Datasets:

ArXiv:
DOI:
License:
zhuwq0 kylewhy commited on
Commit
2fa7223
1 Parent(s): 8a1a19e

add batched configuration (#3)

Browse files

- add batched configuration (fa3c8cdd117884e5405801754ddea6fbc0845399)
- remove batched operation and leave it to sampler (4fd86c8e2af7da1cf379391d953769a14e16cac8)


Co-authored-by: kylewhy <kylewhy@users.noreply.huggingface.co>

Files changed (2) hide show
  1. README.md +79 -6
  2. quakeflow_nc.py +68 -37
README.md CHANGED
@@ -82,36 +82,63 @@ Group: / len:10000
82
  - torch (for PyTorch)
83
 
84
  ### Usage
 
85
  ```python
86
  import h5py
87
  import numpy as np
88
  import torch
89
  from torch.utils.data import Dataset, IterableDataset, DataLoader
90
  from datasets import load_dataset
 
 
 
 
 
 
 
 
 
 
 
 
 
 
91
 
 
 
92
  # load dataset
93
  # ATTENTION: Streaming(Iterable Dataset) is difficult to support because of the feature of HDF5
94
  # So we recommend to directly load the dataset and convert it into iterable later
95
  # The dataset is very large, so you need to wait for some time at the first time
96
- quakeflow_nc = datasets.load_dataset("AI4EPS/quakeflow_nc", split="train")
97
- quakeflow_nc
 
 
 
 
 
 
98
  ```
99
- If you want to use the first several shards of the dataset, you can download the script `quakeflow_nc.py` and change the code below:
 
100
  ```python
101
  # change the 37 to the number of shards you want
102
  _URLS = {
103
  "NCEDC": [f"{_REPO}/ncedc_event_dataset_{i:03d}.h5" for i in range(37)]
104
  }
105
  ```
106
- Then you can use the dataset like this:
107
  ```python
108
- quakeflow_nc = datasets.load_dataset("./quakeflow_nc.py", split="train")
 
109
  quakeflow_nc
110
  ```
 
 
111
  Then you can change the dataset into PyTorch format iterable dataset, and view the first sample:
112
  ```python
 
113
  quakeflow_nc = quakeflow_nc.to_iterable_dataset()
114
- quakeflow_nc = quakeflow_nc.with_format("torch")
115
  # because add examples formatting to get tensors when using the "torch" format
116
  # has not been implemented yet, we need to manually add the formatting
117
  quakeflow_nc = quakeflow_nc.map(lambda x: {key: torch.from_numpy(np.array(value, dtype=np.float32)) for key, value in x.items()})
@@ -136,4 +163,50 @@ for batch in dataloader:
136
  for key in batch.keys():
137
  print(key, batch[key].shape, batch[key].dtype)
138
  break
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
  ```
 
82
  - torch (for PyTorch)
83
 
84
  ### Usage
85
+ Import the necessary packages:
86
  ```python
87
  import h5py
88
  import numpy as np
89
  import torch
90
  from torch.utils.data import Dataset, IterableDataset, DataLoader
91
  from datasets import load_dataset
92
+ ```
93
+ We have 2 configurations for the dataset: `NCEDC` and `NCEDC_full_size`. They all return event-based samples one by one. But `NCEDC` returns samples with 10 stations each, while `NCEDC_full_size` return samples with stations same as the original data.
94
+
95
+ The sample of `NCEDC` is a dictionary with the following keys:
96
+ - `waveform`: the waveform with shape `(3, nt, n_sta)`, the first dimension is 3 components, the second dimension is the number of time samples, the third dimension is the number of stations
97
+ - `phase_pick`: the probability of the phase pick with shape `(3, nt, n_sta)`, the first dimension is noise, P and S
98
+ - `event_location`: the event location with shape `(4,)`, including latitude, longitude, depth and time
99
+ - `station_location`: the station location with shape `(n_sta, 3)`, the first dimension is latitude, longitude and depth
100
+
101
+ Because Huggingface datasets only support dynamic size on first dimension, so the sample of `NCEDC_full_size` is a dictionary with the following keys:
102
+ - `waveform`: the waveform with shape `(n_sta, 3, nt)`,
103
+ - `phase_pick`: the probability of the phase pick with shape `(n_sta, 3, nt)`
104
+ - `event_location`: the event location with shape `(4,)`
105
+ - `station_location`: the station location with shape `(n_sta, 3)`, the first dimension is latitude, longitude and depth
106
 
107
+ The default configuration is `NCEDC`. You can specify the configuration by argument `name`. For example:
108
+ ```python
109
  # load dataset
110
  # ATTENTION: Streaming(Iterable Dataset) is difficult to support because of the feature of HDF5
111
  # So we recommend to directly load the dataset and convert it into iterable later
112
  # The dataset is very large, so you need to wait for some time at the first time
113
+
114
+ # to load "NCEDC"
115
+ quakeflow_nc = load_dataset("AI4EPS/quakeflow_nc", split="train")
116
+ # or
117
+ quakeflow_nc = load_dataset("AI4EPS/quakeflow_nc", name="NCEDC", split="train")
118
+
119
+ # to load "NCEDC_full_size"
120
+ quakeflow_nc = load_dataset("AI4EPS/quakeflow_nc", name="NCEDC_full_size", split="train")
121
  ```
122
+
123
+ If you want to use the first several shards of the dataset, you can download the script `quakeflow_nc.py` and change the code as below:
124
  ```python
125
  # change the 37 to the number of shards you want
126
  _URLS = {
127
  "NCEDC": [f"{_REPO}/ncedc_event_dataset_{i:03d}.h5" for i in range(37)]
128
  }
129
  ```
130
+ Then you can use the dataset like this (Don't forget to specify the argument `name`):
131
  ```python
132
+ # don't forget to specify the script path
133
+ quakeflow_nc = datasets.load_dataset("path_to_script/quakeflow_nc.py", split="train")
134
  quakeflow_nc
135
  ```
136
+
137
+ #### Usage for `NCEDC`
138
  Then you can change the dataset into PyTorch format iterable dataset, and view the first sample:
139
  ```python
140
+ quakeflow_nc = load_dataset("AI4EPS/quakeflow_nc", name="NCEDC", split="train")
141
  quakeflow_nc = quakeflow_nc.to_iterable_dataset()
 
142
  # because add examples formatting to get tensors when using the "torch" format
143
  # has not been implemented yet, we need to manually add the formatting
144
  quakeflow_nc = quakeflow_nc.map(lambda x: {key: torch.from_numpy(np.array(value, dtype=np.float32)) for key, value in x.items()})
 
163
  for key in batch.keys():
164
  print(key, batch[key].shape, batch[key].dtype)
165
  break
166
+ ```
167
+
168
+ #### Usage for `NCEDC_full_size`
169
+
170
+ Then you can change the dataset into PyTorch format dataset, and view the first sample (Don't forget to reorder the keys):
171
+ ```python
172
+ quakeflow_nc = datasets.load_dataset("AI4EPS/quakeflow_nc", split="train", name="NCEDC_full_size")
173
+
174
+ # for PyTorch DataLoader, we need to divide the dataset into several shards
175
+ num_workers=4
176
+ quakeflow_nc = quakeflow_nc.to_iterable_dataset(num_shards=num_workers)
177
+ # because add examples formatting to get tensors when using the "torch" format
178
+ # has not been implemented yet, we need to manually add the formatting
179
+ quakeflow_nc = quakeflow_nc.map(lambda x: {key: torch.from_numpy(np.array(value, dtype=np.float32)) for key, value in x.items()})
180
+ def reorder_keys(example):
181
+ example["waveform"] = example["waveform"].permute(1,2,0).contiguous()
182
+ example["phase_pick"] = example["phase_pick"].permute(1,2,0).contiguous()
183
+ return example
184
+
185
+ quakeflow_nc = quakeflow_nc.map(reorder_keys)
186
+
187
+ try:
188
+ isinstance(quakeflow_nc, torch.utils.data.IterableDataset)
189
+ except:
190
+ raise Exception("quakeflow_nc is not an IterableDataset")
191
+
192
+ data_loader = DataLoader(
193
+ quakeflow_nc,
194
+ batch_size=1,
195
+ num_workers=num_workers,
196
+ )
197
+
198
+ for batch in quakeflow_nc:
199
+ print("\nIterable test\n")
200
+ print(batch.keys())
201
+ for key in batch.keys():
202
+ print(key, batch[key].shape, batch[key].dtype)
203
+ break
204
+
205
+ for batch in data_loader:
206
+ print("\nDataloader test\n")
207
+ print(batch.keys())
208
+ for key in batch.keys():
209
+ batch[key] = batch[key].squeeze(0)
210
+ print(key, batch[key].shape, batch[key].dtype)
211
+ break
212
  ```
quakeflow_nc.py CHANGED
@@ -17,14 +17,9 @@
17
  """QuakeFlow_NC: A dataset of earthquake waveforms organized by earthquake events and based on the HDF5 format."""
18
 
19
 
20
- import csv
21
- import json
22
- import os
23
  import h5py
24
  import numpy as np
25
  import torch
26
- import fsspec
27
- from glob import glob
28
  from typing import Dict, List, Optional, Tuple, Union
29
 
30
  import datasets
@@ -58,16 +53,36 @@ _LICENSE = ""
58
  # This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
59
  _REPO = "https://huggingface.co/datasets/AI4EPS/quakeflow_nc/resolve/main/data"
60
  _URLS = {
61
- "NCEDC": [f"{_REPO}/ncedc_event_dataset_{i:03d}.h5" for i in range(37)]
 
62
  }
63
 
 
 
 
 
 
 
 
 
 
 
 
 
64
 
65
  # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
66
  class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
67
  """QuakeFlow_NC: A dataset of earthquake waveforms organized by earthquake events and based on the HDF5 format."""
68
-
69
  VERSION = datasets.Version("1.1.0")
70
 
 
 
 
 
 
 
 
71
  # This is an example of a dataset with multiple configurations.
72
  # If you don't want/need to define several sub-sets in your dataset,
73
  # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
@@ -79,22 +94,36 @@ class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
79
  # You will be able to load one or the other configurations in the following list with
80
  # data = datasets.load_dataset('my_dataset', 'first_domain')
81
  # data = datasets.load_dataset('my_dataset', 'second_domain')
 
 
82
  BUILDER_CONFIGS = [
83
- datasets.BuilderConfig(name="NCEDC", version=VERSION, description="This part of my dataset covers a first domain"),
 
84
  ]
85
 
86
  DEFAULT_CONFIG_NAME = "NCEDC" # It's not mandatory to have a default configuration. Just use one if it make sense.
87
 
88
  def _info(self):
89
  # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
90
- features=datasets.Features(
91
- {
92
- "waveform": datasets.Array3D(shape=(3, self.nt, self.num_stations), dtype='float32'),
93
- "phase_pick": datasets.Array3D(shape=(3, self.nt, self.num_stations), dtype='float32'),
94
- "event_location": [datasets.Value("float32")],
95
- "station_location": datasets.Array2D(shape=(self.num_stations, 3), dtype="float32"),
96
- }
97
- )
 
 
 
 
 
 
 
 
 
 
 
98
  return datasets.DatasetInfo(
99
  # This is the description that will appear on the datasets page.
100
  description=_DESCRIPTION,
@@ -150,30 +179,26 @@ class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
150
  # ),
151
  ]
152
 
153
- degree2km = 111.32
154
- nt = 8192
155
- feature_nt = 512
156
- feature_scale = int(nt / feature_nt)
157
- sampling_rate=100.0
158
- num_stations = 10
159
 
160
  # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
161
  def _generate_examples(self, filepath, split):
162
  # TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
163
  # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
164
  num_stations = self.num_stations
165
-
166
  for file in filepath:
167
  with h5py.File(file, "r") as fp:
168
  # for event_id in sorted(list(fp.keys())):
169
  for event_id in fp.keys():
170
  event = fp[event_id]
171
  station_ids = list(event.keys())
172
- if len(station_ids) < num_stations:
173
- continue
174
- else:
175
- station_ids = np.random.choice(station_ids, num_stations, replace=False)
176
-
 
 
177
  waveforms = np.zeros([3, self.nt, len(station_ids)])
178
  phase_pick = np.zeros_like(waveforms)
179
  attrs = event.attrs
@@ -182,13 +207,11 @@ class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
182
 
183
  for i, sta_id in enumerate(station_ids):
184
  # trace_id = event_id + "/" + sta_id
185
-
186
  waveforms[:, :, i] = event[sta_id][:,:self.nt]
187
  attrs = event[sta_id].attrs
188
  p_picks = attrs["phase_index"][attrs["phase_type"] == "P"]
189
  s_picks = attrs["phase_index"][attrs["phase_type"] == "S"]
190
  phase_pick[:, :, i] = generate_label([p_picks, s_picks], nt=self.nt)
191
-
192
  station_location.append([attrs["longitude"], attrs["latitude"], -attrs["elevation_m"]/1e3])
193
 
194
  std = np.std(waveforms, axis=1, keepdims=True)
@@ -196,13 +219,21 @@ class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
196
  waveforms = (waveforms - np.mean(waveforms, axis=1, keepdims=True)) / std
197
  waveforms = waveforms.astype(np.float32)
198
 
199
- yield event_id, {
200
- "waveform": torch.from_numpy(waveforms).float(),
201
- "phase_pick": torch.from_numpy(phase_pick).float(),
202
- "event_location": event_location,
203
- "station_location": station_location,
204
- }
205
-
 
 
 
 
 
 
 
 
206
 
207
 
208
  def generate_label(phase_list, label_width=[150, 150], nt=8192):
 
17
  """QuakeFlow_NC: A dataset of earthquake waveforms organized by earthquake events and based on the HDF5 format."""
18
 
19
 
 
 
 
20
  import h5py
21
  import numpy as np
22
  import torch
 
 
23
  from typing import Dict, List, Optional, Tuple, Union
24
 
25
  import datasets
 
53
  # This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
54
  _REPO = "https://huggingface.co/datasets/AI4EPS/quakeflow_nc/resolve/main/data"
55
  _URLS = {
56
+ "NCEDC": [f"{_REPO}/ncedc_event_dataset_{i:03d}.h5" for i in range(37)],
57
+ "NCEDC_full_size": [f"{_REPO}/ncedc_event_dataset_{i:03d}.h5" for i in range(37)],
58
  }
59
 
60
+ class BatchBuilderConfig(datasets.BuilderConfig):
61
+ """
62
+ yield a batch of event-based sample, so the number of sample stations can vary among batches
63
+ Batch Config for QuakeFlow_NC
64
+ :param batch_size: number of samples in a batch
65
+ :param num_stations_list: possible number of stations in a batch
66
+ """
67
+ def __init__(self, batch_size: int, num_stations_list: List, **kwargs):
68
+ super().__init__(**kwargs)
69
+ self.batch_size = batch_size
70
+ self.num_stations_list = num_stations_list
71
+
72
 
73
  # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
74
  class QuakeFlow_NC(datasets.GeneratorBasedBuilder):
75
  """QuakeFlow_NC: A dataset of earthquake waveforms organized by earthquake events and based on the HDF5 format."""
76
+
77
  VERSION = datasets.Version("1.1.0")
78
 
79
+ degree2km = 111.32
80
+ nt = 8192
81
+ feature_nt = 512
82
+ feature_scale = int(nt / feature_nt)
83
+ sampling_rate=100.0
84
+ num_stations = 10
85
+
86
  # This is an example of a dataset with multiple configurations.
87
  # If you don't want/need to define several sub-sets in your dataset,
88
  # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
 
94
  # You will be able to load one or the other configurations in the following list with
95
  # data = datasets.load_dataset('my_dataset', 'first_domain')
96
  # data = datasets.load_dataset('my_dataset', 'second_domain')
97
+
98
+ # default config, you can change batch_size and num_stations_list when use `datasets.load_dataset`
99
  BUILDER_CONFIGS = [
100
+ datasets.BuilderConfig(name="NCEDC", version=VERSION, description="yield event-based samples one by one, the number of sample stations is fixed(default: 10)"),
101
+ datasets.BuilderConfig(name="NCEDC_full_size", version=VERSION, description="yield event-based samples one by one, the number of sample stations is the same as the number of stations in the event"),
102
  ]
103
 
104
  DEFAULT_CONFIG_NAME = "NCEDC" # It's not mandatory to have a default configuration. Just use one if it make sense.
105
 
106
  def _info(self):
107
  # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
108
+ if self.config.name=="NCEDC":
109
+ features=datasets.Features(
110
+ {
111
+ "waveform": datasets.Array3D(shape=(3, self.nt, self.num_stations), dtype='float32'),
112
+ "phase_pick": datasets.Array3D(shape=(3, self.nt, self.num_stations), dtype='float32'),
113
+ "event_location": datasets.Sequence(datasets.Value("float32")),
114
+ "station_location": datasets.Array2D(shape=(self.num_stations, 3), dtype="float32"),
115
+ })
116
+
117
+ elif self.config.name=="NCEDC_full_size":
118
+ features=datasets.Features(
119
+ {
120
+ "waveform": datasets.Array3D(shape=(None, 3, self.nt), dtype='float32'),
121
+ "phase_pick": datasets.Array3D(shape=(None, 3, self.nt), dtype='float32'),
122
+ "event_location": datasets.Sequence(datasets.Value("float32")),
123
+ "station_location": datasets.Array2D(shape=(None, 3), dtype="float32"),
124
+ }
125
+ )
126
+
127
  return datasets.DatasetInfo(
128
  # This is the description that will appear on the datasets page.
129
  description=_DESCRIPTION,
 
179
  # ),
180
  ]
181
 
182
+
 
 
 
 
 
183
 
184
  # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
185
  def _generate_examples(self, filepath, split):
186
  # TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
187
  # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
188
  num_stations = self.num_stations
 
189
  for file in filepath:
190
  with h5py.File(file, "r") as fp:
191
  # for event_id in sorted(list(fp.keys())):
192
  for event_id in fp.keys():
193
  event = fp[event_id]
194
  station_ids = list(event.keys())
195
+
196
+ if self.config.name=="NCEDC":
197
+ if len(station_ids) < num_stations:
198
+ continue
199
+ else:
200
+ station_ids = np.random.choice(station_ids, num_stations, replace=False)
201
+
202
  waveforms = np.zeros([3, self.nt, len(station_ids)])
203
  phase_pick = np.zeros_like(waveforms)
204
  attrs = event.attrs
 
207
 
208
  for i, sta_id in enumerate(station_ids):
209
  # trace_id = event_id + "/" + sta_id
 
210
  waveforms[:, :, i] = event[sta_id][:,:self.nt]
211
  attrs = event[sta_id].attrs
212
  p_picks = attrs["phase_index"][attrs["phase_type"] == "P"]
213
  s_picks = attrs["phase_index"][attrs["phase_type"] == "S"]
214
  phase_pick[:, :, i] = generate_label([p_picks, s_picks], nt=self.nt)
 
215
  station_location.append([attrs["longitude"], attrs["latitude"], -attrs["elevation_m"]/1e3])
216
 
217
  std = np.std(waveforms, axis=1, keepdims=True)
 
219
  waveforms = (waveforms - np.mean(waveforms, axis=1, keepdims=True)) / std
220
  waveforms = waveforms.astype(np.float32)
221
 
222
+ if self.config.name=="NCEDC":
223
+ yield event_id, {
224
+ "waveform": torch.from_numpy(waveforms).float(),
225
+ "phase_pick": torch.from_numpy(phase_pick).float(),
226
+ "event_location": torch.from_numpy(np.array(event_location)).float(),
227
+ "station_location": torch.from_numpy(np.array(station_location)).float(),
228
+ }
229
+ elif self.config.name=="NCEDC_full_size":
230
+
231
+ yield event_id, {
232
+ "waveform": torch.from_numpy(waveforms).float().permute(2,0,1),
233
+ "phase_pick": torch.from_numpy(phase_pick).float().permute(2,0,1),
234
+ "event_location": torch.from_numpy(np.array(event_location)).float(),
235
+ "station_location": torch.from_numpy(np.array(station_location)).float(),
236
+ }
237
 
238
 
239
  def generate_label(phase_list, label_width=[150, 150], nt=8192):