Progress Report II (Post)
Almambet Iskakov, Robert Pienta
12 00 AM ,Mon, Oct 26 2015
Quick Recap
- Our pipeline is almost complete
- We are now working on the final linking (steady state microstructures)
- We have created a series of optimization and cross-validation (CV)
-
Previously we had 11 datasets
-
Now we have 18 datasets
-
Number of PCs (for 90% variance) increased to 5
Cross-Validation (CV)
- To avoid overfitting we use leave-one-out CV
- For each Leave-One-Out test, we measure
- Goodness of fit:
- Mean Squared Error (MSE)
Parameter Optimization
- k-dimensional-grid parameter estimation using CV
- Requires training each model hundreds of times
- Finds approximately optimized parameters per model
Linear P-S Linkage
- Linear Regression
- Ridge (L2 Regression)
- Lasso (L1 Regression)
Example Linear Model Coeffs
For the solidication velocity case:
General Performance
Ongoing Work
- Explore transient datasets and create a time series model
- Increasing the number of samples from a single simulation
- Running an optimized version of each linkage model
- Nonlinear models
- 2 Suspicious datasets
Non-Linear P-S Linkage
- Our nonlinear modeling will be covered in subsequent blog posts
- Linear SVR
- -SVR
- Forest
- Random Forest