OpenEnsembles

The “data” class

class openensembles.data(df, x)[source]

df is a dataframe and x is the x_axis values (or numbers indicating the number entry). Behavior: Only numerical data in df will be carried into a numpy array

Parameters
dfa pandas dataframe

Dataframe with objects in rows and columns representing the feature dimensions

xlist

The x-axis elements. If x is a list of strings, it will be converted here to a list of ints (range 0 to len(x))

Raises
ValueError of the size of x and dimensionality of df do not match
Attributes
dfpandas dataframe

the original dataframe

Ddictionary

A dictionary of numpy data matrices, callable by ‘source_name’.

xlist

a list of integer or float values

x_labelslist

a list of strings (if that was passed in) or the int and float. So that xticklabels could be updated or referenced

params: dict

A dictionary of parameter labels and their values that were used during transformations

Methods

merge(d_list)

Returns an appended object – a merge of the data object (self) and all data objects inside a passed list.

plot_data(source_name[, fig_num])

Plot the data matrix that belongs to source_name

slice(names)

Returns a new data object containing a slice indicated by the list of names given (dictionary keys shared amongst D, params, etc.).

transform(source_name, txfm_fcn, txfm_name, …)

This runs transform (txfm_fcn) on the data matrix defined by source_name with parameters that are variable for each transform.

transforms_available()

Returns a list of all transformations available

merge(d_list)[source]

Returns an appended object – a merge of the data object (self) and all data objects inside a passed list.

Parameters
d_list: list

A list of data objects

Returns
transDictArr: list of dicts

A list of dictionary translation of new labels in merged object, with original labels. List order is same as those passed in

Raises
ValueError

If objects in d_list are not well formed openenembles data objects

Examples

Merge two sets of data objects. Keeps the same dataframe (self.df), i.e. use this when it makes sense to always use that reference dataframe

FINISH EXAMPLES here

plot_data(source_name, fig_num=1, **kwargs)[source]

Plot the data matrix that belongs to source_name

Parameters
source_namestring

name of data source to plot, e.g. ‘parent’

fig_num: int

Set to a different figure number to plot on existing figure. Default fig_num=1

**class_labelslist of ints

this is a vector that assigns points to classes, and will be used to color the points according to assigned class type

**clusters_to_plot: list of ints

If you wish to plot a subset of cluster types (classes), pass that as a list of ints

**titlestring

Title for plot

Raises
ValueError:

If clusters_to_plot not a set in cluster labels

slice(names)[source]

Returns a new data object containing a slice indicated by the list of names given (dictionary keys shared amongst D, params, etc.). Cannot remove ‘parent’ as that is the default dataframe matrix that established data object. To replace parent, instead instantiate a new object on a dataframe created from transformation of interest.

Parameters
names: list

A list of strings matching the names to keep in the new slice

Returns
d: an openensembles data object

A oe.data object that contains only those names passed in

Raises
ValueError

If a name in the list of names does not exist in data object

Examples

Remove ‘zscore’ from the list, keeping everything else

>>> names = d.D.keys() #get all the keys
>>> names = names.remove(['zscore'])
>>> dNew = d.slice(names)
transform(source_name, txfm_fcn, txfm_name, **kwargs)[source]

This runs transform (txfm_fcn) on the data matrix defined by source_name with parameters that are variable for each transform. For example, oe.data.transform(‘parent’, ‘zscore’,’zscore_parent’, axis=0) will run the zscore in a vector-wise manner across the matrix (column-wise) and the new data dictionary access to the transformed data is oe.data[‘zscore_parent’] Successful completion results in the addition of a new entry in the data dictionary with a key according to txfm_name.

Parameters
source_name: string

the name of the source data, for example ‘parent’, or ‘log2’

txfm_fcn: string

the name of the transform function. See transforms.py or run oe.data.transforms_available() for list

txfm_name: string

the name you want to use in the data object dictionary oe.data.D[‘name’] to access transformed data

Other Parameters
**Keep_NaN: boolean

Set to True in order to prevent transformations from being added that produce NaNs. Default Keep_NaN=True this will add transformed data even if NaNs are produced. Set to 0 to prevent addition of data transforms containing NaNs.

**Keep_Inf: boolean

Set to True in order to prevent transformations from being added that produce infinite values Default: Keep_Inf = True (this will add transformed data even if infinite values are produced. Set to 0 to prevent addition of data transforms conta

Raises
ValueError

if the transform function does not exist OR if the data source does not exist by source_name

Warning

NaNs or infinite values are produced

Examples

>>> import pandas as pd
>>> import openensembles as oe
>>> df = pd.read_csv(file)
>>> d = oe.data(df, df.columns
>>> d.transform('parent', 'zscore', 'zscore')
>>> d.transform('zscore', 'PCA', 'pca', n_components=3)
transforms_available()[source]

Returns a list of all transformations available

The “cluster” class

class openensembles.cluster(dataObj)[source]

Initialize a clustering object, which is instantiated with a data object class from OpenEnsembles When clustering is performed, the dictionaries of all attributes are extended using the key given as output_name

Parameters
dataObj

openensembles.data class – consists at least of one data matrix called ‘parent’

Returns
clusterObject

empty openensembles.cluster object

Examples

Load data, zscore it, transform it into the first three principal components and cluster using KMeans with K=4

>>> import pandas as pd
>>> import openensembles as oe
>>> df = pd.read_csv(file)
>>> d = oe.data(df, df.columns
>>> d.transform('parent', 'zscore', 'zscore')
>>> d.transform('zscore', 'PCA', 'pca', n_components=3)
>>> c = oe.cluster(d)
>>> c.cluster('pca', 'kmeans', 'kmeans_pca', 4)
Attributes
dataObj: openensembles.data class

openensembles.data class that was used to instantiate cluster object

labels: dict of lists

A dictionary of lists of clustering solutions (ints). Referred to as output_name in .cluster method

data_source: dict of strings

Name of data source in dataObj

params: dict of dicts

A dictionary of all parameters passed during clustering

clusterNumbers: dict of lists

A listing of the unique set of cluster numbers produced in a clustering

random_state: dict of objects

A listing of the random state objects that can be used to reset the state and

Methods

MI([MI_type])

Calculate the mutual information between all pairs of clustering solutions

algorithms_available()

Call this to list all algorithms currently available in algorithms.py

cluster(source_name, algorithm, output_name)

This runs clustering algorithms on the data matrix defined by source_name with parameters that are variable for each algorithm.

clustering_algorithm_parameters()

This function returns a dictionary with keys equal to parameters of interest {K, linkage, affinity} whose entries indicate algorithms that take those as free parameters.

co_occurrence_matrix([data_source_name])

Calculate the co-occurrence of all pairs of objects across the ensemble

finish_co_occ_linkage(threshold[, linkage])

The finishing technique that calculates a co-occurrence matrix on all cluster solutions in the ensemble and then hierarchically clusters the co-occurrence, treating it as a similarity matrix.

finish_graph_closure(threshold[, clique_size])

The finishing technique that treats the co-occurrence matrix as a graph, that is binarized by the threshold (>=threshold becomes an unweighted, undirected edge in an adjacency matrix).

finish_majority_vote([threshold])

Based on Ana Fred’s 2001 paper: Fred, Ana.

get_cluster_members(solution_name, clusterNum)

Return the dataframe row indexes of a cluster number in solution named by solution_name

merge(c_list)

Returns an appended object – a merge of the cluster object (self) and all cluster objects inside a passed list.

mixture_model([K, iterations])

Finishing Technique to assemble a final, hard parition of the data according to maximizing the likelihood according to the observed clustering solutions across the ensemble.

search_field(field, value)

Find solutions that were made with

slice(names)

Returns a new cluster object containing a slice indicated by the list of names given (dictionary keys shared amongst labels, params, etc.)

MI(MI_type='standard')[source]

Calculate the mutual information between all pairs of clustering solutions

Parameters
MI_type: string {‘standard’, ‘adjusted’, ‘normalized’}

The sklearn.metric mutual information to use, either mutual_info, adjusted_mutual_info, or normalized_mutual_info

Returns
MI class

mutualinformation.MI class, where MI.matrix is the claculated matrix of pairwise mutual information. The diagonal is not guaranteed to be 1 (it depends on the type of MI calculated)

Examples

>>> MI = c.MI(MI_type='adjusted')
>>> MI.plot(sorted=True)
algorithms_available()[source]

Call this to list all algorithms currently available in algorithms.py

cluster(source_name, algorithm, output_name, K=None, Require_Unique=False, random_seed=None, **kwargs)[source]

This runs clustering algorithms on the data matrix defined by source_name with parameters that are variable for each algorithm. Note that K is required for most algorithms.

Parameters
source_name: string

the source data matrix name to operate on in clusterclass dataObj

algorithm: string

name of the algorithm to use, see clustering.py or call oe.cluster.algorithms_available()

output_name: string

this is the dict key for interacting with the results of this clustering solution in any of the cluster class dictionary attributes

K: int

number of clusters to create (ignored for algorithms that define K during clustering). The var_params gets K after, either the parameter passed, or the number of clusters produced if the K was not passed.

Require_Unique: bool

If FALSE and you already have an output_name solution, this will append a number to create a unique name. If TRUE and a solution by that name exists, this will not add solution and raise ValueError. Default Require_Unique=False

random_seed: int or random.getstate()

Pass a random seed or random seed state (random.getstate()) in order to force the starting point of a clustering algorithm to that state. Default is None

Raises
ValueError

if data source is not available by source_name

Warning

This will warn if the number of clusters is differen than what was requested, typically when an algorithm does not accept K as an argument.

Examples

Cluster using KMeans on parent data

>>> c = oe.cluster
>>> c.cluster('parent', 'kmeans','kmeans_parent', K=5) 

Form an iteration to build an ensemble using different values for K

>>> for k in range(2,12):
>>>     name='kmeans_'+k
>>>     c.cluster('parent', 'kmeans', name, k)
clustering_algorithm_parameters()[source]

This function returns a dictionary with keys equal to parameters of interest {K, linkage, affinity} whose entries indicate algorithms that take those as free parameters. For example K-means takes K as an argument, but Affinity Propagation does not, so you will find kmeans is listed in dict[‘K’], but not AffinityPropagation. This is not inclusive of all paramaters of every algorithm, but the common parameters one might want to vary.

Returns
a: dictionary

Keys equal to parameters {K, linkages, distances} and values as lists of algorithms that use that key as a variable

co_occurrence_matrix(data_source_name='parent')[source]

Calculate the co-occurrence of all pairs of objects across the ensemble

Parameters: data_source_name: string

Name of the data source to link to co-occurrence object. Default is ‘parent’

Returns
coMat class

coMat.co_matrix is the NxN matrix, whose entries indicate the number of times the pair of objects in positon (i,j) cluster across the ensemble of clustering solutions available in clustering object.

Examples

>>> coMat = c.co_occurrence_matrix()
>>> coMat.plot()
finish_co_occ_linkage(threshold, linkage='average')[source]

The finishing technique that calculates a co-occurrence matrix on all cluster solutions in the ensemble and then hierarchically clusters the co-occurrence, treating it as a similarity matrix. The clusters are defined by the threshold of the distance used to cut.

Parameters
threshold: float

Linkage distance to use as a cutoff to create partitions

linkage: string

Linkage type. See scipy.cluster.hierarchy

Returns
c: openensembles clustering object

a new clustering object with c.labels[‘co_occ_linkage’] set to the final solution.

Examples

To determine where the cut is visually, at threshold=0.5:

>>> coMat = c.co_occurrence()
>>> coMat.plot(threshold=0.5, linkage='ward')

To create the cut at threshold=0.5

>>> cWard = c.co_occ_linkage(0.5, 'ward')
>>> d.plot_data('parent', cluster_labels=cWard.labels['co_occ_linkage'])
finish_graph_closure(threshold, clique_size=3)[source]

The finishing technique that treats the co-occurrence matrix as a graph, that is binarized by the threshold (>=threshold becomes an unweighted, undirected edge in an adjacency matrix). This graph object is then subjected to clique formation according to clique_size (such as triangles if clique_size=3). The cliques are then combined in the graph to create unique cluster formations.

Returns
c: openenembles clustering object

New cluster object with final solution and name ‘graph_closure’

See also

finishing.py

Examples

>>> cGraph = c.finish_graph_closure(0.5, 3)
>>> d.plot_data('parent', cluster_labels=cGraph.labels['graph_closure'])
finish_majority_vote(threshold=0.5)[source]

Based on Ana Fred’s 2001 paper: Fred, Ana. Finding Consistent Clusters in Data Partitions. In Multiple Classifier Systems, edited by Josef Kittler and Fabio Roli, LNCS 2096, 309-18. Springer, 2001. This algorithm assingns clusters to the same class if they co-cluster at least 50 of the time. It greedily joins clusters with the evidence that at least one pair of items from two different clusters co-cluster a majority of the time. Outliers will get their own cluster.

Parameters
threshold: float

the threshold, or fraction of times objects co-cluster to consider a ‘majority’. Default is 0.5 (50% of the time)

Returns
c: openensembles cluster object

New cluster object with final solution and name ‘majority_vote’

Examples

>>> c_MV = c.majority_vote(threshold=0.7)
>>> labels = c_MV.labels['majority_vote']
get_cluster_members(solution_name, clusterNum)[source]

Return the dataframe row indexes of a cluster number in solution named by solution_name

Parameters
solution_name: string

the name of the clustering solution of interest

clusterNum: int

The cluster number of interest

Returns
indexes: list

a list of indexes of objects with clusterNum in solution_name

Examples

Get a list of objects that belong to each cluster type in a solution

>>> name = 'zscore_agglomerative_ward'
>>> c.cluster('zscore', 'agglomerative', name, K=4, linkage='ward')
>>> labels = {}
>>> for i in c.clusterNumbers[name]:
>>>     labels{i} = c.get_cluster_members(name, i)
merge(c_list)[source]

Returns an appended object – a merge of the cluster object (self) and all cluster objects inside a passed list. This will keep the parent dataobject of the self cluster object. This assumes that the ojbects were instantiated and clustered on the same data source (at least the same mxn features)

Parameters
c_list: list

A list of cluster objects

Returns
transDictArr: list of dicts

A list of dictionary translation of new labels in merged object, with original labels. List order is same as those passed in

Raises
ValueError

If objects in c_list are not well formed cluster objects

Examples

Merge two sets of clustering objects

FINISH EXAMPLES here

mixture_model(K=2, iterations=10)[source]

Finishing Technique to assemble a final, hard parition of the data according to maximizing the likelihood according to the observed clustering solutions across the ensemble. This will operate on all clustering solutions contained in the container cluster class. Operates on entire ensemble of clustering solutions in self, to create a mixture model See finishing.mixture_model for more details.

Parameters
K: int

number of clusters to create. Default K=2

iterations: int

number of iterations of EM algorithm to perform. Default iterations=10

Returns
c: openensembles clustering object

a new clustering object with c.labels[‘mixture_model’] set to the final solution.

Raises
ValueError:

If there are not at least two clustering solutions

References

Topchy, Jain, and Punch, “A mixture model for clustering ensembles Proc. SIAM Int. Conf. Data Mining (2004)”

Examples

>>> cMM = c.mixture_model(4, 10)
>>> d.plot_data('parent', cluster_labels=cMM.labels['mixture_model'])
search_field(field, value)[source]

Find solutions that were made with

Parameters
field: string {‘algorithm’, ‘data_source’, ‘K’, ‘linkage’, ‘distance’, ‘clusterNumber’, etc.}

The name of field, either in algorithm used, data_source selected, or a parameter passed to search for an exact value

value: string or int

The value to search for (where ints are passed for K (desired clusters) or clusterNumber (actual returned clusters))

Returns
names: list of strings

The names of dictionary entries in clustering solutions matching field, value criteria. Returns empty list if nothing was found

Raises
ValueError

If the field was not recognized.

Examples

Find all clustering solutions where K=2 was used >>> names = c.search_field(‘K’, 2)

Find all clustering solutions where the actual cluster numbers were 2 >>> names = c.search_field(‘clusterNumber’, 2)

Find all solutions clustered using kmeans >>> names = c.search_field(‘algorithm’, ‘kmeans’)

Find all clustering solutions where ward linkage was used >>> names = c.search_field(‘linkage’, ‘ward’)

slice(names)[source]

Returns a new cluster object containing a slice indicated by the list of names given (dictionary keys shared amongst labels, params, etc.)

Parameters
names: list

A list of strings matching the names to keep in the new slice

Returns
c: an openensembles clustering object

A oe.cluster object that contains only those names passed in

Raises
ValueError

If a name in the list of names does not exist in cluster object

Examples

Get only the solutions made by agglomerative clustering

>>> names = c.search_field('algorithm', 'agglomerative') #return all solutions with agglomerative
>>> cNew = c.slice(names)

Get only the solutions that were made with K=2 calls

>>> names = c.search_field('K', 2) #return all solution names that used K=2
>>> cNew = c.slice(names)

The “validation” class

class openensembles.validation(dataObj, cObj)[source]

validation is a class to calculate any number of validation metrics on clustering solutions in data. An individual validation metric must be called on a particular instantiation of the data matrix (like ‘parent’ or ‘zscore’) and a specific solution in cObj.

Methods

calculate

calculate(validation_name, cluster_name, source_name='parent')[source]

Calls the function titled by validation_name on the data matrix set by source_name (default ‘parent’) and clustering solution by cluster_name Appends to validation with key value equal to the validation_name+source_name+cluster_name

Returns
output_name: str

The handle name for accessing validation results

merge(v_list)[source]

Returns an appended object – a merge of the validation object (self) and all validation objects inside a passed list. This will keep the dataobject and clusterObjects of the self validation object.

Parameters
v_list: list

A list of validation objects

Returns
transDictArr: list of dicts

A list of dictionary translation of new labels in merged object, with original labels. List order is same as those passed in

Raises
ValueError

If objects in v_list are not well formed value objects

Examples

Merge two sets of validation objects

FINISH EXAMPLES here

validation_metrics_available()[source]

Return all available validation metrics