I'd agree with Peter in general for this kind of stuff; it's probably
better to rewrite the code from scratch a couple of times (maybe a few
different ways), and play some code golf or whatnot.

However if you want to use SRS, I'd advise to reduce the code to higher
level pseudocode (i.e. not "for(i = s.size()-1; i >= 0; i--) if(P[i] > max)
max = P[i]", but "m = max(node weight)" or something).
Then use cloze deletion on each line of the pseudocode. But I'd try to
reduce it to less than 10 lines of pseudocode per card - ideally your
pseudocode would be the minimal information necessary to remind you how to
reconstruct the algorithm.

O

On 29 January 2016 at 17:45, Peter Bienstman <[email protected]>
wrote:

> Hi,
>
> Not sure spaced repetition is the best tool for this. Personally, I would
> try implementing the algorithm from scratch to try and internalise it.
>
> But perhaps people have other suggestions.
>
> Cheers,
>
> Peter
>
>
> On Friday, 29 January 2016 18:41:32 UTC+1, Aditya BSRK wrote:
>>
>> Hi,
>> I am an on and off user of spaced repetition software. I have been
>> considering on how best to use Mnemosyne to remember code for algorithms.
>> For example, take the algorithm for finding the least common ancestor
>> <https://www.topcoder.com/community/data-science/data-science-tutorials/range-minimum-query-and-lowest-common-ancestor/#Another%20easy%20solution%20in%20O(N%20logN,%20O(logN)>
>> of two nodes in a tree. Lifting the code from topcoder:
>>
>>  void process3(int N, int T[MAXN], int P[MAXN][LOGMAXN])
>>   {
>>       int i, j;
>>
>>   //we initialize every element in P with -1
>>       for (i = 0; i < N; i++)
>>           for (j = 0; 1 << j < N; j++)
>>               P[i][j] = -1;
>>
>>   //the first ancestor of every node i is T[i]
>>       for (i = 0; i < N; i++)
>>           P[i][0] = T[i];
>>
>>   //bottom up dynamic programing
>>       for (j = 1; 1 << j < N; j++)
>>          for (i = 0; i < N; i++)
>>              if (P[i][j - 1] != -1)
>>                  P[i][j] = P[P[i][j - 1]][j - 1];
>>   }
>>
>>  int query(int N, int P[MAXN][LOGMAXN], int T[MAXN],
>>   int L[MAXN], int p, int q)
>>   {
>>       int tmp, log, i;
>>
>>   //if p is situated on a higher level than q then we swap them
>>       if (L[p] < L[q])
>>           tmp = p, p = q, q = tmp;
>>
>>   //we compute the value of [log(L[p)]
>>       for (log = 1; 1 << log <= L[p]; log++);
>>       log--;
>>
>>   //we find the ancestor of node p situated on the same level
>>   //with q using the values in P
>>       for (i = log; i >= 0; i--)
>>           if (L[p] - (1 << i) >= L[q])
>>               p = P[p][i];
>>
>>       if (p == q)
>>           return p;
>>
>>   //we compute LCA(p, q) using the values in P
>>       for (i = log; i >= 0; i--)
>>           if (P[p][i] != -1 && P[p][i] != P[q][i])
>>               p = P[p][i], q = P[q][i];
>>
>>       return T[p];
>>   }
>>
>> This is a fairly large algorithm, with some intense logic behind it. What
>> is the best way to be able to recall it on the fly when I need it?
>>
>> Thanks,
>> Aditya
>>
> --
> You received this message because you are subscribed to the Google Groups
> "mnemosyne-proj-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To post to this group, send email to [email protected]
> .
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/mnemosyne-proj-users/c1576a3f-304c-44c5-8597-4a70f8cb384c%40googlegroups.com
> <https://groups.google.com/d/msgid/mnemosyne-proj-users/c1576a3f-304c-44c5-8597-4a70f8cb384c%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
> For more options, visit https://groups.google.com/d/optout.
>

-- 
You received this message because you are subscribed to the Google Groups 
"mnemosyne-proj-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/mnemosyne-proj-users/CAO-0pXDkQYGuXeP2fqBE03pFjQOgiDgSQ9C30TiE4_rZ108JyA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to