finished tensorflow intent classifier. need to clean and fine tune

app/endpoint/controllers.py

@@ -16,19 +16,11 @@
 endpoint = Blueprint('api', __name__, url_prefix='/api')
 
 # Loading ML Models at app startup
-from app.nlu.intent_classifer import IntentClassifier
-
-with app.app_context():
-    PATH = "{}/{}".format(app.config["MODELS_DIR"],
-                          app.config["INTENT_MODEL_NAME"])
-
-    sentence_classifier = IntentClassifier()
-    sentence_classifier.load(PATH)
-
-    synonyms = get_synonyms()
-    entity_extraction = EntityExtractor(synonyms)
-    app.logger.info("Intent Model loaded.")
+from app.nlu.classifiers.tf_intent_classifer import TfIntentClassifier
 
+sentence_classifier = None
+synonyms = None
+entity_extraction = None
 
 # Request Handler
 @endpoint.route('/v1', methods=['POST'])
@@ -202,7 +194,6 @@ def api():
     else:
         return abort(400)
 
-
 def update_model(app, message, **extra):
     """
     Signal hook to be called after training is completed.
@@ -212,20 +203,21 @@ def update_model(app, message, **extra):
     :param extra:
     :return:
     """
-    sentence_classifier.load(PATH)
+    global sentence_classifier
+
+    sentence_classifier = TfIntentClassifier()
+    sentence_classifier.load(app.config["MODELS_DIR"])
     synonyms = get_synonyms()
     global entity_extraction
     entity_extraction = EntityExtractor(synonyms)
     app.logger.info("Intent Model updated")
 
+update_model(app,"Modles updated")
 
 from app.nlu.tasks import model_updated_signal
-
 model_updated_signal.connect(update_model, app)
 
 from app.agents.models import Bot
-
-
 def predict(sentence):
     """
     Predict Intent using Intent classifier

app/nlu/classifiers/tf_intent_classifer.py

@@ -0,0 +1,136 @@
+import numpy as np
+import tensorflow as tf
+import spacy
+from sklearn.preprocessing import LabelBinarizer
+import os
+import cloudpickle
+import time
+
+class TfIntentClassifier():
+
+    def __init__(self):
+        self.model = None
+        self.nlp = spacy.load('en')
+        self.label_encoder = LabelBinarizer()
+        self.graph=None
+        print("im executed")
+
+    def train(self, X, y, models_dir=None, verbose=True):
+        """
+        Train intent classifier for given training data
+        :param X:
+        :param y:
+        :param outpath:
+        :param verbose:
+        :return:
+        """
+
+        def create_model():
+            """
+            Define and return tensorflow model.
+            """
+            model = tf.keras.Sequential()
+            model.add(tf.keras.layers.Dense(512, activation=tf.nn.relu, input_shape=(vocab_size,)))
+            model.add(tf.keras.layers.Dense(num_labels, activation=tf.nn.relu))
+            model.add(tf.keras.layers.Dense(num_labels, activation=tf.nn.softmax))
+
+            model.compile(loss='categorical_crossentropy',
+                          optimizer='rmsprop',
+                          metrics=['accuracy'])
+
+            model.summary()
+
+            return model
+
+        # spacy context vector size
+        vocab_size = 384
+
+        # create spacy doc vector matrix
+        x_train = np.array([list(self.nlp(x).vector) for x in X])
+
+        num_labels = len(set(y))
+        self.label_encoder.fit(y)
+        y_train = self.label_encoder.transform(y)
+
+        del self.model
+        tf.keras.backend.clear_session()
+        time.sleep(3)
+
+        self.model = create_model()
+        # start training
+        self.model.fit(x_train, y_train, shuffle=True, epochs=50, verbose=1)
+
+        if models_dir:
+            tf.keras.models.save_model(
+                self.model,
+                os.path.join(models_dir, "tf_intent_model.hd5")
+
+            )
+            if verbose:
+                print("TF Model written out to {}".format(os.path.join(models_dir, "tf_intent_model.hd5")))
+
+            cloudpickle.dump(self.label_encoder, open(os.path.join(models_dir, "labels.pkl"), 'wb'))
+
+            if verbose:
+                print("Labels written out to {}".format(os.path.join(models_dir, "labels.pkl")))
+
+
+    def load(self, models_dir):
+        try:
+            del self.model
+            tf.keras.backend.clear_session()
+            self.model = tf.keras.models.load_model(os.path.join(models_dir, "tf_intent_model.hd5"),compile=True)
+            self.graph = tf.get_default_graph()
+            print("Tf model loaded")
+            with open(os.path.join(models_dir, "labels.pkl"), 'rb') as f:
+                self.label_encoder = cloudpickle.load(f)
+                print("Labels model loaded")
+
+        except IOError:
+            return False
+
+    def predict(self, text):
+        """
+        Predict class label for given model
+        :param text:
+        :param PATH:
+        :return:
+        """
+        return self.process(text)
+
+    def predict_proba(self, x):
+        """Given a bow vector of an input text, predict most probable label. Returns only the most likely label.
+
+        :param x: raw input text
+        :return: tuple of first, the most probable label and second, its probability"""
+
+        x_predict = [self.nlp(x).vector]
+        with self.graph.as_default():
+            pred_result = self.model.predict(np.array([x_predict[0]]))
+        sorted_indices = np.fliplr(np.argsort(pred_result, axis=1))
+        return sorted_indices, pred_result[:, sorted_indices]
+
+    def process(self, x, return_type="intent", INTENT_RANKING_LENGTH=5):
+        """Returns the most likely intent and its probability for the input text."""
+
+        if not self.model:
+            print("no class")
+            intent = None
+            intent_ranking = []
+        else:
+            intents, probabilities = self.predict_proba(x)
+            intents, probabilities = [self.label_encoder.classes_[intent] for intent in
+                                      intents.flatten()], probabilities.flatten()
+
+            if len(intents) > 0 and len(probabilities) > 0:
+                ranking = list(zip(list(intents), list(probabilities)))[:INTENT_RANKING_LENGTH]
+
+                intent = {"intent": intents[0], "confidence": float("%.2f"%probabilities[0])}
+                intent_ranking = [{"intent": intent_name, "confidence": float("%.2f"%score)} for intent_name, score in ranking]
+            else:
+                intent = {"name": None, "confidence": 0.0}
+                intent_ranking = []
+        if return_type == "intent":
+            return intent
+        else:
+            return intent_ranking

app/nlu/intent_classifer.py

@@ -94,6 +94,7 @@ def predict_proba(self, X):
         import numpy as np
 
         pred_result = self.model.predict_proba(X)
+        print(pred_result)
         # sort the probabilities retrieving the indices of the elements in sorted order
         sorted_indices = np.fliplr(np.argsort(pred_result, axis=1))
         return sorted_indices, pred_result[:, sorted_indices]

app/nlu/tasks.py

@@ -3,7 +3,7 @@
 from app.intents.models import Intent
 
 from app import app
-from app.nlu.intent_classifer import IntentClassifier
+from app.nlu.classifiers.tf_intent_classifer import TfIntentClassifier
 
 from app import my_signals
 model_updated_signal = my_signals.signal('model-updated')
@@ -44,12 +44,10 @@ def train_intent_classifier(intents):
             X.append(example.get("text"))
             y.append(str(intent.id))
 
-    PATH = "{}/{}".format(app.config["MODELS_DIR"],
-                          app.config["INTENT_MODEL_NAME"])
-    intent_classifier = IntentClassifier()
+    intent_classifier = TfIntentClassifier()
     intent_classifier.train(X,
                             y,
-                            outpath=PATH, verbose=False)
+                            models_dir=app.config["MODELS_DIR"], verbose=True)
 
 
 def train_all_ner(story_id, training_data):

dockerfile

@@ -9,6 +9,7 @@ RUN python -m nltk.downloader "averaged_perceptron_tagger"; python
 RUN python -m nltk.downloader "punkt"; python
 RUN python -m nltk.downloader "stopwords"; python
 RUN python -m nltk.downloader "wordnet"; python
+RUN python -m spacy download en; python
 
 EXPOSE 8080
 

model_files/.gitignore

@@ -1 +1,3 @@
-*.model
+*.model
+*.pkl
+*.hd5

requirements.txt

@@ -24,4 +24,7 @@ gevent
 pytest
 cloudpickle
 flask_script
-blinker
+blinker
+h5py
+spacy
+tensorflow

run.py

@@ -1,4 +1,4 @@
 from app import app
 
 if __name__ == '__main__':
-    app.run(host='0.0.0.0', port=8080, debug=True, threaded=True)
+    app.run(host='0.0.0.0', port=8080, debug=False, threaded=False)