das.make_dataset

das.make_dataset.blur_events(event_trace: numpy.ndarray, event_std_seconds: float, samplerate: float) → numpy.ndarray

Blur event trace with a gaussian.

Parameters

• event_trace (np.ndarray) – shape (N,)

• event_std_seconds (float) – With of the Gaussian in seconds

• samplerate (float) – sample rate of event_trace

Returns

blurred event trace

Return type

np.ndarray

das.make_dataset.events_to_probabilities(eventsamples: List[int], desired_len: Optional[int] = None, extent: int = 61)

Converts list of events to one-hot-encoded probability vectors.

Parameters

• eventsamples (List[int]) – List of event “times” in samples.

• desired_len (float, optional) – Length of the probability vector. Events exceeding desired_len will be ignored. Defaults to max(eventsamples) + extent.

• extent (int, optional) – Temporal extent of an event in the probability vector. Each event will be represented as a box with a duration exent samples centered on the event. Defaults to 61 samples (+/-30 samples).

Returns

np.array with shape [desired_len, 2]

where probabilities[:, 0] corresponds to the probability of no event and probabilities[:, 0] corresponds to the probability of an event.

Return type

probabilities

das.make_dataset.infer_class_info(df: pandas.core.frame.DataFrame)

[summary]

Parameters

df ([type]) – [description]

Returns

[description]

Return type

[type]

das.make_dataset.init_store(nb_channels: int, nb_classes: int, store, samplerate: Optional[float] = None, make_single_class_datasets: bool = False, class_names: List[str] = None, class_types: List[str] = None, chunk_len: int = 1000000)

[summary]

Parameters

• nb_channels (int) – [description]

• nb_classes (int) – [description] <- should infer from class_names!

• store – zarr store

• samplerate (float, optional) – [description]. Defaults to None.

• make_single_class_datasets (bool, optional) – make y_suffix and attrs[‘class_names/types_suffix’]. Defaults to None.

• class_names (List[str], optional) – [description]. Defaults to None.

• class_types (List[str], optional) – ‘event’ or ‘segment’. Defaults to None.

• chunk_len (int, optional) – [description]. Defaults to 1_000_000.

Raises

• ValueError – [description]

• ValueError – [description]

Returns

[description]

Return type

[type]

das.make_dataset.make_annotation_matrix(df: pandas.core.frame.DataFrame, nb_samples: int, samplerate: float, class_names: Optional[List[str]] = None) → numpy.ndarray

One-hot encode a list of song timings to a binary matrix.

Parameters

• df (pd.DataFrame) – DataFrame with the following columns: - name: class name of the syllable/song event - start_seconds: start of the song event in the audio recording in seconds. - stop_seconds: stop of the song event in the audio recording in seconds.

• nb_samples ([type]) – Length of the annotation matrix in samples.

• samplerate (float) – Sample rate for the annotation matrix in Hz.

• class_names (List[str], optional) – List of class names. If provided, the annotation matrix will be built only for the events in class_names. Otherwise, the matrix will be build for all class names in the df. Order in class_names determines order in class_matrix

Returns

Binary matrix [nb_samples, nb_classes]

with 1 indicating the presence of a class at a specific sample.

Return type

nd.array

das.make_dataset.make_gaps(y: numpy.ndarray, gap_seconds: float, samplerate: float, start_seconds: Optional[List[float]] = None, stop_seconds: Optional[List[float]] = None) → numpy.ndarray

[summary]

0011112222000111100 -> 0011100222000111100 (gap_fullwidth=2)

Parameters

• y (np.ndarray) – One-hot encoded labels [T, nb_labels]

• gap_seconds (float) – [description]

• samplerate (float) – [description]

• start_seconds –

• stop_seconds –

Returns

[description]

Return type

np.ndarray

das.make_dataset.normalize_probabilities(p: numpy.ndarray) → numpy.ndarray

[summary]

Parameters

p (np.ndarray) – [description]

Returns

[description]

Return type

np.ndarray