marsi.nearest_neighbors package

Submodules

marsi.nearest_neighbors.model module

class marsi.nearest_neighbors.model.ClFail

Bases: object

class marsi.nearest_neighbors.model.KNN(fingerprint, k, mode)

Bases: object

K-Nearest Neighbors runner object.

It is assigned to a model and runs the knn function.

fp

numpy.array – A numpy.array with the fingerprint values.

k

int – The maximum number of neighbors to retrieve.

mode

str – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Methods

__call__(nn)

class marsi.nearest_neighbors.model.RNN(fingerprint, radius, mode)

Bases: object

R-Nearest Neighbors runner object.

It is assigned to a model and runs the rnn function.

fp

numpy.array – A numpy.array with the fingerprint values.

radius

float – A distance radius ]0, 1].

mode

str – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Methods

__call__(nn)

class marsi.nearest_neighbors.model.Distance(fingerprint, mode)

Bases: object

Distance runner object.

It is assigned to a model and runs the distance function.

fp

numpy.array – A numpy.array with the fingerprint values.

mode

str – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Methods

__call__(nn)

class marsi.nearest_neighbors.model.DistributedNearestNeighbors(nns)

Bases: object

Nearest Neighbors distributed implementation.

index

numpy.array – The index of all entries across multiple models.

Attributes

index

Methods

distance_matrix([mode])	Generates a distance matrix between all elements in the models.
distances(fingerprint[, mode, view])	Retrieves the distance a fingerprint and all elements in the model.
feature(index)	Retrieves the fingerprint at a given index.
k_nearest_neighbors(fingerprint[, k, mode, view])	Retrieves the K nearest neighbors to a fingerprint.
radius_nearest_neighbors(fingerprint[, …])	Retrieves the nearest neighbors to a fingerprint within a distance radius.

k_nearest_neighbors(fingerprint, k=5, mode='native', view=<cameo.parallel.SequentialView object>)

Retrieves the K nearest neighbors to a fingerprint.

Parameters:

• fingerprint (list, np.array, tuple) – A fingerprint to use as query.
• k (int) – The number of neighbors to retrieve.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.
• view (cameo.parallel.ParallelView, cameo.parallel.SequentialView) – A parallel mode runner.

Returns:

A dictionary with the InChI Key as key and the distance as value.

Return type:

dict

radius_nearest_neighbors(fingerprint, radius=0.25, mode='native', view=<cameo.parallel.SequentialView object>)

Retrieves the nearest neighbors to a fingerprint within a distance radius.

Parameters:

• fingerprint (list, np.array, tuple) – A fingerprint to use as query.
• radius (float) – A distance radius ]0, 1].
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.
• view (cameo.parallel.ParallelView, cameo.parallel.SequentialView) – A parallel mode runner.

Returns:

A dictionary with the InChI Key as key and the distance as value.

Return type:

dict

distances(fingerprint, mode='native', view=<cameo.parallel.SequentialView object>)

Retrieves the distance a fingerprint and all elements in the model.

Parameters:

• fingerprint (list, np.array, tuple) – A fingerprint to use as query.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.
• view (cameo.parallel.ParallelView, cameo.parallel.SequentialView) – A parallel mode runner.

Returns:

A dictionary with the InChI Key as key and the distance as value.

Return type:

dict

index

distance_matrix(mode='native')

Generates a distance matrix between all elements in the models.

Parameters:mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available. Returns:The distance matrix. Return type:numpy.array

feature(index)

Retrieves the fingerprint at a given index. The index is global for the ensemble of models.

Returns:The fingerprint. Return type:numpy.array Raises:IndexError

class marsi.nearest_neighbors.model.NearestNeighbors(index, features, features_lengths, use_cl=False, opencl_context=None)

Bases: marsi.nearest_neighbors.model_ext.CNearestNeighbors

Attributes

cl_context
data_frame	Create a DataFrame with this model.
features
features_lengths
index
program
start_positions

Methods

distances(fingerprint[, mode])
distances_cl(*args, **kwargs)
distances_py
input_buffer(*args, **kwargs)
knn(fingerprint, k[, mode])	K-Nearest Neighbors
max_memory_allocation_size()
output_buffer(*args, **kwargs)
queue()
rnn(fingerprint, radius[, mode])	Radius-Nearest Neighbors
run_kernel(*args, **kwargs)

cl_context

queue()

program

knn(fingerprint, k, mode='native')

K-Nearest Neighbors

Parameters:

• fingerprint (ndarray) – The fingerprint to search for.
• k (int) – The number of neighbors to return.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Returns:

(Index –> Distance)

Return type:

dict

rnn(fingerprint, radius, mode='native')

Radius-Nearest Neighbors

Parameters:

• fingerprint (ndarray) – The fingerprint to search for.
• radius (float) – The maximum distance of neighbors to return.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Returns:

(Index –> Distance)

Return type:

dict

distances(fingerprint, mode='native')

max_memory_allocation_size()

distances_cl(*args, **kwargs)

input_buffer(*args, **kwargs)

output_buffer(*args, **kwargs)

run_kernel(*args, **kwargs)

index

data_frame

Create a DataFrame with this model.

Returns:A data frame. Return type:pandas.DataFrame

class marsi.nearest_neighbors.model.DBNearestNeighbors(index, session, fingerprint_format, metric='jaccard')

Bases: object

Attributes

data_frame	Create a DataFrame with this model.
features
index
neighbors

Methods

distances(fingerprint[, mode])
knn(fingerprint, k[, mode])	K-Nearest Neighbors
rnn(fingerprint, radius[, mode])	Radius-Nearest Neighbors

neighbors

knn(fingerprint, k, mode='native')

K-Nearest Neighbors

Parameters:

• fingerprint (ndarray) – The fingerprint to search for.
• k (int) – The number of neighbors to return.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Returns:

(Index –> Distance)

Return type:

dict

rnn(fingerprint, radius, mode='native')

Radius-Nearest Neighbors

Parameters:

• fingerprint (ndarray) – The fingerprint to search for.
• radius (float) – The maximum distance of neighbors to return.
• mode (str) – ‘native’ to run python implementation or ‘cl’ to run OpenCL implementation if available.

Returns:

(Index –> Distance)

Return type:

dict

distances(fingerprint, mode='native')

index

features

data_frame

Create a DataFrame with this model.

Returns:A data frame. Return type:pandas.DataFrame

marsi.nearest_neighbors.model_ext module

class marsi.nearest_neighbors.model_ext.CNearestNeighbors

Bases: object

Attributes

features
features_lengths
start_positions

Methods

distances_py

distances_py()

features

features_lengths

start_positions

Module contents

marsi.nearest_neighbors.build_nearest_neighbors_model(database, fpformat='fp4', solubility='high', n_models=5, chunk_size=1000000.0, view=<class 'cameo.parallel.SequentialView'>)

Loads a NN model.

If a ‘default_model.pickle’ exists in data it will load the model. Otherwise it will build a model from the Database. This can take several hours depending on the size of the database.

Parameters:

• database (marsi.io.mongodb.CollectionWrapper) – A Database interface to the metabolites.
• chunk_size (int) – Maximum number of entries per chunk.
• fpformat (str) – The format of the fingerprint (see pybel.fps)
• solubility (str) – One of high, medium, low or all.
• view (cameo.parallel.SequentialView, cameo.parallel.MultiprocesingView) – A view to control parallelization.
• n_models (int) – The number of NearestNeighbors models.

marsi.nearest_neighbors.load_nearest_neighbors_model(chunk_size=1000000.0, fpformat='fp4', solubility='all', session=<sqlalchemy.orm.session.Session object>, view=<cameo.parallel.SequentialView object>, model_size=100000, source='db', costum_query=None)

Loads a NN model.

If a ‘default_model.pickle’ exists in data it will load the model. Otherwise it will build a model from the Database. This can take several hours depending on the size of the database.

Parameters:

• chunk_size (int) – Maximum number of entries per chunk.
• fpformat (str) – The format of the fingerprint (see pybel.fps)
• solubility (str) – One of high, medium, low or all.
• view (cameo.parallel.SequentialView, cameo.parallel.MultiprocesingView) – A view to control parallelization.
• model_size (int) – The size of each NearestNeighbor in the ensemble.