update

bilstm-crf/model/bi_lstm_crf.py

@@ -0,0 +1,97 @@
+import pandas as pd
+import numpy as np
+from keras.preprocessing.sequence import pad_sequences
+from keras.utils import to_categorical
+from sklearn.model_selection import train_test_split
+from keras.models import Model, Input
+from keras.layers import LSTM, Embedding, Dense, TimeDistributed, Dropout, Bidirectional
+from keras_contrib.layers import CRF
+
+
+data = pd.read_csv('../data/ner_dataset.csv', encoding='latin1')
+
+# fill NaN
+data = data.fillna(method='ffill')
+
+# Save all words as a list
+words = list(set(data['Word'].values))
+n_words = len(words)
+
+tags = list(set(data["Tag"].values))
+n_tags = len(tags)
+
+# Sentence class
+class SentenceGetter(object):
+
+    def __init__(self, data):
+        self.n_sent = 1
+        self.data = data
+        self.empty = False
+        agg_func = lambda s: [(w, p, t) for w, p, t in zip(s["Word"].values.tolist(),
+                                                           s["POS"].values.tolist(),
+                                                           s["Tag"].values.tolist())]
+        self.grouped = self.data.groupby("Sentence #").apply(agg_func)
+        self.sentences = [s for s in self.grouped]
+
+    def get_next(self):
+        try:
+            s = self.grouped["Sentence: {}".format(self.n_sent)]
+            self.n_sent += 1
+            return s
+        except:
+            return None
+
+getter = SentenceGetter(data)
+sentences = getter.sentences # get all sentences
+
+# max_len = 75
+word2idx = {w: i + 1 for i, w in enumerate(words)}
+n_words = len(word2idx)
+# word2idx['<unk>'] = len(word2idx) + 1
+tag2idx = {t: i + 1 for i, t in enumerate(tags)}
+tag2idx['<pad>'] = 0
+n_tags = len(tag2idx) # Due to <pad>, here total tag number is from 17 to 18
+
+# Word2inx & Padding for X
+X = [[word2idx[w[0]] for w in s] for s in sentences]
+X = pad_sequences(maxlen=max_len, sequences=X, padding="post", value=0)
+
+# Word2inx & Padding for y
+y = [[tag2idx[w[2]] for w in s] for s in sentences]
+y = pad_sequences(maxlen=max_len, sequences=y, padding="post", value=0)
+
+# Get one-hot labels
+y = [to_categorical(i, num_classes=n_tags) for i in y]
+X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.1)
+
+#==============Bi-LSTM CRF=============
+input = Input(shape=(max_len,))
+model = Embedding(input_dim=n_words + 1, output_dim=20,
+                  input_length=max_len, mask_zero=True)(input)  # 20-dim embedding
+model = Bidirectional(LSTM(units=50, return_sequences=True,
+                           recurrent_dropout=0.1))(model)  # variational biLSTM
+model = TimeDistributed(Dense(50, activation="tanh"))(model)  # a dense layer as suggested by neuralNer
+crf = CRF(n_tags)  # CRF layer
+out = crf(model)  # output
+
+model = Model(input, out)
+model.compile(optimizer="rmsprop", loss=crf.loss_function, metrics=[crf.accuracy])
+model.summary()
+
+history = model.fit(X_tr, np.array(y_tr), batch_size=32, epochs=1,
+                    validation_split=0.1, verbose=1)
+
+# Predictions.
+idx2word = {value: key for key, value in word2idx.items()}
+idx2tag = {value: key for key, value in tag2idx.items()}
+
+p_all = model.predict(np.array(X_te)) # (4796, 75, 18)
+p_all= np.argmax(p_all, axis=-1) # (4796, 75)
+p_all_tags = [[idx2tag[idx] for idx in s if idx!=0] for s in p_all] # ['B-gpe', 'O', 'O', 'O']
+
+true_all = np.argmax(y_te, -1)
+true_all_tags = [[idx2tag[idx] for idx in s if idx!=0] for s in true_all]
+
+# Evaluation
+from seqeval.metrics import f1_score
+f1_score(true_all_tags, p_all_tags)