Cluster

The classes in Cluster defines how each instrument should be grouped in a cluster.

class fipie.NoCluster

Create no cluster and put all instruments into one group which is the classical way of constructing a portfolio

__init__()

Instantiate the clustering algorithm. No extra parameter is needed.

create_clusters(data: pandas.core.frame.DataFrame) → numpy.array

Return a list of cluster indices for each instrument

Parameters

data – instrument returns which the clustering method is based on

Returns

array of cluster indices

Return type

np.array

class fipie.CorrMatrixDistance(max_clusters: int)

Create clusters based on the distance in correlation matrix. The max_node_size parameter controls the maximum number of instruments in a cluster.

cf. https://github.com/TheLoneNut/CorrelationMatrixClustering/blob/master/CorrelationMatrixClustering.ipynb

__init__(max_clusters: int)

Instantiate the clustering algorithm

Parameters

max_clusters (int) – maximum number of clusters (or instruments) one cluster can have.

create_clusters(data: pandas.core.frame.DataFrame) → numpy.array

Return a list of cluster indices for each instrument

Parameters

data – instrument returns which the clustering method is based on

Returns

array of cluster indices

Return type

np.array

pre_process(data: pandas.core.frame.DataFrame) → pandas.core.frame.DataFrame

Pre-process the data time-series before feeding to the clustering algorithm. Remove instruments whose return standard deviation is zero in which case the correlation matrix can not be computed.

Parameters

data (pd.DataFrame) – instrument returns which the clustering method is based on

Returns

pre-processed data

Return type

pd.DataFrame