soeren commited on
Commit
72d08a5
·
1 Parent(s): 13bba81

added annotations

Browse files
.gitignore CHANGED
@@ -4,4 +4,7 @@ Pipfile.lock
4
  data/dataset_audio_annotated_and_embedding_with_probs.parquet.gzip
5
  data/dataset_audio_test.parquet.gzip
6
  data/dataset_audio_train.parquet.gzip
7
- data/dataset_audio_validation.parquet.gzip
 
 
 
 
4
  data/dataset_audio_annotated_and_embedding_with_probs.parquet.gzip
5
  data/dataset_audio_test.parquet.gzip
6
  data/dataset_audio_train.parquet.gzip
7
+ data/dataset_audio_validation.parquet.gzip
8
+
9
+ *.ipynb*
10
+ test.py
data/clip_metadata.py CHANGED
@@ -4,7 +4,7 @@ _SPLIT = "train"
4
 
5
  df = pd.read_parquet("data/dataset_audio_" + _SPLIT +".parquet.gzip")
6
 
7
- clipped_df = df.filter(["Probability", "Predicted Label", "Annotated Labels", "Probability Vector", "embedding_reduced"],
8
- axis=1)
9
 
10
  clipped_df.to_parquet("data/dataset_audio_" + _SPLIT +"_clipped.parquet.gzip")
 
4
 
5
  df = pd.read_parquet("data/dataset_audio_" + _SPLIT +".parquet.gzip")
6
 
7
+ clipped_df = df.filter(["label_string", "probability", "probability_vector", "prediction",
8
+ "prediction_string", "embedding_reduced"], axis=1)
9
 
10
  clipped_df.to_parquet("data/dataset_audio_" + _SPLIT +"_clipped.parquet.gzip")
data/create_enriched_annotated_speechcommands.py CHANGED
@@ -5,8 +5,6 @@
5
  #
6
  # We use the Huggingface transformers library to create an embedding for a an audio dataset
7
  #
8
- #
9
- #
10
 
11
  # ## tldr; Play as callable functions
12
 
@@ -15,12 +13,12 @@
15
  import datasets
16
  from transformers import AutoFeatureExtractor, AutoModel, ASTForAudioClassification
17
  import torch
18
- from renumics import spotlight
19
  import pandas as pd
20
  import umap
21
  import numpy as np
22
 
23
  _SPLIT = "train"
 
24
 
25
  def __set_device():
26
  device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -28,56 +26,47 @@ def __set_device():
28
  torch.cuda.empty_cache()
29
  return device
30
 
31
-
32
- def extract_embeddings(model, feature_extractor):
33
- """Utility to compute embeddings."""
34
- device = model.device
35
-
36
- def pp(batch):
37
- audios = [element["array"] for element in batch["audio"]]
38
- inputs = feature_extractor(raw_speech=audios, return_tensors="pt", padding=True).to(device)
39
- embeddings = model(**inputs).last_hidden_state[:, 0].cpu()
40
-
41
- return {"embedding": embeddings}
42
-
43
-
44
- return pp
45
-
46
-
47
- def huggingface_embedding(dataset, modelname, batched=True, batch_size=8):
48
- # initialize huggingface model
49
- feature_extractor = AutoFeatureExtractor.from_pretrained(modelname, padding=True)
50
- model = AutoModel.from_pretrained(modelname, output_hidden_states=True)
51
-
52
- #compute embedding
53
- device = __set_device()
54
- extract_fn = extract_embeddings(model.to(device), feature_extractor)
55
- updated_dataset = dataset.map(extract_fn, batched=batched, batch_size=batch_size)
56
 
57
- return updated_dataset
58
-
59
-
 
 
 
 
60
 
 
61
 
62
 
63
 
64
- def batch_probabilities(model, feature_extractor):
65
  device = model.device
66
 
67
  def processing(batch):
68
  audios = [element["array"] for element in batch["audio"]]
69
  inputs = feature_extractor(raw_speech=audios, return_tensors="pt", padding=True, sampling_rate=16000).to(device)
70
  outputs = model(**inputs)
71
- return {"logits": outputs.logits}
 
 
72
  return processing
73
 
74
 
75
- def annotate_probabilities(dataset, modelname, batched=True, batch_size= 8):
 
76
  model = ASTForAudioClassification.from_pretrained(modelname)
77
  feature_extractor = AutoFeatureExtractor.from_pretrained(modelname, padding=True)
 
78
 
79
  device = __set_device()
80
- calc_outputs = batch_probabilities (model.to(device), feature_extractor)
81
  output_dataset = dataset.map(calc_outputs, batched = batched, batch_size = batch_size)
82
 
83
  return output_dataset
@@ -94,15 +83,18 @@ def annotate_batch(model, dataset):
94
  logits = [torch.tensor(element) for element in batch["logits"]]
95
  probabilities_per_class = [torch.nn.functional.softmax(logit, dim=-1) for logit in logits]
96
  predicted_class_ids = [torch.argmax(logit).item() for logit in logits]
97
- predicted_labels = [model.config.id2label[predicted_class_id] for predicted_class_id in predicted_class_ids]
98
- # pre-trained model to different amount of classes
99
- # -> id2label only reflects "internal label", not actual dataset label
 
100
  annotated_labels = [labels[element] for element in batch["label"]]
101
  probabilities = []
102
  for index, prob_per_class in enumerate(probabilities_per_class):
103
  probabilities.append(prob_per_class[predicted_class_ids[index]].item())
104
- return {"Probability": probabilities, "Predicted Label": predicted_labels,
105
- "Annotated Labels": annotated_labels, "Probability Vector": probabilities_per_class}
 
 
106
 
107
  return batch_annotation
108
 
@@ -131,6 +123,16 @@ dataset = datasets.load_dataset('speech_commands', 'v0.01', split=_SPLIT)
131
 
132
 
133
  labels = dataset.features["label"].names
 
 
 
 
 
 
 
 
 
 
134
 
135
 
136
  # Let's have a look at all of the labels that we want to predict
@@ -138,23 +140,15 @@ labels = dataset.features["label"].names
138
  print(labels)
139
 
140
 
141
- # ### Compute probabilities and annotate dataset
142
-
143
- # First, calculate logits per sample
144
 
145
 
146
- # calculate logits for each sample and annotate
147
- dataset_annotated = annotate_probabilities(dataset, "MIT/ast-finetuned-speech-commands-v2")
148
 
149
 
150
  # Now annotate labels and probabilities
151
-
152
- dataset_annotated_complete = annotate_dataset(dataset_annotated, "MIT/ast-finetuned-speech-commands-v2")
153
-
154
-
155
- # ### Compute embedding with vision transformer from Huggingface
156
-
157
- dataset_enriched = huggingface_embedding(dataset_annotated_complete, "MIT/ast-finetuned-speech-commands-v2")
158
 
159
 
160
  # ### Reduce embeddings for faster visualization
@@ -165,9 +159,6 @@ reduced_embedding = reducer.fit_transform(embeddings)
165
  dataset_enriched = dataset_enriched.add_column("embedding_reduced", list(reduced_embedding))
166
 
167
 
168
- print(dataset_enriched.features)
169
-
170
-
171
  df = dataset_enriched.to_pandas()
172
 
173
 
 
5
  #
6
  # We use the Huggingface transformers library to create an embedding for a an audio dataset
7
  #
 
 
8
 
9
  # ## tldr; Play as callable functions
10
 
 
13
  import datasets
14
  from transformers import AutoFeatureExtractor, AutoModel, ASTForAudioClassification
15
  import torch
 
16
  import pandas as pd
17
  import umap
18
  import numpy as np
19
 
20
  _SPLIT = "train"
21
+ _MODELNAME = "MIT/ast-finetuned-speech-commands-v2"
22
 
23
  def __set_device():
24
  device = "cuda" if torch.cuda.is_available() else "cpu"
 
26
  torch.cuda.empty_cache()
27
  return device
28
 
29
+ def __permute_model2dataset_probabilities(model_probabilities):
30
+ """
31
+ Model and dataset int mapping is different. Therefore, we need to permute the model's
32
+ logits to match up to the dataset's order.
33
+ Due to small size of vector do this on cpu
34
+ """
35
+ cpu_copy = model_probabilities.to("cpu").detach().numpy()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
36
 
37
+ # only select model2dataset_truncated values as they map model ints to existing dataset ints
38
+ permuted_vector = np.empty(shape=(len(labels),))
39
+ # Problem: model output has more output classes than dataset
40
+ # Hence, only add those outputs that are mapped
41
+ for key in model2dataset_truncated:
42
+ dataset_int = model2dataset_truncated[key]
43
+ permuted_vector[dataset_int] = cpu_copy[key]
44
 
45
+ return permuted_vector
46
 
47
 
48
 
49
+ def batch_probabilities_and_embeddings(model, feature_extractor, classifier):
50
  device = model.device
51
 
52
  def processing(batch):
53
  audios = [element["array"] for element in batch["audio"]]
54
  inputs = feature_extractor(raw_speech=audios, return_tensors="pt", padding=True, sampling_rate=16000).to(device)
55
  outputs = model(**inputs)
56
+
57
+ embeddings = classifier(**inputs).last_hidden_state[:, 0].cpu()
58
+ return {"logits":torch.tensor([__permute_model2dataset_probabilities(logit) for logit in outputs.logits]), "embedding": embeddings}
59
  return processing
60
 
61
 
62
+ def annotate_probabilities_and_embeddings(dataset, modelname, batched=True, batch_size= 8):
63
+
64
  model = ASTForAudioClassification.from_pretrained(modelname)
65
  feature_extractor = AutoFeatureExtractor.from_pretrained(modelname, padding=True)
66
+ classifier = AutoModel.from_pretrained(modelname, output_hidden_states=True)
67
 
68
  device = __set_device()
69
+ calc_outputs = batch_probabilities_and_embeddings(model.to(device), feature_extractor, classifier.to(device))
70
  output_dataset = dataset.map(calc_outputs, batched = batched, batch_size = batch_size)
71
 
72
  return output_dataset
 
83
  logits = [torch.tensor(element) for element in batch["logits"]]
84
  probabilities_per_class = [torch.nn.functional.softmax(logit, dim=-1) for logit in logits]
85
  predicted_class_ids = [torch.argmax(logit).item() for logit in logits]
86
+
87
+ # logits are already permuted to match dataset ordering -> no additional work needed
88
+
89
+ predicted_labels = [labels[predicted_class_id] for predicted_class_id in predicted_class_ids]
90
  annotated_labels = [labels[element] for element in batch["label"]]
91
  probabilities = []
92
  for index, prob_per_class in enumerate(probabilities_per_class):
93
  probabilities.append(prob_per_class[predicted_class_ids[index]].item())
94
+ return {"label_string": annotated_labels, "probability": probabilities,
95
+ "probability_vector": probabilities_per_class,
96
+ "prediction": predicted_class_ids,
97
+ "prediction_string": predicted_labels}
98
 
99
  return batch_annotation
100
 
 
123
 
124
 
125
  labels = dataset.features["label"].names
126
+ # use dict comprehension to build a dict that maps the dataset's string label to the dataset's int label
127
+ label_dict = {label: i for label, i in zip(labels, range(len(labels)))}
128
+ # {key: value for key, value in zip(keys, values)}
129
+ model = ASTForAudioClassification.from_pretrained(_MODELNAME)
130
+ # look up label from model int in the label dict and translate that label into the dataset int
131
+ model2dataset_int_conversion = {i: label_dict[model.config.id2label[i]] if model.config.id2label[i] in label_dict.keys()
132
+ else -1 for i in range(model.config.num_labels)}
133
+
134
+ model2dataset_truncated = {key:value for key,value in model2dataset_int_conversion.items() if value != -1}
135
+
136
 
137
 
138
  # Let's have a look at all of the labels that we want to predict
 
140
  print(labels)
141
 
142
 
143
+ # ### Compute probabilities and embeddings and annotate dataset
 
 
144
 
145
 
146
+ # calculate logits and embedding for each sample and annotate
147
+ dataset_annotated = annotate_probabilities_and_embeddings(dataset, _MODELNAME)
148
 
149
 
150
  # Now annotate labels and probabilities
151
+ dataset_enriched = annotate_dataset(dataset_annotated, _MODELNAME)
 
 
 
 
 
 
152
 
153
 
154
  # ### Reduce embeddings for faster visualization
 
159
  dataset_enriched = dataset_enriched.add_column("embedding_reduced", list(reduced_embedding))
160
 
161
 
 
 
 
162
  df = dataset_enriched.to_pandas()
163
 
164
 
data/dataset_audio_test_clipped.parquet.gzip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d8b6ae3710263e5e83c9f6784e1945b15a97bde88a6b7d1b45587b0d6278345c
3
- size 698777
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:334324791b5c45096971f7bbf6a05afa60160d9ff5a9fc966762b9c2923cff02
3
+ size 642187
data/dataset_audio_train_clipped.parquet.gzip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e8e208d54feef82c4a7e1f120787e4b1074cf7cad75035e3640714636965e84f
3
- size 8525172
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3fed567b3d361fcfff79ad59a2d45233ef663d965ef17a4d9f42d3bc9662693
3
+ size 7864308
data/dataset_audio_validation_clipped.parquet.gzip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:390eaf7ec5fdc8a832651b6255948aeba7291ae1b2afc5de73229aa25c48a73c
3
- size 1566541
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0dfba31f1d0587ebb4b9cbeb96fd10ae93cac9036dc482f4802b1495ab6806c
3
+ size 1443196
speech_commands_enriched.py CHANGED
@@ -168,10 +168,11 @@ class SpeechCommands(datasets.GeneratorBasedBuilder):
168
  "speaker_id": datasets.Value("string"),
169
  "utterance_id": datasets.Value("int8"),
170
  #enriched features:
171
- "Probability": datasets.Value("float64"),
172
- "Predicted Label": datasets.Value("string"),
173
- "Annotated Labels": datasets.Value("string"),
174
- "Probability Vector": datasets.Sequence(feature=datasets.Value("float64"), length=35),
 
175
  "embedding_reduced": datasets.Sequence(feature=datasets.Value("float32"), length=2),
176
  }
177
  ),
@@ -258,14 +259,15 @@ class SpeechCommands(datasets.GeneratorBasedBuilder):
258
  "speaker_id": speaker_id,
259
  "utterance_id": utterance_id,
260
  #enriched features:
261
- "Probability": row[1]["Probability"],
262
- "Predicted Label": row[1]["Predicted Label"],
263
- "Annotated Labels": row[1]["Annotated Labels"],
264
- "Probability Vector": row[1]["Probability Vector"],
 
265
  "embedding_reduced": row[1]["embedding_reduced"]
266
  }
267
 
268
  #for debugging, comment out after
269
- #if __name__ == "__main__":
270
- #ds = datasets.load_dataset("speech_commands_enriched.py", 'v0.01', split="test",
271
- # streaming=False)
 
168
  "speaker_id": datasets.Value("string"),
169
  "utterance_id": datasets.Value("int8"),
170
  #enriched features:
171
+ "label_string": datasets.Value("string"),
172
+ "probability": datasets.Value("float64"),
173
+ "probability_vector": datasets.Sequence(feature=datasets.Value("float64"), length=31),
174
+ "prediction": datasets.ClassLabel(names=self.config.labels),
175
+ "prediction_string":datasets.Value("string"),
176
  "embedding_reduced": datasets.Sequence(feature=datasets.Value("float32"), length=2),
177
  }
178
  ),
 
259
  "speaker_id": speaker_id,
260
  "utterance_id": utterance_id,
261
  #enriched features:
262
+ "label_string": row[1]["label_string"],
263
+ "probability": row[1]["probability"],
264
+ "probability_vector": row[1]["probability_vector"],
265
+ "prediction": row[1]["prediction"],
266
+ "prediction_string": row[1]["prediction_string"],
267
  "embedding_reduced": row[1]["embedding_reduced"]
268
  }
269
 
270
  #for debugging, comment out after
271
+ if __name__ == "__main__":
272
+ ds = datasets.load_dataset("speech_commands_enriched.py", 'v0.01', split="test",
273
+ streaming=False)