Code Review Stack Exchange is a question and answer site for peer programmer code reviews. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I am trying to create an invert index structure in C++. An invert index, in this case, refers to image retrieval system, where each invert index refers to a numbers of "Patch ID". (simply called "ID" in the code) These IDs are basically for describing many kind of pixel gradient patches from all images in the database.

The invert index is a map container and will be very sparse and each index of the invert index will contain a sub-map. A sub-map will have an "Image ID" and frequency counting integer. For example, if we have 2 images with "Image ID" 10 and 20, and both of these images have "Patch ID" 999, appear once in "Image ID" 10 and appear twice in "Image ID" 20, we will have the following:

invertID[999] == sub-map with <10,1> and <20,2>

I've written the code below for this, but somehow I am not quite satisfied.

One instance, I have to put back the sub-map into the invert ID's index as you can see below.

Another thing is I am not quite sure whether I am using the correct type of container for the correct job or not. I have about 5000 images (so that's 5000 "Image ID") and I see that "Patch ID" could vary from 1 to numbers as high as 40 billion.

In this case, performance-wise and design-wise, should I be using map or should I go for some other container? Suggestions are welcome.

Note: this will be run in a server that doesn't have a C++11 compiler.

unsigned long long int ID, index = 0;
map<unsigned long long int, map<int, int>> invertID;


//for the sake of convenient, two for loops below are shown as pseudocode.
for each image{ //loop with imageID, incrementing variable imageID in each iteration

//for each gradient patches in the image
for each gradient patch of this image{ //loop with index, incrementing variable index in each iteration.

    //processID encapsulates the procedure to generate an ID for each interesting point in the image. We basically read values in the respective file and do some calculation.
    ID = processID(index);

    //if this ID exists in InvertID,
    map<unsigned long long int, map<int,int>>::iterator mainMapIt = invertID.find(ID);
    if (mainMapIt != invertID.end()){

        //if this ImageID key exists in InvertID sub-map,
        map<int,int> M = mainMapIt->second;
        map<int,int>::iterator subMapIt = M.find(imageID);
        if (subMapIt != M.end()){
            //the index already exists, so increment the number (frequency) of this ImageID key
            //the index does not exist, so add ImageID key with value 1 into the InvertID
            M[imageID] = 1;

        //get this sub-Map back to the invert ID's index.
        invertID[ID] = M;
        //the ID does not exist in the InvertId, yet, so 
        //create the first empty map with the key as image ID with value 1 and put it into the invertID.
        map<int,int> M;
        M[imageID] = 1;
        invertID[ID] = M;
share|improve this question

You are taking a copy of the inner map, modifying it, then putting it back. That could well be inefficient. You could take the inner map by reference then modify it.

However you can actually make use of the fact that map operator[] inserts if it does not find the element. As that is the behaviour here you can do what you did in 2 lines:

map< int, int > & innerMap = mainMapIt[ID]; // will be created as empty if not found
++innerMap[ imageID ];

On a style issue, one does not commonly use an upper case letter as a variable. You also have ID as a variable but elsewhere your variables begin with a lower case letter. You should be consistent throughout in your style, and ideally adopt what is the most common style for C++ programmers (which is generally agreed to be beginning with a lower-case. Some prefer camel-case and some under-score delimited).

You could also decide instead of nested maps, to have a single map with a key consisting of std::pair<unsigned long long, int >. This might be slightly more efficient as copying maps around could be expensive. It would mean however that you could not as easily search for all the elements from ID.

share|improve this answer

Inverted indexes are often implemented using skip lists instead of maps (which are elaborate search trees). The issue with std::map is that even though the complexity is good, for real lists the overhead of the map is too important. Plain vectors are usually way more efficient for small lists, and skip lists avoid the biggest issue with vectors: search in linear time.

For an inverted index with a small number of items, I've seen a great increase in performance simply from switching from map to vector in C++.

Otherwise, as CashCow said, the [] operator use the default constructor of your value when the key doesn't exist, so it isn't a real issue.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.