Filter: Difference between revisions
no edit summary
(add filter primitive page) |
No edit summary |
||
Line 14:
0 2 4 6
</pre>
There is disticnt behaviour between the filters implemented in K6 and K9. A toy prototype to highlight the difference could be implemented as:
<pre>
f6:{[f;x] x@&f'x} / k6 variant
f9:{[f;x] x@&f x} / k9 variant
</pre>
Both try to abstract the very common pattern in the array programming languages of constructing a boolean mask on the items of the array and using that mask to index back into the array. This effectively filters the array where the boolean predicate returns true.
The difference is that K6 calls <code>f</code> on each element of the array while K9 calls <code>f</code> on the whole array. Considering the [[atomicity]] of most K primitives, applying a function to each element and applying to the whole array may be equivalent, then results agree:
<pre>
/filter greater than 5
f6[5<;!10]
6 7 8 9
f9[5<;!10]
6 7 8 9
</pre>
However, one drawback of the K6 variant is that does not allow the data in <code>x</code> to be used as a reference for constructing the boolean mask. For exemple, consider the problem of filtering all values equal to the maximum in a list. In K9 one could do:
<pre>
{x=|/x}#3 2 1 3 / equivalent to f9[{x=|/x};3 2 1 3]
3 3
</pre>{{Works in|K9}}
In K6, on the other hand, since the filter does not see the whole array, the value to filter for has to be pre-computed or the filter in f9 has to be reimplemented:
<pre>
m:|/3 2 1 3;(m=)#3 2 1 3 / pre-computing the maximum and filtering
3 3
{x@&x=|/x}3 2 1 3 /f9 pattern
3 3
</pre>{{Works in|ngn/k}}
In this regard, K9 filter can be seen as more general. One possible explanation for the prior implementation using [[each]] is that it mirrors the most common form of the filter function implemented in other programming languages<ref>https://en.wikipedia.org/wiki/Filter_(higher-order_function)</ref>.
| |||