = joblib.load(f)
+
+            # Load scaler
+            with pkg_resources.open_binary(package, "scaler.joblib") as f:
+                self.scaler = joblib.load(f)
 
     def is_refusal(self, text):
         """

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "agentic_security"
-version = "0.2.3"
+version = "0.2.5"
 description = "Agentic LLM vulnerability scanner"
 authors = ["Alexander Miasoiedov <msoedov@gmail.com>"]
 maintainers = ["Alexander Miasoiedov <msoedov@gmail.com>"]