libsvm.sql

SELECT 
  -- conversion for libsvm format
  label || ' ' || array_join(array_sort(
    feature_hashing(features), 
    (x, y) -> if(cast(substr(x, 1, strpos(x, ':') - 1) as bigint) < cast(substr(y, 1, strpos(y, ':') - 1) as bigint), 
      -1,
      if(substr(x, 1, strpos(x, ':') - 1) = substr(y, 1, strpos(y, ':') - 1), 0, 1)
    )
  ), ' ') as line
from
  input

sort_features.sql

SELECT 
  -- conversion for libsvm format
  array_sort(features, 
    (x, y) -> if(cast(substr(x, 1, strpos(x, ':') - 1) as double) < cast(substr(y, 1, strpos(y, ':') - 1) as double), 
      -1,
      if(substr(x, 1, strpos(x, ':') - 1) = substr(y, 1, strpos(y, ':') - 1), 0, 1)
    )
  ) as features, 
  label
from
  rf_input

sed -e 's/\"//g' -e 's/\[//' -e 's/\]//' 487600901.tsv | awk '{print $2,$1}' | sed -e 's/,/ /g' > 487600901.libsvm