# Interface: com.pnfsoftware.jeb.core.units.code.android.IMethodEmbeddingManager Interface for an embedding manager for decompiled dex methods. The manager is responsible for generating embeddings as well as comparing good candidate matches to a base embedding. Implementations of this interface must have a no\-arg constructor to allow instantiation. ## Method: findBestMatch - parameter: `target`, type: `com.pnfsoftware.jeb.core.units.code.MethodEmbedding` - parameter: `matches`, type: `java.util.List` - return type: `com.pnfsoftware.jeb.core.units.code.MethodMatch` Description: Given a target embedding and a collection of top candidate matches for that embedding, determine which one is better. This is used to select the best match among several good candidates. Note that this method is optional and may simply return the first embedding in the provided collection of candidates \(that is, the one with the best vector similarity score according to the vector database\). However, it can be implemented to perform a more in\-depth analysis of the candidates and select the best one among them \(that may not have the best vector similarity\), for example, by using the extra data stored provided during embedding generation. parameter: target: the target embedding representing the method for which we are looking for a match parameter: matches: a non\-empty list of candidate matches for the target embedding; the list is ordered by descending similarity scores; these candidates passed the thresholds defined by [#getDatabaseQuerySettings()](#getDatabaseQuerySettings()), but they may still be of varying quality; this collection is never null and contains at least one candidate return: the best embedding ## Method: generateEmbedding - parameter: `m`, type: `com.pnfsoftware.jeb.core.units.code.android.ir.IDMethodContext` - parameter: `extras`, type: `java.util.Map` - parameter: `unstoredExtras`, type: `java.util.Map` - return type: `float[]` Description: Generate a method embedding, given an IR method context for that method. parameter: m: method context parameter: extras: a sink map that may receive extra information about the embedding; the extra \(key, value\) pairs stored in this method when generating the embedding will be stored in the final computed [MethodEmbedding](MethodEmbedding) object; watch out\! the accepted value types are: Integer, Long, Float, Double, String, byte\[\] parameter: unstoredExtras: a sink map that may receive extra information about the embedding, but that will not be stored in the final computed [MethodEmbedding](MethodEmbedding) object, and will not be stored in the vector database return: the embedding object, or null if no decent embedding can be generated for the given method \(e.g. the method is too small or too simple to be meaningfully represented as an embedding\) ## Method: getDatabaseQuerySettings - return type: `int[]` Description: Get vector database query thresholds. return: an 3\-element array \(k, n, s\) of integers defining the thresholds for candidate matches: - `k`: the number of embeddings to find and consider - - `n`: the top number of hits, among a maximum pool of \`k\` embeddings, to be returned when searching for a match; \`k\` must be greater than \`n\` - - `s`: the minimal similarity score \(in percentages, from \-100 to 100\) between two vectors to be considered a good candidate match; if the score is below this threshold, the match will be discarded and not returned as a candidate - ## Method: getSimilarityMethod - return type: `java.lang.String` Description: Get the similarity method name. return: the similarity method used by the vector database to find good matches approximating a target embedding; currently, the only supported value is "COSINE", and this method must return it\) ## Method: getVectorDimension - return type: `int` Description: Get the vector dimension. return: the count of dimensions for embeddings generating by [#generateEmbedding](#generateEmbedding); all vectors generated and used by this manager must have this dimension