updated readme

README.md

@@ -154,6 +154,10 @@ a ClassLabel for the label and a ClassLabel for the class.
  'domain': 0 (source)
  'label': 0 (normal)
  'class': 1 (fan)
+ 'dev_train_lof_anomaly': 0
+ 'dev_train_lof_anomaly_score': 1.241023
+ 'add_train_lof_anomaly': 1
+ 'add_train_lof_anomaly_score': 1.806289
  'ast-finetuned-audioset-10-10-0.4593-embeddings': [0.8152204155921936,
    1.5862374305725098, ...,   
    1.7154160737991333]
@@ -174,8 +178,9 @@ The length of each audio file is 10 seconds.
 - `domain`: an integer whose value may be either _0_, indicating that the audio sample is from the _source_ domain, _1_, indicating that the audio sample is from the _target_.
 - `class`: an integer as class label.
 - `label`: an integer whose value may be either _0_, indicating that the audio sample is _normal_, _1_, indicating that the audio sample contains an _anomaly_.
+- '*_lof_anomaly': an integer as anomaly indicator. The anomaly prediction is computed with the [Local Outlier Factor](https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.LocalOutlierFactor.html) algorithm based on the "*"-dataset.
+- '*_lof_anomaly_score': a float as anomaly score. The anomaly score is computed with the [Local Outlier Factor](https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.LocalOutlierFactor.html) algorithm based on the "*"-dataset.
 - `embeddings_ast-finetuned-audioset-10-10-0.4593`: an `datasets.Sequence(Value("float32"), shape=(1, 768))` representing audio embeddings that are generated with an [Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer#transformers.ASTFeatureExtractor).
-- `embeddings_dcase2023_task2_baseline_ae`: an `datasets.Sequence(Value("float32"), shape=(1, 512))` representing audio embeddings that are generated with the [**DCASE 2023 Challenge Task 2 Baseline Auto Encoder**](https://github.com/nttcslab/dcase2023_task2_baseline_ae). **Seven individual class-specific AEs** are trained. Dimensionality Reduction is applied with **PCA** separately for each class with a fit on the respecting training set of samples.
 
 ### Data Splits
 
@@ -217,6 +222,16 @@ The data consists of the normal/anomalous operating sounds of seven types of rea
 - Slide rail
 - Valve
 
+The "additional training data" and "evaluation data" datasets contain the following classes:
+
+- bandsaw
+- grinder
+- shaker
+- ToyDrone 
+- ToyNscale
+- ToyTank
+- Vacuum
+
 ### Source Data
 
 #### Definition