mindmeld.models.taggers.taggers module

This module contains all code required to perform sequence tagging.

class mindmeld.models.taggers.taggers.BoundaryCounts

Bases: object

This class stores the counts of the boundary evaluation metrics.

le

int -- Label error count. This is when the span is the same but the entity label is incorrect

be

int -- Boundary error count. This is when the entity type is correct but the span is incorrect

lbe

int -- Label boundary error count. This is when both the entity type and span are incorrect, but there was an entity predicted

tp

int -- True positive count. When an entity was correctly predicted

tn

int -- True negative count. Count of times it was correctly predicted that there is no entity

fp

int -- False positive count. When an entity was predicted but one shouldn't have been

fn

int -- False negative count. When an entity was not predicted where one should have been

to_dict()

Converts the object to a dictionary

class mindmeld.models.taggers.taggers.Tagger(**parameters)

Bases: object

A class for all sequence tagger models implemented in house. It is important to follow this interface exactly when implementing a new model so that your model is configured and trained as expected in the MindMeld pipeline. Note that this follows the sklearn estimator interface so that GridSearchCV can be used on our sequence models.

static dump(model_path)

Dumps any tagger specific data to disk and returns a model_path (modified if required). This is a no-op since we do not have to do anything special to dump default serializable models for SKLearn.

Parameters:model_path (str) -- The path to dump the model to

extract_and_predict(examples, config, resources)

Does both feature extraction and prediction. Often necessary for sequence models when the prediction of the previous example is used as a feature for the next example. If this is not the case, extract is simply called before predict here. Note that the MindMeld config and resources are passed in each time to make the underlying model implementation stateless.

Parameters:

• examples (list of mindmeld.core.Query) -- A list of queries to extract features for and predict
• config (ModelConfig) -- The ModelConfig which may contain information used for feature extraction
• resources (dict) -- Resources which may be used for this model's feature extraction

Returns:

A list of predicted labels (in encoded format)

Return type:

(list of classification labels)

extract_features(examples, config, resources)

Extracts all features from a list of MindMeld examples. Processes the data and returns the features in the format that is expected as an input to fit(). Note that the MindMeld config and resources are passed in each time to make the underlying model implementation stateless.

Parameters:

• examples (list of mindmeld.core.Query) -- A list of queries to extract features for
• config (ModelConfig) -- The ModelConfig which may contain information used for feature extraction
• resources (dict) -- Resources which may be used for this model's feature extraction

Returns:

tuple containing:

• (list of feature vectors): X
• (list of labels): y
• (list of groups): A list of groups to be used for splitting with sklearn GridSearchCV

Return type:

(tuple)

fit(X, y)

Trains the model. X and y are the format of what is returned by extract_features. There is no restriction on their type or content. X should be the fully processed data with extracted features that are ready to be used to train the model. y should be a list of classes as encoded by the label_encoder

Parameters:

• X (list) -- Generally a list of feature vectors, one for each training example
• y (list) -- A list of classification labels (encoded by the label_encoder, NOT MindMeld entity objects)

Returns:

self

get_params(deep=True)

Gets a dictionary of all of the current model parameters and their values

Parameters:deep (bool) -- Not used, needed for sklearn compatibility Returns:A dictionary of the model parameter names as keys and their set values Return type:(dict)

static load(model_path)

Load the model state to memory. This is a no-op since we do not have to do anything special to load default serializable models for SKLearn.

Parameters:model_path (str) -- The path to dump the model to

predict(X, dynamic_resource=None)

Predicts the labels from a feature matrix X. Again X is the format of what is returned by extract_features.

Parameters:X (list) -- A list of feature vectors, one for each example Returns:a list of predicted labels (in an encoded format) Return type:(list of classification labels)

predict_proba(examples, config, resources)

Parameters:

• examples (list of mindmeld.core.Query) -- A list of queries to extract features for and predict
• config (ModelConfig) -- The ModelConfig which may contain information used for feature extraction
• resources (dict) -- Resources which may be used for this model's feature extraction

Returns:

A list of predicted labels (in encoded format) and confidence scores

Return type:

(list of lists)

set_params(**parameters)

Sets the model parameters. Defaults should be set for all parameters such that a model is initialized with reasonable default parameters if none are explicitly passed in.

Parameters:**parameters -- Arbitrary keyword arguments. The keys are model parameter names and the values are what they should be set to Returns:self

setup_model(config)

"Not implemented.

static unload()

is_serializable

mindmeld.models.taggers.taggers.extract_sequence_features(example, example_type, feature_config, resources)

Extracts feature dicts for each token in an example.

Parameters:

• example (mindmeld.core.Query) -- a query
• example_type (str) -- The type of example
• feature_config (dict) -- The config for features
• resources (dict) -- Resources of this model

Returns:

features

Return type:

(list of dict)

mindmeld.models.taggers.taggers.get_boundary_counts(expected_sequence, predicted_sequence, boundary_counts)

Gets the boundary counts for the expected and predicted sequence of entities.

mindmeld.models.taggers.taggers.get_entities_from_tags(query, tags, system_entity_recognizer)

From a set of joint IOB tags, parse the app and system entities.

This performs the reverse operation of get_tags_from_entities.

Parameters:

• query (Query) -- Any query instance.
• tags (list of str) -- Joint app and system tags, like those created by get_tags_from_entities.
• system_entity_recognizer (SystemEntityRecognizer) --

Returns:

(list of QueryEntity) The tuple containing the list of entities.

mindmeld.models.taggers.taggers.get_tags_from_entities(query, entities, scheme='IOB')

Get joint app and system IOB tags from a query's entities.

Parameters:

• query (Query) -- A query instance.
• entities (List of QueryEntity) -- A list of queries found in the query

Returns:

The tags for each token in the query. A tag has four parts separated by '|'. The first two are the IOB status for app entities followed by the type of app entity or '' if the IOB status is 'O'. The last two are like the first two, but for system entities.

Return type:

(list of str)