# Meeting Log

## Meeting 7 July
##### Author: Amanda

### TODO
1. Integrate our code with GAT code

## Meeting 9 June
##### Author: Amanda

### TODO
1. Finish ppt

### Notes:

## Meeting 5 May
##### Author: Jessica

### TODO
1. Complete data generation for GAT anomaly
2. Create a supervised learning system for domain classification
3. Look up: contrastive learning, seimese network, triplet loss for option 2 when supervised learning system fails

### Notes:
1. Contrastive Learning

## Meeting 21 April
##### Author: Jessica

### TODO
1. Label data for grouping machines using clustering (Jessica)
* Try using statistical, temporal and spectral features
* Don't use K-Means
* Expected output: `number of cluster` < `number of machines`
2. Label timestamps for detecting anomalies (Mew. Amanda)
* Use GAT with sliding window ranging 10 ~ 30
* add an `is_anomaly` to the original csv file
* 0 means normal
* 1 means anomaly

### Notes:
1. Label data for grouping machine

2. Particular things to take note of

3. Output Dimensions

## Meeting 7 April
##### Author: Jessica
### Changes
1. Explore ways to do MTSAD using a GAT-based model with the pipeline drawn in the notes below

### TODO
1. Create a report to justify that ML-based model for MTSAD is impossible
2. Familiarize with GAT code so that it will be easier to make changes
3. Perform clustering on historical data (piece-wise using size 290)
4. Create weak label using GAT on historical data (use sliding window with size 290) for the training data
5. Code review for GAT...?

### Notes:

## Meeting 31 March
##### Author: Jessica
### TODO
1. Search for a predefined function for seasonality and stationary
2. Find the connection between historical and input batch data
3. Formulate a function to quantify the "anomaly score" for the proposed method and a;sp the "customized score" for different cases
4. Check the LoF function and update the graphs at HackMD

## Meeting 24 March
##### Author: New
### TODO
1. Generate the output from GAT, VAR, Spatial and Temporal model for 20 folders.
2. Making an analysis on historical data.

## Meeting 17 March
##### Author: Jessica
### TODO
1. Implement GAT ASAP
2. Write report after getting the results on both GAT and VAR

## Meeting 10 March
##### Author: Jessica
### Changes
1. Get ground truth from deep learning models (**GAT** and VAR)
2. PPT and HackMD Report would be postponed to **March 24**

### TODO
1. Find more statistical insights from clustering (Jessica)
2. Implement VAR (Amanda)
3. Implement GAT (New)

## Meeting 3 March
##### Author: Jessica
### TODO
1. Get more data until number of folders is 20 (Jay)
2. Write a report in HackMD about the progress this week. Don't forget to write in your comments and thoughts on the findings. **Deadline: 17 March** (New, Jessica)
3. Find first version of hyperparameters for the ARIMA model (New)
4. Obtain a the raw results of of temporal and spatial model separately (New, Jessica)
5. Conduct the statistical analysis on data. Focus on *Network_Out* (Jessica)
6. Catch Amanda up on the work when she comes (Jay, New)

### Notes

###### tags: `Meeting Log`