//特征名称
var features = Array("weight", "height", "age")
//字段转换成特征向量
var splitDatas = new VectorAssembler()
.setInputCols(features)
.setOutputCol("vector_features")
.transform(dataFrame.select("id", features:_*))
.randomSplit(Array(0.4, 0.3, 0.3))
//训练模型
var model:BucketedRandomProjectionLSHModel = new BucketedRandomProjectionLSH()
.setInputCol("vector_features") //待变换的特征
.setOutputCol("bkt_lsh") //变换后的特征名称
.setBucketLength(10d) //每个哈希桶的长度，更大的桶降低了假阴性率
.setNumHashTables(5) //哈希表的数量，散列表数量的增加降低了错误的否定率，如果降低它的值则会提高运行性能
.setSeed(100L) //随机种子
.fit(splitDatas.apply(0)) //训练
//通过模型转换数据
var transform = model.transform(splitDatas.apply(0))
transform.show(10, 100)
transform.printSchema()
//推荐信息,获取相关性较高的数据
var recommend= model.approxSimilarityJoin(splitDatas.apply(1), splitDatas.apply(2), 2, "distCol")
.select(
col("datasetA").getField("id").as("id"),
col("datasetB").getField("id").as("recommend_id"),
col("datasetA").getField("age").as("age"),
col("datasetB").getField("age").as("recommend_age"),
col("datasetA").getField("weight").as("weight"),
col("datasetB").getField("weight").as("recommend_weight"),
col("datasetA").getField("height").as("height"),
col("datasetB").getField("height").as("recommend_height"),
col("distCol")
)
recommend.orderBy("id", "distCol").show(100, 1000)
连续特征代码示例
// df 数据集是dataframe,并且words字段是格式是 ["我","爱","北京"，"天安门"]的词列表Array[String]
val word2Vec = new Word2Vec()
.setInputCol("words")
.setOutputCol("wordvec")
.setVectorSize(10)
.setMinCount(0)
val wvModel = word2Vec.fit(df)
val w2vDf = wvModel.transform(df)
val brp = new BucketedRandomProjectionLSH()
.setBucketLength(4d)
.setNumHashTables(10)
.setInputCol("wordvec")
.setOutputCol("hashes")
val brpModel = brp.fit(w2vDf)
val tsDf = brpModel.transform(w2vDf)
val key_value = List(0.13868775751827092,-0.11639275898904025,0.19808808788014898,0.2722799372859299,0.08275626220836721,-0.2846828463712129,0.2887565325463897,-0.05958885527697617,0.042977130971848965,-0.03787828763497287)
val key = Vectors.dense(key_value.toArray)
val a = brpModel.approxNearestNeighbors(tsDf, key , 3).toDF()
