Lazily load and return a global instance of ``CharacterSet``.

Create a new instance of ``LabelData`` from integer labels and optionally together with its corresponding timestamps. Args:

`hashlib._Hash` object of the sequence of decoded labels in the path leading to this beam state for efficiently keying into a dict. I

Convert a string to the corresponding sequence of labels after cleaning up and stardardizing. Also see ``str_to_keys()``.

Stable log-sum-exp to sum probabilities in log-space. Ref for example http://gregorygundersen.com/blog/2020/02/09/log-sum-exp/. We could use

Return a normalized string by substituting unsupported characters with supported ones while leaving unicode characters corresponding t

Initialize a new BeamState with empty sequence (CTC blank label), probability of 1 for ending in blank and 0 for non-blank. The label

Fetch the categorical label corresponding to the given key.

Convert a sequence of keys to its corresponding textual representation after standardizing.

Return a normalized AND filtered canonical string corresponding to the ``unicode_str`` such that the characters of the returned string

Returns the next BeamState by extending `prev_state` with `label` and applying LM as appropriate. If `label` is None, we treat that a

r"""Convert a string to the corresponding sequence of supported keys after cleaning up and standardizing. Supports conversion from bo

Offline decoding API that operates over a batch of emission logits. This simply loops over each batch element and calls `decode` in sequence.

Fetch the key corresponding to the given categorical label.

Reset decoder state.

Normalize the given sequence of keys in any representation. NOTE: This doesn NOT filter out out-of-vocabulary keys, directly call `cl

Normalize and filter the given sequence of keys in any representation. Every single key returned is guaranteed to be a member of ``all

Online decoding API that updates decoder state and returns the decoded sequence thus far. Args: emissions (`np.ndarray`):

Return a `pd.Series` consisting of users who have at least `min_sessions` sessions.

Create a new instance of ``LabelData`` from a raw string, after normalizing and cleaning up as per ``CharacterSet``. Args:

Split `df` into train, val and test partitions satisfying the provided per-user constraints.

Fetch the unicode value corresponding to the given key.

Convert a sequence of labels to its corresponding textual representation.

Helper to safely compute the weighted LM score to be added. Handles deletion scenario by undoing the LM score corresponding to the pr

Sample `n` rows per user from `df`.

Convert a unicode value to its corresponding textual representation after standardizing.

Sequence of values on the path from root to this node.

Takes in KenLM state and a token label, and returns a tuple of the next KenLM state on applying the token as well as the LM score. Fo

Convert the HDF5 emg2qwerty session file located at `session_path` to BIDS format and output to `bids_root` directory.

To be called at the end of the sequence to finish any pending LM states by adding end-of-word </s> tokens.

Create a new instance of ``LabelData`` from a single keystroke, after normalizing and cleaning up as per ``CharacterSet``. Args:

Create a new instance of ``LabelData`` from a sequence of keystrokes between the provided start and end timestamps, after normalizing

Create a new instance of ``LabelData`` from a single prompt, after normalizing and cleaning up as per ``CharacterSet``. The returned o

Create a new instance of ``LabelData`` from a sequence of prompts between the provided start and end timestamps, after normalizing and

Read data in HDF5 and get an MNE Raw object with keystrokes and prompts saved as mne.Annotations.

Convert a single key to its corresponding textual representation after standardizing.

Sample `n` users for personalization by giving precedence to those with the most number of sessions.