def __init__(self, settings_instance: Settings):
self.settings = settings_instance
self.category_codes = {
'business': 0,
'entertainment': 1,
'politics': 2,
'sport': 3,
'tech': 4,
}
self.category_names = {v: k for k, v in self.category_codes.items()}
self.model = None
self.tf_idf = None
self.text_processing = TextProcessing()
def classify(flag=None):
#Pre-process tweets
wordId,tweetId=tp.process(2)
#TF-IDF
docwords,docCatIds=svmp.cacheTweetsInList(wordId,tweetId,2)
#SVM Processing
#1. Create libSVM file
svmp.createTestFile(docwords,docCatIds,tweetId)
#2. Train the SVM
if flag is not None:
return svmp.testSVM(1)
else:
svmp.testSVM()
def classify():
labels =[]
#sampletext = request.form.get('txt1')
wordId,tweetId=tp.process(2)
#TF-IDF
docwords,docCatIds=svmp.cacheTweetsInList(wordId,tweetId,2)
#SVM Processing
#1. Create libSVM file
svmp.createTestFile(docwords,docCatIds,tweetId)
#2. Train the SVM
#svmp.trainLibLinear()
labels = svmp.testSVM(1)
#clean_dict(start_pos,end_pos)
return labels
class NewsPredictor:
def __init__(self, settings_instance: Settings):
self.settings = settings_instance
self.category_codes = {
'business': 0,
'entertainment': 1,
'politics': 2,
'sport': 3,
'tech': 4,
}
self.category_names = {v: k for k, v in self.category_codes.items()}
self.model = None
self.tf_idf = None
self.text_processing = TextProcessing()
def load_model(self, path_to_model: str, path_to_tf_idf: str) -> None:
with open(path_to_model, 'rb') as f_model, open(path_to_tf_idf, 'rb') as f_tf_idf:
self.model = pickle.load(f_model)
self.tf_idf = pickle.load(f_tf_idf)
def predict(self, texts: List[str]) -> List[Dict[str, Any]]:
df = self.text_processing.pre_process_text_list(texts)
features = self.tf_idf.transform(df['Text_Processed']).toarray()
prediction_probabilities = self.model.predict_proba(features)
predictions = []
for prob in prediction_probabilities:
max_index = np.argmax(prob)
max_prob = prob[max_index]
prediction = {
'prediction': self.category_names[max_index],
'probability': round(max_prob, 6),
}
predictions.append(prediction)
return predictions
