`gunshotmatch_pipeline.decision_tree`

Prepare data and train decision trees.

Classes:

DecisionTreeVisualiser(classifier, ...)

Class for exporting visualisations of a decision tree or random forest.

Functions:

`data_from_projects`(projects[, normalize])	Returns a `DataFrame` containing decision tree data for the given projects.
`data_from_unknown`(unknown, feature_names[, ...])	Returns a `DataFrame` containing decision tree data for the given unknown.
`dotsafe_name`(name)	Return a dot (graphviz) suitable name for a sample, with special characters escaped.
`fit_decision_tree`(data, classifier)	Fit the classifier to the data.
`get_feature_names`(data)	Return the feature names for the given data.
`predict_unknown`(unknown, classifier, ...)	Predict classes for an unknown sample from a decision tree or random forest.
`simulate_data`(project[, normalize, n_simulated])	Generate simulated peak area data for a project.
`visualise_decision_tree`(data, classifier, ...)	Visualise a decision tree with graphviz.

class DecisionTreeVisualiser(classifier, feature_names, factorize_map)[source]

Bases: object

Class for exporting visualisations of a decision tree or random forest.

New in version 0.8.0.

Parameters

classifier (ClassifierMixin) – Decision tree or random forest classifier.
feature_names (List[str]) – The compounds the decision tree was trained on.
factorize_map (List[str]) – List of class names in the order they appear as classes in the classifier.

Methods:

`__eq__`(other)	Return `self == other`.
`__getstate__`()	Used for pickling.
`__ne__`(other)	Check equality and either forward a NotImplemented or return the result negated.
`__repr__`()	Return a string representation of the `DecisionTreeVisualiser`.
`__setattr__`(name, val)	Implement `setattr(self, name)`.
`__setstate__`(state)	Used for pickling.
`from_data`(data, classifier, factorize_map)	Alternative constructor from the pandas dataframe the classifier was trained on.
`visualise_tree`([filename, filetype])	Visualise the decision tree or random forest as an image.

Attributes:

`classifier`	Decision tree or random forest classifier.
`factorize_map`	List of class names in the order they appear as classes in the classifier.
`feature_names`	The compounds the decision tree was trained on.

__eq__(other)

Return self == other.

Return type: bool

__getstate__()

Used for pickling.

Automatically created by attrs.

__ne__(other)

Check equality and either forward a NotImplemented or return the result negated.

Return type: bool

__repr__()

Return a string representation of the DecisionTreeVisualiser.

Return type: str

__setattr__(name, val): Implement setattr(self, name).

__setstate__(state)

Used for pickling.

Automatically created by attrs.

classifier

Type: ClassifierMixin

Decision tree or random forest classifier.

factorize_map

Type: List[str]

List of class names in the order they appear as classes in the classifier.

feature_names

Type: List[str]

The compounds the decision tree was trained on.

classmethod from_data(data, classifier, factorize_map)[source]

Alternative constructor from the pandas dataframe the classifier was trained on.

Parameters

data (DataFrame)
classifier (ClassifierMixin) – Decision tree or random forest classifier.
factorize_map (List[str]) – List of class names in the order they appear as classes in the classifier.

Return type

DecisionTreeVisualiser

visualise_tree(filename='decision_tree_graphivz', filetype='svg')[source]

Visualise the decision tree or random forest as an image.

Parameters

filename (str) – Output filename without extension; for random forest, the base filename (followed by -tree-n). Default 'decision_tree_graphivz'.
filetype (str) – Output filetype (e.g. svg, png, pdf). Default 'svg'.

data_from_projects(projects, normalize=False)[source]

Returns a DataFrame containing decision tree data for the given projects.

Parameters

projects (Projects)
normalize (bool) – Default False.

Return type

Tuple[DataFrame, List[str]]

data_from_unknown(unknown, feature_names, normalize=False)[source]

Returns a DataFrame containing decision tree data for the given unknown.

Parameters

unknown (UnknownSettings)
feature_names (Collection[str]) – The compounds the decision tree was trained on. Extra compounds in the unknown will be excluded.
normalize (bool) – Default False.

Return type

DataFrame

dotsafe_name(name)[source]

Return a dot (graphviz) suitable name for a sample, with special characters escaped.

Parameters: name (str)
Return type: str

New in version 0.5.0.

fit_decision_tree(data, classifier)[source]

Fit the classifier to the data.

Parameters

data (DataFrame)
classifier (ClassifierMixin)

Return type

List[str]

Returns

List of feature names

get_feature_names(data)[source]

Return the feature names for the given data.

Parameters: data (DataFrame)
Return type: List[str]

predict_unknown(unknown, classifier, factorize_map, feature_names)[source]

Predict classes for an unknown sample from a decision tree or random forest.

Parameters

unknown (UnknownSettings)
classifier (ClassifierMixin)
factorize_map (List[str]) – List of class names in the order they appear as classes in the classifier.
feature_names (List[str]) – The compounds the decision tree was trained on. Extra compounds in the unknown will be excluded.

Return type

Iterator[Tuple[str, float]]

Returns

An iterator of predicted class names and their probabilities, ranked from most to least likely.

New in version 0.9.0.

simulate_data(project, normalize=False, n_simulated=10)[source]

Generate simulated peak area data for a project.

Parameters

project (Project)
normalize (bool) – Default False.
n_simulated (int) – The number of values to simulate. Default 10.

Return type

DataFrame

visualise_decision_tree(data, classifier, factorize_map, filename='decision_tree_graphivz', filetype='svg')[source]

Visualise a decision tree with graphviz.

Parameters

data (DataFrame)
classifier (ClassifierMixin)
factorize_map (List[str]) – List of class names in the order they appear as classes in the classifier.
filename (str) – Output filename without extension; for random forest, the base filename (followed by -tree-n). Default 'decision_tree_graphivz'.
filetype (str) – Output filetype (e.g. svg, png, pdf). Default 'svg'.

`gunshotmatch_pipeline.decision_tree.export`

Export and load decision trees to/from JSON-safe dictionaries..

New in version 0.6.0.

Functions:

`serialise_decision_tree`(model)	Serialise a decision tree to a JSON-safe dictionary.
`deserialise_decision_tree`(model_dict)	Deserialise a decision tree.
`verify_saved_decision_tree`(in_process, from_file)	Verify the saved `DecisionTreeClassifier` matches the model in memory.
`serialise_random_forest`(model)	Serialise a random forest to a JSON-safe dictionary.
`deserialise_random_forest`(model_dict)	Deserialise a random forest.
`verify_saved_random_forest`(in_process, from_file)	Verify the saved `RandomForestClassifier` matches the model in memory.

serialise_decision_tree(model)[source]

Serialise a decision tree to a JSON-safe dictionary.

Parameters: model (DecisionTreeClassifier) – Trained decision tree.
Return type: Dict[str, Any]

deserialise_decision_tree(model_dict)[source]

Deserialise a decision tree.

Parameters: model_dict (Dict[str, Any]) – JSON-safe representation of the decision tree.
Return type: DecisionTreeClassifier

verify_saved_decision_tree(in_process, from_file)[source]

Verify the saved DecisionTreeClassifier matches the model in memory.

Will raise an AssertionError if the data do not match.

Parameters

in_process (DecisionTreeClassifier) – The DecisionTreeClassifier already in memory.
from_file (DecisionTreeClassifier) – A DecisionTreeClassifier loaded from disk.

New in version 0.7.0.

serialise_random_forest(model)[source]

Serialise a random forest to a JSON-safe dictionary.

Parameters: model (RandomForestClassifier) – Trained random forest.
Return type: Dict[str, Any]

deserialise_random_forest(model_dict)[source]

Deserialise a random forest.

Parameters: model_dict (Dict[str, Any]) – JSON-safe representation of the random forest.
Return type: RandomForestClassifier

verify_saved_random_forest(in_process, from_file)[source]

Verify the saved RandomForestClassifier matches the model in memory.

Will raise an AssertionError if the data do not match.

Parameters

in_process (RandomForestClassifier) – The RandomForestClassifier already in memory.
from_file (RandomForestClassifier) – A RandomForestClassifier loaded from disk.

New in version 0.7.0.

`gunshotmatch_pipeline.decision_tree.predictions`

Represents random forest classifier predictions for testing classifier performance.

New in version 0.9.0.

Classes:

PredictionResult(name, class_name, predictions)

Represents the predicted classes from a random forest classifier.

Functions:

`dump_predictions`(predictions[, indent])	Return a JSON representation of the predictions.
`load_predictions`(predictions_json)	Load predictions from the given JSON string.

namedtuple PredictionResult(name, class_name, predictions)[source]

Bases: NamedTuple

Represents the predicted classes from a random forest classifier.

Fields

name (str) – The sample name e.g. “Unknown Western Double A”
class_name (str) – The class name, i.e. the ammo type e.g. “Western Double A”
predictions (Tuple[Tuple[str, float], …]) – List of predictions (pairs of (<class name>, <probability>))

property correct

Returns whether the top prediction matches the actual class name.

Return type: bool

__repr__(): Return a nicely formatted representation string

dump_predictions(predictions, indent=2)[source]

Return a JSON representation of the predictions.

Parameters

predictions (List[PredictionResult])
indent (Optional[int]) – Default 2.

Return type

str

load_predictions(predictions_json)[source]

Load predictions from the given JSON string.

Parameters: predictions_json (str)
Return type: List[PredictionResult]

gunshotmatch_pipeline.decision_tree

gunshotmatch_pipeline.decision_tree.export

gunshotmatch_pipeline.decision_tree.predictions

`gunshotmatch_pipeline.decision_tree`

`gunshotmatch_pipeline.decision_tree.export`

`gunshotmatch_pipeline.decision_tree.predictions`