Each team member folder contains an Excel file with daily logs grouped by project week.
A brief description of tasks completed in each week is provided below. The tables provide the general tasks that were completed and list individual contributions.
Week 1: Project Definition and Proposal
The first project week consisted of:
- Select/Define Project
- Create Project Management Tools
- Write Project Proposal
- Background Data Review
All team-members contributed to the above tasks.
Week 2: Data and Systems Understanding
Goals for the second project week included:
- Select Project Data and Complete EDA (still in progress at the end of Week 2)
- Understand Data Systems
- Learn InfluxDB and Telegraf
- Simulate Streaming in InfluxDB
- Anomaly Detection Research
Additionally, a Telegraf parser was worked on allowing streaming data directly into InfluxDB.
Nate | Mitch | Ryan | All |
---|---|---|---|
Learn InfluxDB and Telegraf, Simulate Streaming in InfluxDB, Investigate Streaming Detection Frameworks, Project Management | Anomaly Detection Research | Learn InfluxDB and Telegraf, EDA, XML Parser | Get a Local Setup of InfluxDB and Telegraf Working in Docker, Work on Streaming Data Parser |
Week 3: Anomaly Detection Model
Goals for the third project week included:
- Complete EDA from Week 2
- Data Cleaning/Feature Processing
- Build Streaming Framework for Anomaly Detection Model
- Meet with Domain Experts
- Start Reviewing Azure (secondary goal)
Helping implement the Telegraf parser to support getting streaming data into InfluxDB was also worked on.
Nate | Mitch | Ryan | All |
---|---|---|---|
Build Streaming Framework and Test Locally, Project Management, Build Anomaly Labeller App, Manually Label Anomalies | Ongoing Anomaly Detection Research, Start Building LSTM | Ongoing EDA, Help with Anomaly Detection Research, Support Telegraf Parsing, Manually Label Anomalies | Meet with Domain Experts |
Notes:
- Felt that additional time was required on anomaly detection research before building the model. Accordingly, building the model was moved to Week 4.
- Also, with the ongoing EDA this week data cleaning will was largely moved to Week 4.
- An additional item not included in the schedule was research of good anomaly detection performance measures, this will be completed early next week.
- Also needed to manually download data and self-label what we consider to be anomalies in the datasets. Five sensors were downloaded and labelled (2 of which required comments from EWS).
Week 4: Anomaly Detection Model Continued
The original proposal goal for the fourth project week was implementing the anomaly detection model. However, as discussed above for Week 3, several tasks were behind the proposal schedule. Therefore, the goals for this week were to complete the anomaly detection model and evaluate the labelled datasets from the previous week.
- Mid-Project Status Presentation
- Complete Anomaly Detection Model Pipeline (including understanding various LSTM architectures)
- Evaluate Downloaded Datasets
The five sensors downloaded were evaluated using the various labelling permutations discussed in /data/labelled-skyspark-data
. It was discussed that additional self-labelling would not be completed as it is very time intensive. Instead, assessment using the five sensors would be completed and additional assessment would just be completed on unlabelled data.
An additional task looked at was research on using a Spectral Residual transformation on the data as a step before the LSTM model.
Nate | Mitch | Ryan | All |
---|---|---|---|
Project Status Report, Spectral Residual Research, Support on Model Pipeline, Research LSTM architectures | LSTM Model | Cleaning Pipeline | Model Evaluation |
Week 5: Model Tuning and Implementation
The original proposal goal for the fifth project week was completing any tasks that gave the most value to the project. The week was used to complete several tasks behind schedule including:
- Implementing the anomaly detection framework with InfluxDB
- Tuning model performance
Reporting was also started this week.
Nate | Mitch | Ryan | All |
---|---|---|---|
Project Management, Report | LSTM Model Tuning | Model Implementation | Status Presentation |
Week 6: Dashboard
The original proposal goal for the sixth project week was completing the Dashboard for the project. The week was also used to implement a test environment with the package in InfluxDB as it was recognized that UDL would not have historical or streaming SkySpark data available in InfluxDB in the project timeline.
Final model tuning and test result organization for the report was also worked on.
Nate | Mitch | Ryan | All |
---|---|---|---|
Project Management, Report, Dashboard Support | Model Results | Test Environment, Dashboard | - |
The Dashboard did not get completed this week and will continue next week. It was also decided to continue additional testing for the project next week. As the report is in good shape, this is considered reasonable.
Week 7: Reporting
This week included final reporting and project wrap-up. As the the model was only implemented in a test environment, a good walk-through notebook was created as part of project wrap-up.
Additional time was also spent at the end of the week troubleshooting an issue that was noted with the non-deterministic nature of the LSTM.
Nate | Mitch | Ryan | All |
---|---|---|---|
Project Management, Report | Model Results in Report | Code and Dashboard Wrap-up | Walk-through Notebook Wrap-up, Final Presentation |