Please enable JavaScript.
Coggle requires JavaScript to display documents.
Machine learning for application optimization - Zoran Sevarac (Basic idea…
Machine learning for application optimization - Zoran Sevarac
We need systematic process
Basic idea is build module to solve it
Regression problem
Complex unknown function
Not possible to make magic
Check input parameters
Measure outputs
Average response time
Total execution time
Garbage collection
Apdex score
Depends to goal
Website
Desktop app
Mobile app
Procedure
Data collection
Prepare and load
Run
Apache JMeter
Data preprocessing
Clean data
Check errors
Visualize
Filter
Correlations
Normalize
Transform to one hot binary vectors
Prepare for model
Build model
Neural network which converts inputs to outputs
Sequence of layers
Each layer performs its operations
Not important to know what it does
Loss function
Network training
Iterative minimization - backpropagation
Seeks for layers that produce min if loss function
Optimization
Model tuning
Hidden layers
Optimization algorithm
Other options
Automated procedures
Evaluate model in test data
Evaluate model in production
Check performance in production
Pipeline
Apps
Deep Netts
Tensor flow
Neuroph - for beginners
Others
Case studies
Billing application
10 hours for 5 millions bills
After optimization 15% faster