Adds run fusion to anserini #2489
Conversation
…yserini convention
| * See the License for the specific language governing permissions and | ||
| * limitations under the License. | ||
| */ | ||
| package io.anserini.fusion; |
| this.runtag = runtag; | ||
| } | ||
|
|
||
| private class Document implements Comparable<Document>{ |
There was a problem hiding this comment.
| private class Document implements Comparable<Document>{ | |
| private class Document implements Comparable<Document> { |
| String docid; | ||
| double score; | ||
|
|
||
| public Document(String docid, double score) |
There was a problem hiding this comment.
House style, brace on previous line
| this.score = score; | ||
| } | ||
|
|
||
| @Override public int compareTo(Document a) |
There was a problem hiding this comment.
House style, annotation on separate line.
|
|
||
| } | ||
|
|
||
| public void writeTopic(String qid, HashMap<String,Double> results) { |
There was a problem hiding this comment.
| public void writeTopic(String qid, HashMap<String,Double> results) { | |
| public void writeTopic(String qid, Map<String,Double> results) { |
No need to specify impl
|
|
||
| public void writeTopic(String qid, HashMap<String,Double> results) { | ||
| int rank = 1; | ||
| ArrayList<Document> documents = new ArrayList<>(); |
There was a problem hiding this comment.
| ArrayList<Document> documents = new ArrayList<>(); | |
| List<Document> documents = new ArrayList<>(); |
ditto
| } | ||
| } | ||
|
|
||
| public class FuseRuns { |
There was a problem hiding this comment.
In Python, it doesn't violate best practice to have multiple classes in a single file.
In Java it's different... best practice is one class per file. If you need extra classes, organize as inner classes.
So above class should be it's own file?
|
|
||
| } | ||
|
|
||
| public static TreeMap<String, HashMap<String,DocScore>> createRunMap(String filename) throws FileNotFoundException, IOException { |
There was a problem hiding this comment.
Again, think about when you really need to specify impl?
|
|
||
| TreeMap<String, HashMap<String, Double>> finalRun; | ||
|
|
||
| if (fuseArgs.method.equals(FusionMethods.AVERAGE)) { |
There was a problem hiding this comment.
Hrm... I'm not sure I like the design.
Should there be like an actual RunsFuser class (I don't like the name but we can quibble about it separately) that performs the actual fusion. Right now, you have FusionMethods that's dispatched from a main class.
So, I'm thinking - move the logic of fusion into RunsFuser?
I like the design pattern of a lightweight main class that really just parses args, and dispatches to class that handles the heavy lifting.
E.g., HnswDenseSearcher is the main searcher class https://github.com/castorini/anserini/blob/master/src/main/java/io/anserini/search/HnswDenseSearcher.java
And SearchHnswDenseVectors.java does arg parsing and dispatches:
https://github.com/castorini/anserini/blob/master/src/main/java/io/anserini/search/SearchHnswDenseVectors.java
Thoughts? Open to discussion...
(Updated, fixed links)
This adds run fusion to anserini to match pyserini's implementation and closes #2355. To fuse 2 runs we use the following command
/bin/run.sh io.anserini.fusion.FuseRuns -method methodname -runs runa runb -output outputfileCurrently the supported methods are
rrf,interpolation, andaverage