Description

Maps columns of indices to strings, based on the model fitted by {@link StringIndexer}.

While {@link StringIndexerModel} maps string to index, IndexToString maps index to string. However, IndexToString does not have a corresponding {@link com.alibaba.alink.pipeline.EstimatorBase}. Instead, IndexToString uses model data in StringIndexerModel to perform predictions.

IndexToString use the name of the {@link StringIndexerModel} to get the model data. The referenced {@link StringIndexerModel} should be created before calling transform method.

A common use case is as follows:

StringIndexer stringIndexer = new StringIndexer() .setModelName(“name_a”) // The fitted StringIndexerModel will have name “name_a”. .setSelectedCol(…);

StringIndexerModel model = stringIndexer.fit(…); // This model will have name “name_a”.

IndexToString indexToString = new IndexToString() .setModelName(“name_a”) // Should match the name of one StringIndexerModel. .setSelectedCol(…) .setOutputCol(…);

indexToString.transform(…); // Will relies on a StringIndexerModel with name “name_a” to do transformation.

The reason we use model name registration mechanism here is to make possible stacking both StringIndexer and IndexToString into a {@link Pipeline}. For examples,

StringIndexer stringIndexer = new StringIndexer() .setModelName(“si_model_0”).setSelectedCol(“label”);

MultilayerPerceptronClassifier mlpc = new MultilayerPerceptronClassifier() .setVectorCol(“features”).setLabelCol(“label”).setPredictionCol(“predicted_label”);

IndexToString indexToString = new IndexToString() .setModelName(“si_model_0”).setSelectedCol(“predicted_label”);

Pipeline pipeline = new Pipeline().add(stringIndexer).add(mlpc).add(indexToString);

pipeline.fit(…);

Parameters

Name Description Type Required？ Default Value
modelName Name of the model String ✓
selectedCol Name of the selected column used for processing String ✓
reservedCols Names of the columns to be retained in the output table String[] null
outputCol Name of the output column String null

Script Example

Code

  1. data = np.array([
  2.     ["football"],
  3.     ["football"],
  4.     ["football"],
  5.     ["basketball"],
  6.     ["basketball"],
  7.     ["tennis"],
  8. ])
  10. df_data = pd.DataFrame({
  11.     "f0": data[:, 0],
  12. })
  14. data = dataframeToOperator(df_data, schemaStr='f0 string', op_type="batch")
  16. stringIndexer = StringIndexerTrainBatchOp() \
  17.     .setModelName("string_indexer_model") \
  18.     .setSelectedCol("f0") \
  19.     .setStringOrderType("frequency_asc")
  21. model = stringIndexer.linkFrom(data)
  22. string2int = StringIndexerPredictBatchOp() \
  23.     .setSelectedCol("f0").setOutputCol("f0_indexed")
  24. indexed = string2int.linkFrom(model, data)
  26. predictor = IndexToStringPredictBatchOp().setSelectedCol("f0_indexed").setOutputCol("f0_indxed_unindexed");
  27. predictor.linkFrom(model, indexed).print()

Results

  1. f0|f0_indexed|f0_indxed_unindexed
  2. --|----------|-------------------
  3. football|2|football
  4. football|2|football
  5. football|2|football
  6. basketball|1|basketball
  7. basketball|1|basketball
  8. tennis|0|tennis