medcat.config.config_rel_cat

Classes:

• ConfigRelCAT –

  The RelCAT part of the config

• General –

  The General part of the RelCAT config

• Model –

  The model part of the RelCAT config

• Train –

  The train part of the RelCAT config

ConfigRelCAT

Bases: ComponentConfig

The RelCAT part of the config

Methods:

• load –

  Load the config from a file.

Attributes:

• general (General) –
• model (Model) –
• model_config –
• train (Train) –

general class-attribute instance-attribute

general: General = General()

model class-attribute instance-attribute

model: Model = Model()

model_config class-attribute instance-attribute

model_config = ConfigDict(extra='allow', validate_assignment=True)

train class-attribute instance-attribute

train: Train = Train()

load classmethod

load(load_path: str = './') -> ConfigRelCAT

Load the config from a file.

Parameters:

• load_path

  (str, default: './' ) –

  Path to RelCAT config. Defaults to "./".

Returns:

• ConfigRelCAT ( ConfigRelCAT ) –

  The loaded config.

Source code in medcat-v2/medcat/config/config_rel_cat.py

@classmethod
def load(cls, load_path: str = "./") -> "ConfigRelCAT":
    """Load the config from a file.

    Args:
        load_path (str): Path to RelCAT config. Defaults to "./".

    Returns:
        ConfigRelCAT: The loaded config.
    """
    config = cls()
    if os.path.exists(load_path):
        if os.path.sep + "config" not in load_path:
            load_path = os.path.join(load_path, "config")
        config: ConfigRelCAT = deserialise(load_path)  # type: ignore
        logging.info("Loaded config.json")

    return config

General

Bases: SerialisableBaseModel

The General part of the RelCAT config

Methods:

• convert_keys_to_int –

Attributes:

• addl_rels_max_sample_size (int) –

  Limit the number of 'Other' samples selected for training/test. This is

• annotation_schema_tag_ids (list[int]) –

  If a foreign non-MCAT trainer dataset is used, you can insert your own

• cntx_left (int) –

  Number of tokens to take from the left of the concept

• cntx_right (int) –

  Number of tokens to take from the right of the concept

• create_addl_rels (bool) –

  When processing relations from a MedCAT export/docs, relations labeled

• create_addl_rels_by_type (bool) –

  When creating the 'Other' relation class, actually split this class

• device (str) –

  The device to use (CPU or GPU).

• idx2labels (dict[int, str]) –
• labels2idx (dict[str, int]) –
• language (str) –

  Used for Spacy lang setting

• limit_samples_per_class (int) –

  Number of samples per class, this limit is applied for train samples,

• log_level (int) –

  The log level for RelCAT.

• lowercase (bool) –

  If true all input text will be lowercased

• max_seq_length (int) –

  The maximum sequence length.

• model_config –
• model_name (str) –

  The name of the model used.

• pin_memory (bool) –

  If True the data loader will copy the tensors to the GPU pinned memory.

• relation_type_filter_pairs (list) –

  Map from category values to ID, if empty it will be

• seed (int) –

  The seed for random number generation.

• task (str) –

  The task for RelCAT.

• tokenizer_name (str) –

  The name of the tokenizer user.

• tokenizer_other_special_tokens (dict[str, str]) –

  The special tokens used by the tokenizer. The {PAD} is

• tokenizer_relation_annotation_special_tokens_tags (list[str]) –
• tokenizer_special_tokens (bool) –

  Tokenizer.

• vocab_size (Optional[int]) –
• window_size (int) –

  Max acceptable dinstance between entities (in characters),

addl_rels_max_sample_size class-attribute instance-attribute

addl_rels_max_sample_size: int = 200

Limit the number of 'Other' samples selected for training/test. This is applied per encountered medcat project, sample_size/num_projects.

annotation_schema_tag_ids class-attribute instance-attribute

annotation_schema_tag_ids: list[int] = [30522, 30523, 30524, 30525]

If a foreign non-MCAT trainer dataset is used, you can insert your own Rel entity token delimiters into the tokenizer, copy those token IDs here, and also resize your tokenizer embeddings and adjust the hidden_size of the model, this will depend on the number of tokens you introduce for example: 30522 - [s1], 30523 - [e1], 30524 - [s2], 30525 - [e2], 30526 - [BLANK], 30527 - [ENT1], 30528 - [ENT2], 30529 - [/ENT1], 30530 - [/ENT2] Please note that the tokenizer special tokens are supposed to be in pairs of two for example [s1] and [e1], [s2] and [e2], the [BLANK] is just an example placeholder token. If you have more than four tokens here then you need to make sure they are present in the text, otherwise the pipeline will throw an error in the get_annotation_schema_tag() function.

cntx_left class-attribute instance-attribute

cntx_left: int = 15

Number of tokens to take from the left of the concept

cntx_right class-attribute instance-attribute

cntx_right: int = 15

Number of tokens to take from the right of the concept

create_addl_rels class-attribute instance-attribute

create_addl_rels: bool = False

When processing relations from a MedCAT export/docs, relations labeled as 'Other' are created from all the annotations pairs available

create_addl_rels_by_type class-attribute instance-attribute

create_addl_rels_by_type: bool = False

When creating the 'Other' relation class, actually split this class into subclasses based on concept types

device class-attribute instance-attribute

device: str = 'cpu'

The device to use (CPU or GPU).

NB! For these changes to take effect, the pipe would need to be recreated.

idx2labels class-attribute instance-attribute

idx2labels: dict[int, str] = {}

labels2idx class-attribute instance-attribute

labels2idx: dict[str, int] = {}

language class-attribute instance-attribute

language: str = 'en'

Used for Spacy lang setting

limit_samples_per_class class-attribute instance-attribute

limit_samples_per_class: int = -1

Number of samples per class, this limit is applied for train samples, o if train samples are 100 then test would be 20.

log_level class-attribute instance-attribute

log_level: int = INFO

The log level for RelCAT.

NB! For these changes to take effect, the pipe would need to be recreated.

lowercase class-attribute instance-attribute

lowercase: bool = True

If true all input text will be lowercased

max_seq_length class-attribute instance-attribute

max_seq_length: int = 512

The maximum sequence length.

NB! For these changes to take effect, the pipe would need to be recreated.

model_config class-attribute instance-attribute

model_config = ConfigDict(protected_namespaces=())

model_name class-attribute instance-attribute

model_name: str = 'bert-base-uncased'

The name of the model used.

NB! For these changes to take effect, the pipe would need to be recreated.

pin_memory class-attribute instance-attribute

pin_memory: bool = True

If True the data loader will copy the tensors to the GPU pinned memory.

relation_type_filter_pairs class-attribute instance-attribute

relation_type_filter_pairs: list = []

Map from category values to ID, if empty it will be autocalculated during training

seed class-attribute instance-attribute

seed: int = 13

The seed for random number generation.

NB! For these changes to take effect, the pipe would need to be recreated.

task class-attribute instance-attribute

task: str = 'train'

The task for RelCAT.

tokenizer_name class-attribute instance-attribute

tokenizer_name: str = 'bert'

The name of the tokenizer user.

NB! For these changes to take effect, the pipe would need to be recreated.

tokenizer_other_special_tokens class-attribute instance-attribute

tokenizer_other_special_tokens: dict[str, str] = {'pad_token': '[PAD]'}

The special tokens used by the tokenizer. The {PAD} is for Lllama tokenizer.

tokenizer_relation_annotation_special_tokens_tags class-attribute instance-attribute

tokenizer_relation_annotation_special_tokens_tags: list[str] = ['[s1]', '[e1]', '[s2]', '[e2]']

tokenizer_special_tokens class-attribute instance-attribute

tokenizer_special_tokens: bool = False

Tokenizer.

NB! For these changes to take effect, the pipe would need to be recreated.

vocab_size class-attribute instance-attribute

vocab_size: Optional[int] = None

window_size class-attribute instance-attribute

window_size: int = 300

Max acceptable dinstance between entities (in characters), care when using this as it can produce sentences that are over 512 tokens (limit is given by tokenizer)

convert_keys_to_int classmethod

convert_keys_to_int(value)

Source code in medcat-v2/medcat/config/config_rel_cat.py

@classmethod
def convert_keys_to_int(cls, value):
    if isinstance(value, dict):
        return {int(k): v for k, v in value.items()}
    return value

Model

Bases: SerialisableBaseModel

The model part of the RelCAT config

Attributes:

• dropout (float) –
• emb_grad (bool) –

  If True the embeddings will also be trained

• freeze_layers (bool) –

  If we update the weights during training

• hidden_layers (int) –

  hidden_size * 5, 5 being the number of tokens,

• hidden_size (int) –

  The hidden size.

• ignore_cpos (bool) –

  If set to True center positions will be ignored when calculating

• input_size (int) –
• llama_use_pooled_output (bool) –

  If set to True, used only in Llama model, it will add the extra tensor

• model_config –
• model_size (int) –

  The size of the model.

• num_directions (int) –

  2 - bidirectional model, 1 - unidirectional

• padding_idx (int) –

dropout class-attribute instance-attribute

dropout: float = 0.2

emb_grad class-attribute instance-attribute

emb_grad: bool = True

If True the embeddings will also be trained

freeze_layers class-attribute instance-attribute

freeze_layers: bool = True

If we update the weights during training

hidden_layers class-attribute instance-attribute

hidden_layers: int = 3

hidden_size * 5, 5 being the number of tokens, default (s1,s2,e1,e2+CLS).

NB! For these changes to take effect, the pipe would need to be recreated.

hidden_size class-attribute instance-attribute

hidden_size: int = 768

The hidden size.

NB! For these changes to take effect, the pipe would need to be recreated.

ignore_cpos class-attribute instance-attribute

ignore_cpos: bool = False

If set to True center positions will be ignored when calculating representation

input_size class-attribute instance-attribute

input_size: int = 300

llama_use_pooled_output class-attribute instance-attribute

llama_use_pooled_output: bool = False

If set to True, used only in Llama model, it will add the extra tensor formed from selecting the max of the last hidden layer

model_config class-attribute instance-attribute

model_config = ConfigDict(extra='allow', validate_assignment=True, protected_namespaces=())

model_size class-attribute instance-attribute

model_size: int = 5120

The size of the model.

NB! For these changes to take effect, the pipe would need to be recreated.

num_directions class-attribute instance-attribute

num_directions: int = 2

2 - bidirectional model, 1 - unidirectional

padding_idx class-attribute instance-attribute

padding_idx: int = -1

Train

Bases: SerialisableBaseModel

The train part of the RelCAT config

Attributes:

• adam_betas (tuple[float, float]) –
• adam_epsilon (float) –
• adam_weight_decay (float) –
• auto_save_model (bool) –

  Should the model be saved during training for best results

• batch_size (int) –

  batch size

• batching_minority_limit (Union[list[int], int]) –

  Maximum number of samples the minority class can have.

• batching_samples_per_class (list) –

  Number of samples per class in each batch

• class_weights (Union[list[float], None]) –
• enable_class_weights (bool) –
• gradient_acc_steps (int) –
• lr (float) –

  Learning rate

• max_grad_norm (float) –
• model_config –
• multistep_lr_gamma (float) –
• multistep_milestones (list[int]) –
• nclasses (int) –

  Number of classes that this model will output

• nepochs (int) –

  Epochs

• score_average (str) –

  What to use for averaging F1/P/R across labels

• shuffle_data (bool) –

  Used only during training, if set the dataset will be shuffled before

• stratified_batching (bool) –

  Train the model with stratified batching

• test_size (float) –

adam_betas class-attribute instance-attribute

adam_betas: tuple[float, float] = (0.9, 0.999)

adam_epsilon class-attribute instance-attribute

adam_epsilon: float = 1e-08

adam_weight_decay class-attribute instance-attribute

adam_weight_decay: float = 0

auto_save_model class-attribute instance-attribute

auto_save_model: bool = True

Should the model be saved during training for best results

batch_size class-attribute instance-attribute

batch_size: int = 25

batch size

batching_minority_limit class-attribute instance-attribute

batching_minority_limit: Union[list[int], int] = 0

Maximum number of samples the minority class can have. Since the minority class elements need to be repeated, this is used to facilitate that example: batching_samples_per_class - [6,6,6,8,8,8,6,8,8] batching_minority_limit - 6

batching_samples_per_class class-attribute instance-attribute

batching_samples_per_class: list = []

Number of samples per class in each batch example for batch size 64: [6,6,6,8,8,8,6,8,8]

class_weights class-attribute instance-attribute

class_weights: Union[list[float], None] = None

enable_class_weights class-attribute instance-attribute

enable_class_weights: bool = False

gradient_acc_steps class-attribute instance-attribute

gradient_acc_steps: int = 1

lr class-attribute instance-attribute

lr: float = 0.0001

Learning rate

max_grad_norm class-attribute instance-attribute

max_grad_norm: float = 1.0

model_config class-attribute instance-attribute

model_config = ConfigDict(extra='allow', validate_assignment=True)

multistep_lr_gamma class-attribute instance-attribute

multistep_lr_gamma: float = 0.8

multistep_milestones class-attribute instance-attribute

multistep_milestones: list[int] = [2, 4, 6, 8, 12, 15, 18, 20, 22, 24, 26, 30]

nclasses class-attribute instance-attribute

nclasses: int = 2

Number of classes that this model will output

nepochs class-attribute instance-attribute

nepochs: int = 1

Epochs

score_average class-attribute instance-attribute

score_average: str = 'weighted'

What to use for averaging F1/P/R across labels

shuffle_data class-attribute instance-attribute

shuffle_data: bool = True

Used only during training, if set the dataset will be shuffled before train/test split

stratified_batching class-attribute instance-attribute

stratified_batching: bool = False

Train the model with stratified batching

test_size class-attribute instance-attribute

test_size: float = 0.2