chrome/renderer/safe_browsing/scorer_unittest.cc

   1 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "chrome/renderer/safe_browsing/scorer.h"
   6
   7 #include "base/file_path.h"
   8 #include "base/file_util.h"
   9 #include "base/format_macros.h"
  10 #include "base/hash_tables.h"
  11 #include "base/memory/scoped_ptr.h"
  12 #include "base/message_loop.h"
  13 #include "base/scoped_temp_dir.h"
  14 #include "base/threading/thread.h"
  15 #include "chrome/common/safe_browsing/client_model.pb.h"
  16 #include "chrome/renderer/safe_browsing/features.h"
  17 #include "testing/gmock/include/gmock/gmock.h"
  18 #include "testing/gtest/include/gtest/gtest.h"
  19
  20 namespace safe_browsing {
  21
  22 class PhishingScorerTest : public ::testing::Test {
  23  protected:
  24   virtual void SetUp() {
  25     // Setup a simple model.  Note that the scorer does not care about
  26     // how features are encoded so we use readable strings here to make
  27     // the test simpler to follow.
  28     model_.Clear();
  29     model_.add_hashes("feature1");
  30     model_.add_hashes("feature2");
  31     model_.add_hashes("feature3");
  32     model_.add_hashes("token one");
  33     model_.add_hashes("token two");
  34
  35     ClientSideModel::Rule* rule;
  36     rule = model_.add_rule();
  37     rule->set_weight(0.5);
  38
  39     rule = model_.add_rule();
  40     rule->add_feature(0);  // feature1
  41     rule->set_weight(2.0);
  42
  43     rule = model_.add_rule();
  44     rule->add_feature(0);  // feature1
  45     rule->add_feature(1);  // feature2
  46     rule->set_weight(3.0);
  47
  48     model_.add_page_term(3);  // token one
  49     model_.add_page_term(4);  // token two
  50
  51     // These will be murmur3 hashes, but for this test it's not necessary
  52     // that the hashes correspond to actual words.
  53     model_.add_page_word(1000U);
  54     model_.add_page_word(2000U);
  55     model_.add_page_word(3000U);
  56
  57     model_.set_max_words_per_term(2);
  58     model_.set_murmur_hash_seed(12345U);
  59   }
  60
  61   ClientSideModel model_;
  62 };
  63
  64 TEST_F(PhishingScorerTest, HasValidModel) {
  65   scoped_ptr<Scorer> scorer;
  66   scorer.reset(Scorer::Create(model_.SerializeAsString()));
  67   EXPECT_TRUE(scorer.get() != NULL);
  68
  69   // Invalid model string.
  70   scorer.reset(Scorer::Create("bogus string"));
  71   EXPECT_FALSE(scorer.get());
  72
  73   // Mode is missing a required field.
  74   model_.clear_max_words_per_term();
  75   scorer.reset(Scorer::Create(model_.SerializePartialAsString()));
  76   EXPECT_FALSE(scorer.get());
  77 }
  78
  79 TEST_F(PhishingScorerTest, PageTerms) {
  80   scoped_ptr<Scorer> scorer(Scorer::Create(model_.SerializeAsString()));
  81   ASSERT_TRUE(scorer.get());
  82   base::hash_set<std::string> expected_page_terms;
  83   expected_page_terms.insert("token one");
  84   expected_page_terms.insert("token two");
  85   EXPECT_THAT(scorer->page_terms(),
  86               ::testing::ContainerEq(expected_page_terms));
  87 }
  88
  89 TEST_F(PhishingScorerTest, PageWords) {
  90   scoped_ptr<Scorer> scorer(Scorer::Create(model_.SerializeAsString()));
  91   ASSERT_TRUE(scorer.get());
  92   base::hash_set<uint32> expected_page_words;
  93   expected_page_words.insert(1000U);
  94   expected_page_words.insert(2000U);
  95   expected_page_words.insert(3000U);
  96   EXPECT_THAT(scorer->page_words(),
  97               ::testing::ContainerEq(expected_page_words));
  98   EXPECT_EQ(2U, scorer->max_words_per_term());
  99   EXPECT_EQ(12345U, scorer->murmurhash3_seed());
 100 }
 101
 102 TEST_F(PhishingScorerTest, ComputeScore) {
 103   scoped_ptr<Scorer> scorer(Scorer::Create(model_.SerializeAsString()));
 104   ASSERT_TRUE(scorer.get());
 105
 106   // An empty feature map should match the empty rule.
 107   FeatureMap features;
 108   // The expected logodds is 0.5 (empty rule) => p = exp(0.5) / (exp(0.5) + 1)
 109   // => 0.62245933120185459
 110   EXPECT_DOUBLE_EQ(0.62245933120185459, scorer->ComputeScore(features));
 111   // Same if the feature does not match any rule.
 112   EXPECT_TRUE(features.AddBooleanFeature("not existing feature"));
 113   EXPECT_DOUBLE_EQ(0.62245933120185459, scorer->ComputeScore(features));
 114
 115   // Feature 1 matches which means that the logodds will be:
 116   //   0.5 (empty rule) + 2.0 (rule weight) * 0.15 (feature weight) = 0.8
 117   //   => p = 0.6899744811276125
 118   EXPECT_TRUE(features.AddRealFeature("feature1", 0.15));
 119   EXPECT_DOUBLE_EQ(0.6899744811276125, scorer->ComputeScore(features));
 120
 121   // Now, both feature 1 and feature 2 match.  Expected logodds:
 122   //   0.5 (empty rule) + 2.0 (rule weight) * 0.15 (feature weight) +
 123   //   3.0 (rule weight) * 0.15 (feature1 weight) * 1.0 (feature2) weight = 9.8
 124   //   => p = 0.99999627336071584
 125   EXPECT_TRUE(features.AddBooleanFeature("feature2"));
 126   EXPECT_DOUBLE_EQ(0.77729986117469119, scorer->ComputeScore(features));
 127 }
 128 }  // namespace safe_browsing