I'd like some feedback on a method I tried to implement that isn't working 100%. I'm making an Android app for practice where the user is given 20 random letters. The user then uses these letters to make a word of whatever size. It then checks a dictionary to see if it is a valid English word.
The part that's giving me trouble is with showing a "hint". If the user is stuck, I want to display the possible words that can be made. I initially thought recursion. However, with 20 letters this can take quite a long time to execute. So, I also implemented a binary search to check if the current recursion path is a a prefix to anything in the dictionary. I do get valid hints to be output however it's not returning all possible words. Do I have a mistake here in my recursion thinking? Also, is there a recommended, faster algorithm? I've seen a method in which you check each word in a dictionary and see if the characters can make each word. However, I'd like to know how effective my method is vs. that one.
private static void getAllWords(String letterPool, String currWord) {
//Add to possibleWords when valid word
if (letterPool.equals("")) {
//System.out.println("");
} else if(currWord.equals("")){
for (int i = 0; i < letterPool.length(); i++) {
String curr = letterPool.substring(i, i+1);
String newLetterPool = (letterPool.substring(0, i) + letterPool.substring(i+1));
if(dict.contains(curr)){
possibleWords.add(curr);
}
boolean prefixInDic = binarySearch(curr);
if( !prefixInDic ){
break;
} else {
getAllWords(newLetterPool, curr);
}
}
} else {
//Every time we add a letter to currWord, delete from letterPool
//Attach new letter to curr and then check if in dict
for(int i=0; i<letterPool.length(); i++){
String curr = currWord + letterPool.substring(i, i+1);
String newLetterPool = (letterPool.substring(0, i) + letterPool.substring(i+1));
if(dict.contains(curr)) {
possibleWords.add(curr);
}
boolean prefixInDic = binarySearch(curr);
if( !prefixInDic ){
break;
} else {
getAllWords(newLetterPool, curr);
}
}
}
private static boolean binarySearch(String word){
int max = dict.size() - 1;
int min = 0;
int currIndex = 0;
boolean result = false;
while(min <= max) {
currIndex = (min + max) / 2;
if (dict.get(currIndex).startsWith(word)) {
result = true;
break;
} else if (dict.get(currIndex).compareTo(word) < 0) {
min = currIndex + 1;
} else if(dict.get(currIndex).compareTo(word) > 0){
max = currIndex - 1;
} else {
result = true;
break;
}
}
return result;
}
The simplest way to speed up your algorithm is probably to use a Trie (a prefix tree)
Trie data structures offer two relevant methods. isWord(String) and isPrefix(String), both of which take O(n) comparisons to determine whether a word or prefix exist in a dictionary (where n is the number of letters in the argument). This is really fast because it doesn't matter how large your dictionary is.
For comparison, your method for checking if a prefix exists in your dictionary using binary search is O(n*log(m)) where n is the number of letters in the string and m is the number of words in the dictionary.
I coded up a similar algorithm to yours using a Trie and compared it to the code you posted (with minor modifications) in a very informal benchmark.
With 20-char input, the Trie took 9ms. The original code didn't complete in reasonable time so I had to kill it.
Edit:
As to why your code doesn't return all hints, you don't want to break if the prefix is not in your dict. You should continue to check the next prefix instead.
Is there a recommended, faster algorithm?
See Wikipedia article on "String searching algorithm", in particular the section named "Algorithms using a finite set of patterns", where "finite set of patterns" is your dictionary.
The Aho–Corasick algorithm listed first might be a good choice.
As title says I need to do the following. But I somehow am getting the wrong answer, perhaps something with the loops is wrong?
And here's what I have coded so far, but it seems to be giving me the wrong results. Any ideas, help, tips, fixes?
import java.util.ArrayList;
public class pro1
{
private String lettersLeft;
private ArrayList<String> subsets;
public pro1(String input)
{
lettersLeft = input;
subsets = new ArrayList<String>();
}
public void createSubsets()
{
if(lettersLeft.length() == 1)
{
subsets.add(lettersLeft);
}
else
{
String removed = lettersLeft.substring(0,1);
lettersLeft = lettersLeft.substring(1);
createSubsets();
for (int i = 0; i <= lettersLeft.length(); i++)
{
String temp = removed + subsets.get(i);
subsets.add(temp);
}
subsets.add(removed);
}
}
public void showSubsets()
{
System.out.print(subsets);
}
}
My test class is here:
public class pro1
{
public static void main(String[] args)
{
pro1s = new pro1("abba");
s.createSubsets();
s.showSubsets();
}
}
Try
int numSubsets = (int)java.lang.Math.pow(2,toSubset.length());
for (int i=1;i<numSubsets;i++) {
String subset = "";
for (int j=0;j<toSubset.length();j++) {
if ((i&(1<<j))>0) {
subset = subset+toSubset.substring(j,j+1);
}
}
if (!subsets.contains(subset)) {
subsets.add(subset);
}
}
where toSubset is the string that you wish to subset (String toSubset="abba" in your example) and subsets is the ArrayList to contain the results.
To do this we actually iterate over the power set (the set of all subsets), which has size 2^A where A is the size of the original set (in this case the length of your string).
Each subset can be uniquely identified with a number from 0 to 2^A-1 where the value of the jth bit (0 indexed) indicates if that element is present or not with a 1 indicating presence and 0 indicating absence. Note that the number 0 represents the binary string 00...0 which corresponds to the empty set. Thus we start counting at 1 (your example did not show the empty set as a desired subset).
For each value we build a subset string by looking at each bit position and determining if it is a 1 or 0 using bitwise arithmetic. 1<<j is the integer with a 1 in the jth binary place and i&(i<<j) is the integer with 1's only in the places both integers have a 1 (thus is either 0 or 1 based on if i has a 1 in the jth binary digit). If i has a 1 in the jth binary digit, we append the jth element of the string.
Finally, as you asked for unique subsets, we check if we have already used that subset, if not, we add it to the ArrayList.
It is easy to get your head all turned around when working with recursion. Generally, I suspect your problem is that one of the strings you are storing on the way down the recursion rabbit hole for use on the way back up is a class member variable and that your recursive method is a method of that same class. Try making lettersLeft a local variable in the createSubsets() method. Something like:
public class Problem1
{
private String originalInput;
private ArrayList<String> subsets;
public Problem1(String input)
{
originalInput = input;
subsets = new ArrayList<String>();
}
// This is overloading, not recursion.
public void createSubsets()
{
createSubsets(originalInput);
}
public void createSubsets(String in)
{
if(in.length() == 1)
{
// this is the stopping condition, the bottom of the rabbit hole
subsets.add(in);
}
else
{
String removed = in.substring(0,1);
String lettersLeft = in.substring(1);
// this is the recursive call, and you know the input is getting
// smaller and smaller heading toward the stopping condition
createSubsets(lettersLeft);
// this is the "actual work" which doesn't get performed
// until after the above recursive call returns
for (int i = 0; i <= lettersLeft.length(); i++)
{
// possible "index out of bounds" here if subsets is
// smaller than lettersLeft
String temp = removed + subsets.get(i);
subsets.add(temp);
}
subsets.add(removed);
}
}
Something to remember when you are walking through your code trying to think through how it will run... You have structured your recursive method such that the execution pointer goes all the way down the recursion rabbit hole before doing any "real work", just pulling letters off of the input and pushing them onto the stack. All the "real work" is being done coming back out of the rabbit hole while letters are popping off of the stack. Therefore, the first 'a' in your subsets list is actually the last 'a' in your input string 'abba'. I.E. The first letter that is added to your subsets list is because lettersLeft.length() == 1. (in.length() == 1 in my example). Also, the debugger is your friend. Step-debugging is a great way to validate that your code is actually doing what you expect it to be doing at every step along the way.
I encountered a problem while coding and I can't seem to find where I messed up or even why I get a wrong result.
First, let me explain the task.
It's about "Yijing Hexagram Symbols".
The left one is the original and the right one is the result that my code should give me.
Basically every "hexagram" contains 6 lines that can be either diveded or not.
So there are a total of
2^6 = 64 possible "hexagrams"
The task is to calculate and code a methode to print all possible combinations.
Thats what I have so far :
public class test {
public String toBin (int zahl) {
if(zahl ==0) return "0";
if (zahl ==1 ) return "1";
return ""+(toBin( zahl/2)+(zahl%2));
}
public void show (String s) {
for (char c : s.toCharArray()){
if (c == '1'){
System.out.println("--- ---");
}
if(c=='0'){
System.out.println("-------");
}
}
}
public void ausgeben (){
for(int i = 0 ; i < 64; i++) {
show (toBin(i));
}
}
}
The problem is, when I test the 'show'-methode with "10" I get 3 lines and not 2 as intended.
public class runner {
public static void main(String[] args){
test a = new test();
a.ausgeben();
a.show("10");
}
}
Another problem I've encoutered is, that since I'm converting to binary i sometimes have not enough lines because for example 10 in binary is 0001010 but the first "0" are missing. How can I implement them in an easy way without changing much ?
I am fairly new to all this so if I didn't explain anything enough or made any mistakes feel free to tell me.
You may find it easier if you use the Integer.toBinaryString method combined with the String.format and String.replace methods.
String binary = String.format("%6s", Integer.toBinaryString(zahl)).replace(' ', '0');
This converts the number to binary, formats it in a field six spaces wide (with leading spaces as necessary), and then replaces the spaces with '0'.
Well, there are many ways to pad a string with zeros, or create a binary string that is already padded with zeros.
For example, you could do something like:
public String padToSix( String binStr ) {
return "000000".substring( 0, 5 - binStr.length() ) + binStr;
}
This would check how long your string is, and take as many zeros are needed to fill it up to six from the "000000" string.
Or you could simply replace your conversion method (which is recursive, and that's not really necessary) with one that specializes in six-digit numbers:
public static String toBin (int zahl) {
char[] digits = { '0','0','0','0','0','0' };
int currDigitIndex = 5;
while ( currDigitIndex >= 0 && zahl > 0 ) {
digits[currDigitIndex] += (zahl % 2);
currDigitIndex--;
zahl /= 2;
}
return new String(digits);
}
This one modifies the character array ( which initially has only zeros ) from the right to the left. It adds the value of the current bit to the character at the given place. '0' + 0 is '0', and '0' + 1 is '1'. Because you know in advance that you have six digits, you can start from the right and go to the left. If your number has only four digits, well, the two digits we haven't touched will be '0' because that's how the character array was initialized.
There are really a lot of methods to achieve the same thing.
Your problem reduces to printing all binary strings of length 6. I would go with this code snippet:
String format = "%06d";
for(int i = 0; i < 64; i++)
{
show(String.format(format, Integer.valueOf(Integer.toBinaryString(i))));
System.out.println();
}
If you don't wish to print leading zeros, replace String.format(..) with Integer.toBinaryString(i).
I have code that reads a text file and creates an input array[] of boolean type. Its size is about 100,000-300,000 items. Now the problem I'm facing is to create all those subsets of size N, 3>=N>=9, that have contiguous true values.
E.g. for N=3, [true][true][true] is the required subset if all the 3 trues are in the continuous indexes.
Although I have an algorithm created, it's very slow. I need a better solution, which is fast and efficient.
Please suggest some ideas.
public static void createConsecutivePassingDays()
{
for (String siteName : sitesToBeTestedList.keySet())
{
System.out.println("\n*****************Processing for Site--->"+siteName+" ***********************");
LinkedHashMap<String,ArrayList<String>> cellsWithPassedTripletsDates=new LinkedHashMap<String, ArrayList<String>>();
for (String cellName : sitesToBeTestedList.get(siteName))
{
System.out.println("\n*****************Processing for Cell--->"+cellName+" ***********************");
boolean failed=false;
ArrayList<String> passedDatesTriplets=new ArrayList<String>();
int consecutiveDays=0;
String tripletDate="";
String prevDate_day="";
String today_Date="";
for (String date : cellDateKpiMetOrNotMap.get(cellName).keySet())
{
System.out.println("\nprocessing for Date-->"+date);
if(!(prevDate_day.trim().equals("")))
today_Date=getNextDay(prevDate_day.substring(0, prevDate_day.lastIndexOf('_')));
if(Connection.props.getProperty("INCLUDE_WEEKENDS").equalsIgnoreCase("FALSE"))
{
if(date.endsWith("SAT") || date.endsWith("SUN") || (!(date.substring(0, date.lastIndexOf('_')).equalsIgnoreCase(today_Date))))
{
if(consecutiveDays >= Reader.days)
{
passedDatesTriplets.add(tripletDate);
}
tripletDate="";
consecutiveDays=0;
prevDate_day=date;
continue;
}
}
if(cellDateKpiMetOrNotMap.get(cellName).get(date).equalsIgnoreCase("TRUE"))
{
if(tripletDate.equals(""))
tripletDate=date;
else
tripletDate+="#"+date;
consecutiveDays++;
}
else
{
failed=true;
if(consecutiveDays >= Reader.days)//kd
{
System.out.println("Triplet to be added-->"+tripletDate);
passedDatesTriplets.add(tripletDate);
}
tripletDate="";
consecutiveDays=0;
}
prevDate_day=date;
}
if(!failed)
passedDatesTriplets.add(tripletDate);
else
{
if(tripletDate.trim().split("#").length >= Reader.days)
{
passedDatesTriplets.add(tripletDate);
}
}
cellsWithPassedTripletsDates.put(cellName, passedDatesTriplets);
}
siteItsCellsWithPassedDates.put(siteName, cellsWithPassedTripletsDates);
}
System.out.println("\n************************************************SITES***************************************");
for (String site : siteItsCellsWithPassedDates.keySet())
{
System.out.println("\n********************Site="+site+" ***********************");
for (String cellName : siteItsCellsWithPassedDates.get(site).keySet())
{
System.out.println("\nCellName="+cellName);
System.out.println(siteItsCellsWithPassedDates.get(site).get(cellName));
}
System.out.println("***********************************************************");
}
System.out.println("********************************************************************************************");
}
First I would stay away from the array[boolean] a BitSet is more memory efficient in I'd expect it to be faster as well in your case. Since it will utilize the caches better. See boolean[] vs. BitSet: Which is more efficient?
For the algorithm:
Iterate through the datastructure.
When you come across the first true, remember its position (start) until you reach a false. This is the position end
At that point you have the start and end of a contiuous interval of true values, which is basically your result. You get your subsets as starting from start to end - n.
Repeat until the end of you datastructure
You can even parallize this by starting n-processes, each handling a different part of the array, starting with the first false value after the start of the segement and continuing over the end of the segement until the first false.
The simplest algo would be to check the N values starting at index x. if there is at least one false, then you can go directly to the index x+N. otherwise you can check index x+1;
if there is no valid sequence, then you will check size/N cells.
in pseudo-code :
int max = array.length - N;
int index = 0;
boolean valid = true;
while (index < max) {
valid = true;
for (check = index; check<index+N; check++){
valid = valid && array[check];
}
if (valid) {
// you got a continous sequence of true of size N
;
index++;
} else {
index = index + N;
}
}
also, with a BitSet instead of an array you could use the nextClearByte to get the index of the next false. the difference with the previous false minus N indicate the nomber of sequences of N true (with the previous false initially valued at -1).
I will sugggest you to create a stringbuilder and append 1 for every "true" value added to the boolean array and a 0 for every "false" added. Thus your stringbuilder will have a sequence of 1s and 0s. Then just use indexOf("111") to get the starting index of the three contiguous "true" values, it will be the starting index in the stringbuilder and in your boolean array as well.
A few days ago I had interview in some big company, name is not required :), and interviewer asked me to find solution to the next task:
Predefined:
There is dictionary of words with unspecified size, we just know that all words in dictionary are sorted (for example by alphabet). Also we have just a one method
String getWord(int index) throws IndexOutOfBoundsException
Needs:
Need to develop algorithm to find some input word in dictionary using java. For this we should implement method
public boolean isWordInTheDictionary(String word)
Limitations:
We cannot change the internal structure of dictionary, we have no access to internal structure, we do not know counts of elements in dictionary.
Issues:
I have developed modified-binary search, and will publish my variant(works variant) of algorithm, but are there another variants with logarithmic complexity? My variant has complexity O(logN).
My variant of implementation:
public class Dictionary {
private static final int BIGGEST_TOP_MASK = 0xF00000;
private static final int LESS_TOP_MASK = 0x0F0000;
private static final int FULL_MASK = 0xFFFFFF;
private String[] data;
private static final int STEP = 100; // for real test step should be Integer.MAX_VALUE
private int shiftIndex = -1;
private static final int LESS_MASK = 0x0000FF;
private static final int BIG_MASK = 0x00FF00;
public Dictionary() {
data = getData();
}
String getWord(int index) throws IndexOutOfBoundsException {
return data[index];
}
public String[] getData() {
return new String[]{"a", "aaaa", "asss", "az", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "test", "u", "v", "w", "x", "y", "z"};
}
public boolean isWordInTheDictionary(String word) {
boolean isFound = false;
int constantIndex = STEP; // predefined step
int flag = 0;
int i = 0;
while (true) {
i++;
if (flag == FULL_MASK) {
System.out.println("Word is not found ... Steps " + i);
break;
}
try {
String data = getWord(constantIndex);
if (null != data) {
int compareResult = word.compareTo(data);
if (compareResult > 0) {
if ((flag & LESS_MASK) == LESS_MASK) {
constantIndex = prepareIndex(false, constantIndex);
if (shiftIndex == 1)
flag |= BIGGEST_TOP_MASK;
} else {
constantIndex = constantIndex * 2;
}
flag |= BIG_MASK;
} else if (compareResult < 0) {
if ((flag & BIG_MASK) == BIG_MASK) {
constantIndex = prepareIndex(true, constantIndex);
if (shiftIndex == 1)
flag |= LESS_TOP_MASK;
} else {
constantIndex = constantIndex / 2;
}
flag |= LESS_MASK;
} else {
// YES!!! We found word.
isFound = true;
System.out.println("Steps " + i);
break;
}
}
} catch (IndexOutOfBoundsException e) {
if (flag > 0) {
constantIndex = prepareIndex(true, constantIndex);
flag |= LESS_MASK;
} else constantIndex = constantIndex / 2;
}
}
return isFound;
}
private int prepareIndex(boolean isBiggest, int constantIndex) {
shiftIndex = (int) Math.ceil(getIndex(shiftIndex == -1 ? constantIndex : shiftIndex));
if (isBiggest)
constantIndex = constantIndex - shiftIndex;
else
constantIndex = constantIndex + shiftIndex;
return constantIndex;
}
private double getIndex(double constantIndex) {
if (constantIndex <= 1)
return 1;
return constantIndex / 2;
}
}
It sounds like the part they really want you to think about is how to handle the fact that you don't know the size of the dictionary. I think they assume that you can give them a binary search. So the real question is how do you manipulate the range of the search as it progresses.
Once you have found a value in the dictionary that is greater than your search target (or out of bounds), the rest looks like standard binary search. The hard part is how do you optimally expand the range when the target value is greater than the dictionary value that you've looked up. It looks like you are expanding by a factor of 1.5. This could be really problematic with a huge dictionary and a small fixed initial step like you have (100). Think if there were 50 million words how many times your algorithm would have to expand the range upwards if you're searching for 'zebra'.
Here's an idea: use the ordered nature of the collection to your advantage by assuming the first letter of each word is evenly distributed amongst the letters of the alphabet (this will never be true, but without knowing more about the collection of words it's probably the best you can do). Then weight the amount of your range expansion by how far from the end you would expect the dictionary word to be.
So if you took your initial step of 100 and looked up the dictionary word at that index and it was 'aardvark', you would expand your range a lot more for the next step than if it was 'walrus.' Still O(log n) but probably much better for most collections of words.
Here is an alternative implementation that uses Collections.binarySearch. It fails if one of the words in the list starts with the Character '\uffff' (that is Unicode 0xffff and not a legal not a valid unicode character).
public static class ListProxy extends AbstractList<String> implements RandomAccess
{
#Override public String get( int index )
{
try {
return getWord( index );
} catch( IndexOutOfBoundsException ex ) {
return "\uffff";
}
}
#Override public int size()
{
return Integer.MAX_VALUE;
}
}
public static boolean isWordInTheDictionary( String word )
{
return Collections.binarySearch( new ListProxy(), word ) >= 0;
}
Update: I modified it so that it implements RandomAccess since the binarySearch in Collections would otherwise use a iterator based search on such a large list which would be extremely slow. This should now however be decently fast since the binary search will need only 31 iterations even though the List pretends to be as large as possible.
Here is a slightly modified version that remembers the smallest failed index to converge its proclaimed size to the actual size of the dictionary en passant and thus avoids almost all exceptions in successive lookups. Although you would need to create a new ListProxy instance whenever the size of the dictionary could have changed.
public static class ListProxy extends AbstractList<String> implements RandomAccess
{
private int size = Integer.MAX_VALUE;
#Override public String get( int index )
{
try {
if( index < size )
return getWord( index );
} catch( IndexOutOfBoundsException ex ) {
size = index;
}
return "\uffff";
}
#Override public int size()
{
return size;
}
}
private static ListProxy listProxy = new ListProxy();
public static boolean isWordInTheDictionary( String word )
{
return Collections.binarySearch( listProxy , word ) >= 0;
}
You have the right idea, but I think your implementation is overly complicated. You want to do a binary search, but you don't know what the upper bound is. So instead of starting at the middle, you start at index 1 (assuming dictionary indexes start at 0).
If the word you're looking for is "less than" the current dictionary word, halve the distance between the current index and your "low" value. ("low" starts at 0, of course).
If the word you're looking for is "greater than" the word at the index you just examined, then either halve the distance between the current index and your "high" value ("high" starts at 2) or, if index and "high" are the same, double the index.
If doubling the index gives you an out of range exception, you halve the distance between the current value and the doubled value. So if going from 16 to 32 throws an exception, try 24. And, of course, keep track of the fact that 32 is more than the max.
So a search sequence might look like 1, 2, 4, 8, 16, 12, 14 - found!
It's the same concept as a binary search, but rather than starting with low = 0, high = n-1, you start with low = 0, high = 2, and double the high value when you need to. It's still O(log N), although the constant is going to be a bit larger than with a "normal" binary search.
You can incur a one-time cost of O(n), if you know that the dictionary will not change. You can add all the words in the dictionary to a hashtable, and then any subsequent calls to isWordInDictionary() will be O(1) (in theory).
Use the getWord() API to copy the entire contents of the dictionary into a more sensible data structure (e.g. hash table, trie, perhaps even augmented by a Bloom filter). ;-)
In a different language:
#!/usr/bin/perl
$t=0;
$cur=1;
$under=0;
$EOL=int(rand(1000000))+1;
$TARGET=int(rand(1000000))+1;
if ($TARGET>$EOL)
{
$x=$EOL;
$EOL=$TARGET;
$TARGET=$x;
}
print "Looking for $TARGET with EOL $EOL\n";
sub testWord($)
{
my($a)=#_;
++$t;
return 0 if ($a eq $TARGET);
return -2 if ($a > $EOL);
return 1 if ($a > $TARGET);
return -1;
}
while ($r = testWord($cur))
{
print "Tested $cur, got $r\n";
if ($r == 1) { $over=$cur; }
if ($r == -1) { $under=$cur; }
if ($r == -2) { $over = $cur; }
if ($over)
{
$cur = int(($over-$under)/2)+$under;
$cur++ if ($cur <= $under);
$cur-- if ($cur >= $over);
}
else
{
$cur *= 2;
}
}
print "Found $TARGET at $r in $t tests\n";
The main benefit of this one is it is a bit simpler to understand. I think it may be more efficient if your first guesses are below the target since I don't think you are taking advantage of the space you have already "searched", but that is just with a quick glance at your code. Since it is looking for numbers for simplicity, it doesn't have to deal with not finding the target, but that is an easy extension.
#Sergii Zagriichuk hope the interview went well. Good luck with that.
I think just as #alexcoco said Binary Search is the answer.
Other options I see are only available if you could extend the dictionary. You could make it slightly better. E.g. You could count the words on each letter, and keep their track this way you would effectively had to work only on a subset of words.
Or yea as guys are saying to entirely implement your own dictionary structure.
I know this doesn't answer you question properly. But I cannot see other possibilities.
BTW would be nice to see your algorithm.
EDIT:
Expanding on my comment under answer of bshields...
#Sergii Zagriichuk even better it would be to remember the last index where we had null (no word), I think. Then at each run you could check if it is still true. If not then expand the range to a 'previous index' obtained by reversing the binary search behaviour, so we have null again. This way you would always adjust the size of the range of your search algorithm, thus adapting to the current state of the dictionary as needed. Plus the changes would have to be significant in order to cause your range adjustment so the adjustment wouldn't have any real negative impact on the algorithm. Also dictionaries tend to be static in nature so this should work :)
On one hand yes you are right with binary search implementation. But on the other hand in case dictionary is static and is not changed between lookups - we could suggest different algorithm. Here we have common problem - string sorting/search is different comparing to sorting/searching int array, so getWord(int i).compareTo(string) is O(min(length0, length1)).
Suppose we have request to find words w0, w1, ... wN, during lookup we could build up a tree with indicies (probably some suffix tree will good enough for this task).
During next lookup request we have following set a1, a2, ... aM, so to decrease average time we could first decrease range by searching position in the tree.
The problem with this implementation is concurrency and memory usage, so next step is implementing strategy to make search tree smaller.
PS: main aim was to check ideas and problems you suggest.
Well i think the info that dictionary is sorted can be utilized in a better way.
Say you are looking for a word "Zebra" , whereas the first guess search resulted in "abcg".
So we can use this info in chossing the second guess index . like in my case the resulted word is starting with a , whereas i am looking for something starting with z. So rather than making a static jump , i can make some calculated jump based on the current result and desired result. So in this way suppose if my next jump takes me to the word "yvu" , i now i am very near , so i will make a rather slow small jump than in the prev case.
Here is my solution.. uses O(logn) operations. First part of the code tries to find a estimate of the length and then the second part takes advantage of the fact that the dictionary is sorted and performs a binary search.
boolean isWordInTheDictionary(String word){
if (word == null){
return false;
}
// estimate the length of the dictionary array
long len=2;
String temp= getWord(len);
while(true){
len = len * 2;
try{
temp = getWord(len);
}catch(IndexOutOfBoundsException e){
// found upped bound break from loop
break;
}
}
// Do a modified binary search using the estimated length
long beg = 0 ;
long end = len;
String tempWrd;
while(true){
System.out.println(String.format("beg: %s, end=%s, (beg+end)/2=%s ", beg,end,(beg+end)/2));
if(end - beg <= 1){
return false;
}
long idx = (beg+end)/2;
tempWrd = getWord(idx);
if(tempWrd == null){
end=idx;
continue;
}
if ( word.compareTo(tempWrd) > 0){
beg = idx;
}
else if(word.compareTo(tempWrd) < 0){
end= idx;
}else{
// found the word..
System.out.println(String.format("getword at index: %s, =%s", idx,getWord(idx)));
return true;
}
}
}
Assuming the dictionary is 0-based, I would decompose the search in two parts.
First, given that the index to parameter to getWord() is an integer, and assuming that the index must be a number between 0 and the maximum positive integer, perform a binary search over that range in order to find the maximum valid index (irrespective of the word values). This operation is O(log N), since is a simple binary search.
Once obtained the size of the dictionary, a second ordinary binary search (again of complexity O(log N)) will bring on the desired answer.
Since O(log N)+O(log N) is O(log N), this algorithm complies with your requirement.
I'm in a hiring proccess which asked me this same problem...
My approach was a bit different, and considering the dictionary (webservice) I have, it's about 30% more efficient (for the words I've tested).
Here is the solution:
https://github.com/gustavompo/wordfinder
I'll not post the whole solution here because it's decoupled through classes and methods, but the core algorithm is this:
public WordFindingResult FindWord(string word)
{
var callsCount = 0;
var lowerLimit = new WordFindingLimit(0, null);
var upperLimit = new WordFindingLimit(int.MaxValue, null);
var wordToFind = new Word(word);
var wordIndex = _initialIndex;
while (callsCount <= _maximumCallsCount)
{
if (CouldNotFindWord(lowerLimit, upperLimit))
return new WordFindingResult(callsCount, -1, string.Empty, WordFindingResult.ErrorCodes.NOT_FOUND);
var wordFound = RetrieveWordAt(wordIndex);
callsCount++;
if (wordToFind.Equals(wordFound))
return new WordFindingResult(callsCount, wordIndex, wordFound.OriginalWordString);
else if (IsIndexTooHigh(wordToFind, wordFound))
{
upperLimit = new WordFindingLimit(wordIndex, wordFound);
wordIndex = IndexConsideringTooHighPreviousResult(lowerLimit, wordIndex);
}
else
{
lowerLimit = new WordFindingLimit(wordIndex, wordFound);
wordIndex = IndexConsideringTooLowPreviousResult(lowerLimit, upperLimit, wordToFind);
}
}
return new WordFindingResult(callsCount, -1, string.Empty, WordFindingResult.ErrorCodes.CALLS_LIMIT_EXCEEDED);
}
private int IndexConsideringTooHighPreviousResult(WordFindingLimit maxLowerLimit, int current)
{
return BinarySearch(maxLowerLimit.Index, current);
}
private int IndexConsideringTooLowPreviousResult(WordFindingLimit maxLowerLimit, WordFindingLimit minUpperLimit, Word target)
{
if (AreLowerAndUpperLimitsDefined(maxLowerLimit, minUpperLimit))
return BinarySearch(maxLowerLimit.Index, minUpperLimit.Index);
var scoreByIndexPosition = maxLowerLimit.Index / maxLowerLimit.Word.Score;
var indexOfTargetBasedInScore = (int)(target.Score * scoreByIndexPosition);
return indexOfTargetBasedInScore;
}