closes #51

build.sh

@@ -14,11 +14,16 @@ fi
 echo "Downloading NLTK stopwords and punctuation packages..."
 python3 -c "import nltk; nltk.download('stopwords'); nltk.download('punkt')"
 
-if [ ! -e 'src/bin/bayes_model.pkl' ]; then
+if [ ! -e 'src/bin/bayes_model_sk.pkl' ]; then
  echo "Training Bayes model..."
  python3 ./src/scripts/train_bayes.py
 fi
 
+if [ ! -e 'src/bin/logreg_model_sk.pkl' ]; then
+ echo "Training LogReg model..."
+ python3 ./src/scripts/train_logreg.py
+fi
+
 if [ ! -d 'src/bin/rnn' ]; then
  echo "Training RNN model..."
  python3 ./src/scripts/train_rnn.py

src/classifier.py

@@ -10,6 +10,9 @@
 with open("src/bin/bayes_model_sk.pkl", "rb") as f:
  bayes_model = pickle.load(f)
 
+with open("src/bin/logreg_model_sk.pkl", "rb") as f:
+ logreg_model = pickle.load(f)
+
 with open("src/bin/vectorizer.pkl", "rb") as f:
  vectorizer = pickle.load(f)
 
@@ -25,14 +28,23 @@ def predict_bayes(text: str) -> dict[str, float]:
  }
 
 
+def predict_logreg(text: str) -> dict[str, float]:
+ features = vectorizer.transform([text])
+ probs = logreg_model.predict_proba(features)[0]
+ return {
+ "male": probs[1],
+ "female": probs[0],
+ }
+
+
 def predict_rnn(text: str) -> dict[str, float]:
  pred_arr = rnn_model.predict(np.array([text]))
  pred = float(pred_arr[0, 0])
  # F = 0, M = 1, P(!A) == 1 - P(A)
  return {"male": pred, "female": 1 - pred}
 
 
-pred_funcs = {"bayes": predict_bayes, "rnn": predict_rnn}
+pred_funcs = {"bayes": predict_bayes, "rnn": predict_rnn, "logreg": predict_logreg}
 
 
 def predict(text: str, classifier: str = "bayes") -> dict[str, str | dict[str, float]]:

src/scripts/train_logreg.py

@@ -0,0 +1,64 @@
+import nltk
+import numpy as np
+import os
+import pickle
+from sklearn.feature_extraction.text import CountVectorizer
+from sklearn.linear_model import LogisticRegressionCV
+
+MODEL_PATH = "src/bin/logreg_model_sk.pkl"
+VECTORIZER_PATH = "src/bin/vectorizer.pkl"
+
+STOPWORDS = nltk.corpus.stopwords.words("norwegian")
+np.random.seed(42)
+
+
+def read_file(path: str) -> str:
+ with open(path) as f:
+ content = f.read()
+ return content
+
+
+def _load_corpus() -> tuple[list[str] | np.ndarray]:
+ dir_f = "corpus/data/train/F/"
+ dir_m = "corpus/data/train/M/"
+ files_f = [dir_f + i for i in os.listdir(dir_f)]
+ files_m = [dir_m + i for i in os.listdir(dir_m)]
+ labels_f = np.full(len(files_f), "F")
+ labels_m = np.full(len(files_m), "M")
+
+ raw_data = list(map(read_file, files_f + files_m))
+ labels = np.concatenate((labels_f, labels_m))
+ return raw_data, labels
+
+
+def _save_model(model: LogisticRegressionCV, path: str) -> None:
+ with open(path, "wb+") as f:
+ pickle.dump(model, f)
+
+
+def train() -> None:
+ raw_data, labels = _load_corpus()
+
+ vectorizer = CountVectorizer(
+ stop_words=STOPWORDS,
+ ngram_range=[1, 3], # for usage of trigrams and bigrams
+ max_features=5000,
+ )
+
+ features = vectorizer.fit_transform(raw_data)
+ # shuffle data in case it is not permutation invariant
+ perms = np.random.permutation(len(labels))
+ # Does CSRMatrix handle this?
+ features = features[perms]
+ labels = labels[perms]
+
+ clf = LogisticRegressionCV(multi_class="multinomial", cv=5, max_iter=5000)
+
+ clf.fit(features, labels)
+
+ _save_model(clf, MODEL_PATH)
+ _save_model(vectorizer, VECTORIZER_PATH)
+
+
+if __name__ == "__main__":
+ train()

src/scripts/validation.py

@@ -38,11 +38,31 @@ def predict_on_str(text: str) -> int:
  return lambda x: [predict_on_str(s) for s in x]
 
 
+def get_predict_logreg() -> list[int]:
+ model_path = "src/bin/logreg_model_sk.pkl"
+ vectorizer_path = "src/bin/vectorizer.pkl"
+
+ with open(model_path, "rb") as f:
+ bayes_model = pickle.load(f)
+
+ with open(vectorizer_path, "rb") as f:
+ vectorizer = pickle.load(f)
+
+ def predict_on_str(text: str) -> int:
+ text = text.decode("UTF8")
+ features = vectorizer.transform([text])
+ pred = bayes_model.predict(features)
+ return 0 if pred[0] == "F" else 1
+
+ return lambda x: [predict_on_str(s) for s in x]
+
+
 if __name__ == "__main__":
  dev_set, test_set = get_dev_test()
 
  predict_rnn = get_predict_rnn("src/bin/rnn")
  predict_bayes = get_predict_bayes()
+ predict_logreg = get_predict_logreg()
 
  for name, (X, y) in [("dev", dev_set), ("test", test_set)]:
  X = X.numpy()
@@ -53,3 +73,7 @@ def predict_on_str(text: str) -> int:
  y_pred = predict_bayes(X)
  print("Results for", name, "with bayes:")
  print(classification_report(y, y_pred))
+
+ y_pred = predict_logreg(X)
+ print("Results for", name, "with logreg:")
+ print(classification_report(y, y_pred))