TEST(naive_bayes, serialization) {
NaiveBayes bayes;
std::vector<std::string> words1{"hello", "world", "world"};
std::vector<std::string> words2{"f**k", "world", "world"};
std::vector<Example> examples = {
Example(words1, "positive"),
Example(words2, "negative")
};
bayes.fit(examples);
auto scores = bayes.scores(examples[0]);
TempFile tmp_file;
std::ofstream ofs(tmp_file.filename);
boost::archive::text_oarchive oa(ofs);
oa << bayes;
ofs.close();
NaiveBayes bayes_restore;
std::ifstream ifs(tmp_file.filename);
boost::archive::text_iarchive ia(ifs);
ia >> bayes_restore;
auto scores_restore = bayes_restore.scores(examples[0]);
EXPECT_EQ(scores.get("positive"), scores_restore.get("positive"));
EXPECT_EQ(scores.get("negative"), scores_restore.get("negative"));
}
TEST(naive_bayes, example) {
NaiveBayes bayes;
std::vector<std::string> words1{"hello", "world", "world"};
std::vector<std::string> words2{"f**k", "world", "world"};
std::vector<Example> examples = {
Example(words1, "positive"),
Example(words2, "negative")
};
bayes.fit(examples);
print_scores(bayes.scores(examples[0]));
print_scores(bayes.scores(examples[1]));
}
TEST(naive_bayes, nomatching_words_example) {
// When nothing in dict matches, we want an empty map returned,
// rather than smoothed class distributions.
NaiveBayes bayes;
std::vector<std::string> words1{"hello", "world"};
std::vector<std::string> words2{"good", "morning"};
std::vector<Example> examples = {
Example(words1, "positive"),
Example(words2, "negative")
};
bayes.fit(examples);
std::vector<std::string> missing_words{"escape", "notfound"};
Example missing_example(missing_words, "positive");
auto scores = bayes.scores(missing_example);
EXPECT_EQ(0, scores.size());
EXPECT_EQ(0, scores.sum());
}
