- This code is the second stage of whole project called "Realtime event detection using deep learning technique" here
- As our proposed system, the output of CNN based model will be accumulated tweet data which play a role as the input of event detection model. Typically, this time series data look like underlying discrete signal. Value of each signal is accumulated frequency of tweets in given interval. The moving average can be used to filter noise in time series data. Our problem is similarto anomaly detection, novel detection, that also makes sense for the case of disaster event detection.The LSTM based prediction model refer to given current and past data to estimate next time pointin the time-series. Then, the errors distribution is used to event detection model which are built by anomaly likelihood.
- Keras
- TensorFlow
- sickit-learn
- GPyOpt
-
Step1: Configuration:
- First set the configuration settings in
configuration/config.py.- Xserver is set as True
- experiment_id: example
Result of running LSTM 2with input data isEarthquake1 - data_folder: example
data_folder: resources/data/earthquake/Earthquake2/
- Parametter of LSTM in
configuration/multi_step_lstm_config.- bath size, look back, look aheed,....
- Architecture
- First set the configuration settings in
-
Step2: Data Pre-processing:
- We need form raw data to turn to proper format for LSTM architecture that refer to look back and look head.
- Use notebook in the designed folder /notebook/... Note: Just run the first half of these notebook (Preprocessing), the rest is for visulization and time detection
-
Prediction Model Execution:
- The main LSTM models used are in the file models/lstm.py. Training the model and generating predictions two main files
are implemented by:
lstm_predictor.py, this file uses the default LSTM implementation by keras. - Run to get the output file
- The main LSTM models used are in the file models/lstm.py. Training the model and generating predictions two main files
are implemented by:
-
Step3:
- Comeback notebook folder and run the rest of code to time event detection and perform visualization.
- Example
