from surf

.gitignore

@@ -33,3 +33,4 @@ tmp_*/
 model.rds
 scorecard.rds
 first_harp_date.df
+.ipynb_checkpoints

inst/surf_2019_02_demo/.gitignore

@@ -1,3 +1,5 @@
 .ipynb_checkpoints
 surf_201902.nb.html
 surf_201902.html
+01_surf_201902.html
+01_surf_201902.tex

inst/surf_2019_02_demo/01_surf_201902.rmd

@@ -3,6 +3,7 @@ title: "Simple Fannie Mae Example"
 output:
   html_document:
     df_print: paged
+  pdf_document: default
 ---
 
 ```{r}
@@ -53,7 +54,9 @@ select
   count(*) as cnt,
   n_default/cnt as odr # observed default rate
 from
-  table;
+  table
+group by
+  monthly.rpt.prd;
 ```
 This analysis only uses two columns, namely `default_12m` and `monthly.rpt.prd`. So I use `srckeep` to ensure that only those two columns are loaded.
 
@@ -78,6 +81,26 @@ system.time(a_wh2 <- a_wh1 %>% collect) # 60~70 plugged in
 a_wh2
 ```
 
+```{r}
+a_wh1 %>% 
+  srckeep(c("monthly.rpt.prd", "default_12m")) %>% 
+  map(function(chunk) {
+    chunk[1,]
+  }) %>% 
+  collect
+```
+
+
+```{r}
+a_wh1 %>% 
+  srckeep(c("monthly.rpt.prd", "default_12m")) %>% 
+  map(~{
+    .x[1,.SD]
+  }) %>% 
+  collect
+```
+
+
 once `collect` is called the resultant data is stored as a data.frame.
 
 However this is not the correct result, as the group by was performed within each chunk. Hence we need a second stage group by. The second group by takes no time at all, as everything was done in memory

presentation/.gitignore

@@ -1 +1,2 @@
 *.pptx
+*.pdf