To Do List - Githubissues

[ ] boxplot for segmentation outlier
[ ] residual to tree methods (check variable missing)
[ ] single Tree visualisation on 2 dimension split, which can compare with different tree methods.
[ ] Travis.yml update to R template from current c template
[ ] EDA report
[ ] Quick Xgboost run for important variable
[ ] Quick abnormal detection run

Done:

[x] Need a function to visualise the interactions or correlations: Can use the networkD3 package for the network plot, or use the DiagrammeR , igraph and vivagraph http://www.buildingwidgets.com/blog/?offset=1435611529118. [#26]
[x] Code efficient by either Rcpp or cleaner and elegant coding in R
[x] different residuals can be used on residual Plot (e.g. student, pearson, cooker's, etc.) [#52]
[x] deviance residual vs predicted target value [#52]
[x] different residuals can be used on residual Plot (e.g. student, pearson, cooker's, etc.) [#52]

Not Now:

Correlation matrix calculation for numerical matrix or data frame. Or find a quicker way for correlation calculation on big data. [#24]
Distributed / parallel Calculation
Need a model comparison function for all ML methods. A structure or template is needed to conduct this model comparison. A PMML structure would be a good starting point, even though not all ML packages support it. Or use caret package straight away.
which terms in the model have been affect by new factors and in what way?
1. Coefficient
2. Changes by level (vis)
3. Is it because of correlation?
4. What this new factor has explained? This is more like feature selection, and may to specific to glm / lm method.
Trends analysis:
1. Within the model to assess the consistency
2. Between datasets on different times to assess the development
3. There should have a function to give this feature. A starting point is the AbnormalDetection package and changepoint package. More timeseries analysis. Cannot find specific / achievable criteria.
[ ] historgram of deviance residual (symmetry). (REASON: currentl resiPlot has contour plot for this.)
[ ] deviance residual (sqrt(weighted deviance) * sign(a-e)). (REASON: #52 provides an open slot for this.)
[ ] deviance plot for outlier visualisation. (REASON: currentl resiPlot has contour plot for this.)
[ ] support multinomial analysis. (REASON: not urgent need.)
[ ] RcppParalle for CramersV function (REASON: not urgent need. And parallel overheads may have inverse effect.)

SixiangHu / DataMan

To Do List #18