def overlap(input_graph, vertex_pair=None):
"""
Compute the Overlap Coefficient between each pair of vertices connected by
an edge, or between arbitrary pairs of vertices specified by the user.
Overlap Coefficient is defined between two sets as the ratio of the volume
of their intersection divided by the smaller of their two volumes. In the
context of graphs, the neighborhood of a vertex is seen as a set. The
Overlap Coefficient weight of each edge represents the strength of
connection between vertices based on the relative similarity of their
neighbors. If first is specified but second is not, or vice versa, an
exception will be thrown.
Parameters
----------
graph : cugraph.Graph
cuGraph graph descriptor, should contain the connectivity information
as an edge list (edge weights are not used for this algorithm). The
adjacency list will be computed if not already present.
vertex_pair : cudf.DataFrame
A GPU dataframe consisting of two columns representing pairs of
vertices. If provided, the overlap coefficient is computed for the
given vertex pairs, else, it is computed for all vertex pairs.
Returns
-------
df : cudf.DataFrame
GPU data frame of size E (the default) or the size of the given pairs
(first, second) containing the Overlap coefficients. The ordering is
relative to the adjacency list, or that given by the specified vertex
pairs.
df['source'] : cudf.Series
The source vertex ID (will be identical to first if specified).
df['destination'] : cudf.Series
The destination vertex ID (will be identical to second if
specified).
df['overlap_coeff'] : cudf.Series
The computed Overlap coefficient between the source and destination
vertices.
Examples
--------
>>> gdf = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(gdf, source='0', destination='1')
>>> df = cugraph.overlap(G)
"""
if (type(vertex_pair) == cudf.DataFrame):
null_check(vertex_pair[vertex_pair.columns[0]])
null_check(vertex_pair[vertex_pair.columns[1]])
elif vertex_pair is None:
pass
else:
raise ValueError("vertex_pair must be a cudf dataframe")
df = overlap_wrapper.overlap(input_graph, None, vertex_pair)
return df
def subgraph(G, vertices):
"""
Compute a subgraph of the existing graph including only the specified
vertices. This algorithm works for both directed and undirected graphs,
it does not actually traverse the edges, simply pulls out any edges that
are incident on vertices that are both contained in the vertices list.
Parameters
----------
G : cugraph.Graph
cuGraph graph descriptor
vertices : cudf.Series
Specifies the vertices of the induced subgraph
Returns
-------
Sg : cugraph.Graph
A graph object containing the subgraph induced by the given vertex set.
Examples
--------
>>> gdf = cudf.read_csv('datasets/karate.csv',
delimiter = ' ',
dtype=['int32', 'int32', 'float32'],
header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(gdf, source='0', destination='1')
>>> verts = numpy.zeros(3, dtype=numpy.int32)
>>> verts[0] = 0
>>> verts[1] = 1
>>> verts[2] = 2
>>> sverts = cudf.Series(verts)
>>> Sg = cugraph.subgraph(G, sverts)
"""
null_check(vertices)
if G.renumbered:
vertices = G.lookup_internal_vertex_id(vertices)
result_graph = type(G)()
df = subgraph_extraction_wrapper.subgraph(G, vertices)
if G.renumbered:
df = G.unrenumber(df, "src")
df = G.unrenumber(df, "dst")
if G.edgelist.weights:
result_graph.from_cudf_edgelist(df,
source="src",
destination="dst",
edge_attr="weight")
else:
result_graph.from_cudf_edgelist(df, source="src", destination="dst")
return result_graph
def renumber(source_col, dest_col):
"""
Take a (potentially sparse) set of source and destination vertex ids and
renumber the vertices to create a dense set of vertex ids using all values
contiguously from 0 to the number of unique vertices - 1.
Input columns can be either int64 or int32. The output will be mapped to
int32, since many of the cugraph functions are limited to int32. If the
number of unique values in source_col and dest_col > 2^31-1 then this
function will return an error.
Return from this call will be three cudf Series - the renumbered
source_col, the renumbered dest_col and a numbering map that maps the new
ids to the original ids.
Parameters
----------
source_col : cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the source index for each edge.
Source indices must be an integer type.
dest_col : cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the destination index for each edge.
Destination indices must be an integer type.
numbering_map : cudf.Series
This cudf.Series wraps a gdf column of size V (V: number of vertices).
The gdf column contains a numbering map that maps the new ids to the
original ids.
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> source_col, dest_col, numbering_map = cugraph.renumber(sources,
>>> destinations)
>>> G = cugraph.Graph()
>>> G.add_edge_list(source_col, dest_col, None)
"""
csg.null_check(source_col)
csg.null_check(dest_col)
(source_col, dest_col,
numbering_map) = graph_new_wrapper.renumber(source_col, dest_col)
return source_col, dest_col, numbering_map
def subgraph(G, vertices):
"""
Compute a subgraph of the existing graph including only the specified
vertices. This algorithm works for both directed and undirected graphs,
it does not actually traverse the edges, simply pulls out any edges that
are incident on vertices that are both contained in the vertices list.
Parameters
----------
G : cugraph.Graph
cuGraph graph descriptor
vertices : cudf.Series
Specifies the vertices of the induced subgraph
Returns
-------
Sg : cugraph.Graph
A graph object containing the subgraph induced by the given vertex set.
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> G = cugraph.Graph()
>>> G.add_edge_list(sources, destinations, None)
>>> verts = numpy.zeros(3, dtype=numpy.int32)
>>> verts[0] = 0
>>> verts[1] = 1
>>> verts[2] = 2
>>> sverts = cudf.Series(verts)
>>> Sg = cugraph.subgraph(G, sverts)
"""
null_check(vertices)
result_graph = Graph()
subgraph_extraction_wrapper.subgraph(G.graph_ptr, vertices,
result_graph.graph_ptr)
return result_graph
def symmetrize(source_col, dest_col, value_col=None):
"""
Take a COO set of source destination pairs along with associated values and
create a new COO set of source destination pairs along with values where
all edges exist in both directions.
Return from this call will be a COO stored as two cudf Series - the
symmetrized source column and the symmetrized dest column, along with
an optional cudf Series containing the associated values (only if the
values are passed in).
Parameters
----------
source_col : cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the source index for each edge.
Source indices must be an integer type.
dest_col : cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the destination index for each edge.
Destination indices must be an integer type.
value_col : cudf.Series (optional)
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains values associated with this edge.
For this function the values can be any type, they are not
examined, just copied.
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> values = cudf.Series(M['2'])
>>> src, dst, val = cugraph.symmetrize(sources, destinations, values)
>>> G = cugraph.Graph()
>>> G.add_edge_list(src, dst, val)
"""
csg.null_check(source_col)
csg.null_check(dest_col)
input_df = cudf.DataFrame({"source": source_col, "destination": dest_col})
if value_col is not None:
csg.null_check(value_col)
input_df.insert(len(input_df.columns), "value", value_col)
output_df = symmetrize_df(input_df, "source", "destination")
if value_col is not None:
return (
output_df["source"],
output_df["destination"],
output_df["value"],
)
return output_df["source"], output_df["destination"]
def traveling_salesperson(
pos_list,
restarts=100000,
beam_search=True,
k=4,
nstart=None,
verbose=False,
):
"""
Finds an approximate solution to the traveling salesperson problem (TSP).
cuGraph computes an approximation of the TSP problem using hill climbing
optimization.
The current implementation does not support a weighted graph.
Parameters
----------
pos_list: cudf.DataFrame
Data frame with initial vertex positions containing three columns:
'vertex' ids and 'x', 'y' positions.
restarts: int
Number of starts to try. The more restarts, the better the solution
will be approximated. The number of restarts depends on the problem
size and should be kept low for instances above 2k cities.
beam_search: bool
Specify if the initial solution should use KNN for an approximation
solution.
k: int
Beam width to use in the search.
nstart: int
Vertex id to use as starting position.
verbose: bool
Logs configuration and iterative improvement.
Returns
-------
route : cudf.Series
cudf.Series of size V containing the ordered list of vertices
than needs to be visited.
"""
if not isinstance(pos_list, cudf.DataFrame):
raise TypeError("Instance should be cudf.DataFrame")
null_check(pos_list['vertex'])
null_check(pos_list['x'])
null_check(pos_list['y'])
if nstart is not None and not pos_list[pos_list['vertex'] == nstart].index:
raise ValueError("nstart should be in vertex ids")
route, cost = traveling_salesperson_wrapper.traveling_salesperson(
pos_list, restarts, beam_search, k, nstart, verbose)
return route, cost
def symmetrize(source_col, dest_col, value_col=None, multi=False,
symmetrize=True):
"""
Take a COO set of source destination pairs along with associated values
stored in a single GPU or distributed
create a new COO set of source destination pairs along with values where
all edges exist in both directions.
Return from this call will be a COO stored as two cudf Series or
dask_cudf.Series -the symmetrized source column and the symmetrized dest
column, along with
an optional cudf Series containing the associated values (only if the
values are passed in).
Parameters
----------
source_col : cudf.Series or dask_cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the source index for each edge.
Source indices must be an integer type.
dest_col : cudf.Series or dask_cudf.Series
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains the destination index for each edge.
Destination indices must be an integer type.
value_col : cudf.Series or dask_cudf.Series (optional)
This cudf.Series wraps a gdf_column of size E (E: number of edges).
The gdf column contains values associated with this edge.
For this function the values can be any type, they are not
examined, just copied.
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> values = cudf.Series(M['2'])
>>> src, dst, val = cugraph.symmetrize(sources, destinations, values)
"""
input_df = None
weight_name = None
if type(source_col) is dask_cudf.Series:
# FIXME convoluted way of just wrapping dask cudf Series in a ddf
input_df = source_col.to_frame()
input_df = input_df.rename(columns={source_col.name: "source"})
input_df["destination"] = dest_col
else:
input_df = cudf.DataFrame(
{"source": source_col, "destination": dest_col}
)
csg.null_check(source_col)
csg.null_check(dest_col)
if value_col is not None:
weight_name = "value"
input_df.insert(len(input_df.columns), "value", value_col)
output_df = None
if type(source_col) is dask_cudf.Series:
output_df = symmetrize_ddf(
input_df, "source", "destination", weight_name
).persist()
else:
output_df = symmetrize_df(input_df, "source", "destination", multi,
symmetrize)
if value_col is not None:
return (
output_df["source"],
output_df["destination"],
output_df["value"],
)
return output_df["source"], output_df["destination"]
def jaccard(input_graph, vertex_pair=None):
"""
Compute the Jaccard similarity between each pair of vertices connected by
an edge, or between arbitrary pairs of vertices specified by the user.
Jaccard similarity is defined between two sets as the ratio of the volume
of their intersection divided by the volume of their union. In the context
of graphs, the neighborhood of a vertex is seen as a set. The Jaccard
similarity weight of each edge represents the strength of connection
between vertices based on the relative similarity of their neighbors. If
first is specified but second is not, or vice versa, an exception will be
thrown.
NOTE: If the vertex_pair parameter is not specified then the behavior
of cugraph.jaccard is different from the behavior of
networkx.jaccard_coefficient.
cugraph.jaccard, in the absence of a specified vertex pair list, will
use the edges of the graph to construct a vertex pair list and will
return the jaccard coefficient for those vertex pairs.
networkx.jaccard_coefficient, in the absence of a specified vertex
pair list, will return an upper triangular dense matrix, excluding
the diagonal as well as vertex pairs that are directly connected
by an edge in the graph, of jaccard coefficients. Technically, networkx
returns a lazy iterator across this upper triangular matrix where
the actual jaccard coefficient is computed when the iterator is
dereferenced. Computing a dense matrix of results is not feasible
if the number of vertices in the graph is large (100,000 vertices
would result in 4.9 billion values in that iterator).
If your graph is small enough (or you have enough memory and patience)
you can get the interesting (non-zero) values that are part of the networkx
solution by doing the following:
>>> gdf = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(gdf, source='0', destination='1')
>>> pairs = cugraph.get_two_hop_neighbors(G)
>>> df = cugraph.jaccard(G, pairs)
But please remember that cugraph will fill the dataframe with the entire
solution you request, so you'll need enough memory to store the 2-hop
neighborhood dataframe.
Parameters
----------
graph : cugraph.Graph
cuGraph graph descriptor, should contain the connectivity information
as an edge list (edge weights are not used for this algorithm). The
graph should be undirected where an undirected edge is represented by a
directed edge in both direction. The adjacency list will be computed if
not already present.
vertex_pair : cudf.DataFrame
A GPU dataframe consisting of two columns representing pairs of
vertices. If provided, the jaccard coefficient is computed for the
given vertex pairs. If the vertex_pair is not provided then the
current implementation computes the jaccard coefficient for all
adjacent vertices in the graph.
Returns
-------
df : cudf.DataFrame
GPU data frame of size E (the default) or the size of the given pairs
(first, second) containing the Jaccard weights. The ordering is
relative to the adjacency list, or that given by the specified vertex
pairs.
df['source'] : cudf.Series
The source vertex ID (will be identical to first if specified)
df['destination'] : cudf.Series
The destination vertex ID (will be identical to second if
specified)
df['jaccard_coeff'] : cudf.Series
The computed Jaccard coefficient between the source and destination
vertices
Examples
--------
>>> gdf = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(gdf, source='0', destination='1')
>>> df = cugraph.jaccard(G)
"""
if type(input_graph) is not Graph:
raise Exception("input graph must be undirected")
# FIXME: Add support for multi-column vertices
if type(vertex_pair) == cudf.DataFrame:
for col in vertex_pair.columns:
null_check(vertex_pair[col])
if input_graph.renumbered:
vertex_pair = input_graph.add_internal_vertex_id(
vertex_pair, col, col)
elif vertex_pair is None:
pass
else:
raise ValueError("vertex_pair must be a cudf dataframe")
df = jaccard_wrapper.jaccard(input_graph, None, vertex_pair)
if input_graph.renumbered:
df = input_graph.unrenumber(df, "source")
df = input_graph.unrenumber(df, "destination")
return df
def jaccard(input_graph, first=None, second=None):
"""
Compute the Jaccard similarity between each pair of vertices connected by
an edge, or between arbitrary pairs of vertices specified by the user.
Jaccard similarity is defined between two sets as the ratio of the volume
of their intersection divided by the volume of their union. In the context
of graphs, the neighborhood of a vertex is seen as a set. The Jaccard
similarity weight of each edge represents the strength of connection
between vertices based on the relative similarity of their neighbors. If
first is specified but second is not, or vice versa, an exception will be
thrown.
Parameters
----------
graph : cugraph.Graph
cuGraph graph descriptor, should contain the connectivity information
as an edge list (edge weights are not used for this algorithm). The
graph should be undirected where an undirected edge is represented by a
directed edge in both direction. The adjacency list will be computed if
not already present.
first : cudf.Series
Specifies the first vertices of each pair of vertices to compute for,
must be specified along with second.
second : cudf.Series
Specifies the second vertices of each pair of vertices to compute for,
must be specified along with first.
Returns
-------
df : cudf.DataFrame
GPU data frame of size E (the default) or the size of the given pairs
(first, second) containing the Jaccard weights. The ordering is
relative to the adjacency list, or that given by the specified vertex
pairs.
df['source'] : cudf.Series
The source vertex ID (will be identical to first if specified)
df['destination'] : cudf.Series
The destination vertex ID (will be identical to second if
specified)
df['jaccard_coeff'] : cudf.Series
The computed Jaccard coefficient between the source and destination
vertices
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> G = cugraph.Graph()
>>> G.add_edge_list(sources, destinations, None)
>>> df = cugraph.jaccard(G)
"""
if (type(first) == cudf.Series and type(second) == cudf.Series):
null_check(first)
null_check(second)
elif first is None and second is None:
pass
else:
raise ValueError("Specify first and second or neither")
df = jaccard_wrapper.jaccard(input_graph.graph_ptr, first, second)
return df
def force_atlas2(input_graph,
max_iter=500,
pos_list=None,
outbound_attraction_distribution=True,
lin_log_mode=False,
prevent_overlapping=False,
edge_weight_influence=1.0,
jitter_tolerance=1.0,
barnes_hut_optimize=True,
barnes_hut_theta=0.5,
scaling_ratio=2.0,
strong_gravity_mode=False,
gravity=1.0,
verbose=False,
callback=None):
"""
ForceAtlas2 is a continuous graph layout algorithm for handy network
visualization.
NOTE: Peak memory allocation occurs at 17*V.
Parameters
----------
input_graph : cugraph.Graph
cuGraph graph descriptor with connectivity information.
Edge weights, if present, should be single or double precision
floating point values.
max_iter : integer
This controls the maximum number of levels/iterations of the Force
Atlas algorithm. When specified the algorithm will terminate after
no more than the specified number of iterations.
No error occurs when the algorithm terminates early in this manner.
Good short-term quality can be achieved with 50-100 iterations.
Above 1000 iterations is discouraged.
pos_list: cudf.DataFrame
Data frame with initial vertex positions containing two columns:
'x' and 'y' positions.
outbound_attraction_distribution: bool
Distributes attraction along outbound edges.
Hubs attract less and thus are pushed to the borders.
lin_log_mode: bool
Switch Force Atlas model from lin-lin to lin-log.
Makes clusters more tight.
prevent_overlapping: bool
Prevent nodes to overlap.
edge_weight_influence: float
How much influence you give to the edges weight.
0 is “no influence” and 1 is “normal”.
jitter_tolerance: float
How much swinging you allow. Above 1 discouraged.
Lower gives less speed and more precision.
barnes_hut_theta: float
Float between 0 and 1. Tradeoff for speed (1) vs
accuracy (0) for Barnes Hut only.
scaling_ratio: float
How much repulsion you want. More makes a more sparse graph.
Switching from regular mode to LinLog mode needs a readjustment
of the scaling parameter.
gravity : float
Attracts nodes to the center. Prevents islands from drifting away.
verbose: bool
Output convergence info at each interation.
callback: GraphBasedDimRedCallback
An instance of GraphBasedDimRedCallback class to intercept
the internal state of positions while they are being trained.
Example of callback usage:
from cugraph.layout import GraphBasedDimRedCallback
class CustomCallback(GraphBasedDimRedCallback):
def on_preprocess_end(self, positions):
print(positions.copy_to_host())
def on_train_end(self, positions):
print(positions.copy_to_host())
def on_train_end(self, positions):
print(positions.copy_to_host())
Returns
-------
pos : cudf.DataFrame
GPU data frame of size V containing three columns:
the vertex identifiers and the x and y positions.
"""
if pos_list is not None:
null_check(pos_list['vertex'])
null_check(pos_list['x'])
null_check(pos_list['y'])
if prevent_overlapping:
raise Exception("Feature not supported")
if input_graph.is_directed():
input_graph = input_graph.to_undirected()
pos = force_atlas2_wrapper.force_atlas2(
input_graph,
max_iter=max_iter,
pos_list=pos_list,
outbound_attraction_distribution=outbound_attraction_distribution,
lin_log_mode=lin_log_mode,
prevent_overlapping=prevent_overlapping,
edge_weight_influence=edge_weight_influence,
jitter_tolerance=jitter_tolerance,
barnes_hut_optimize=barnes_hut_optimize,
barnes_hut_theta=barnes_hut_theta,
scaling_ratio=scaling_ratio,
strong_gravity_mode=strong_gravity_mode,
gravity=gravity,
verbose=verbose,
callback=callback)
return pos
def pagerank(G,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-5,
nstart=None):
"""
Find the PageRank score for every vertex in a graph. cuGraph computes an
approximation of the Pagerank eigenvector using the power method. The
number of iterations depends on the properties of the network itself; it
increases when the tolerance descreases and/or alpha increases toward the
limiting value of 1. The user is free to use default values or to provide
inputs for the initial guess, tolerance and maximum number of iterations.
Parameters
----------
graph : cugraph.Graph
cuGraph graph descriptor, should contain the connectivity information
as an edge list (edge weights are not used for this algorithm).
The transposed adjacency list will be computed if not already present.
alpha : float
The damping factor alpha represents the probability to follow an
outgoing edge, standard value is 0.85.
Thus, 1.0-alpha is the probability to “teleport” to a random vertex.
Alpha should be greater than 0.0 and strictly lower than 1.0.
personalization : cudf.Dataframe
GPU Dataframe containing the personalization information.
personalization['vertex'] : cudf.Series
Subset of vertices of graph for personalization
personalization['values'] : cudf.Series
Personalization values for vertices
max_iter : int
The maximum number of iterations before an answer is returned. This can
be used to limit the execution time and do an early exit before the
solver reaches the convergence tolerance.
If this value is lower or equal to 0 cuGraph will use the default
value, which is 100.
tolerance : float
Set the tolerance the approximation, this parameter should be a small
magnitude value.
The lower the tolerance the better the approximation. If this value is
0.0f, cuGraph will use the default value which is 1.0E-5.
Setting too small a tolerance can lead to non-convergence due to
numerical roundoff. Usually values between 0.01 and 0.00001 are
acceptable.
nstart : cudf.Dataframe
GPU Dataframe containing the initial guess for pagerank.
nstart['vertex'] : cudf.Series
Subset of vertices of graph for initial guess for pagerank values
nstart['values'] : cudf.Series
Pagerank values for vertices
Returns
-------
PageRank : cudf.DataFrame
GPU data frame containing two cudf.Series of size V: the vertex
identifiers and the corresponding PageRank values.
df['vertex'] : cudf.Series
Contains the vertex identifiers
df['pagerank'] : cudf.Series
Contains the PageRank score
Examples
--------
>>> gdf = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> G = cugraph.Graph()
>>> G.from_cudf_edgelist(gdf, source='0', destination='1')
>>> pr = cugraph.pagerank(G, alpha = 0.85, max_iter = 500, tol = 1.0e-05)
"""
if personalization is not None:
null_check(personalization["vertex"])
null_check(personalization["values"])
if G.renumbered is True:
personalization = G.add_internal_vertex_id(personalization,
"vertex", "vertex")
if nstart is not None:
if G.renumbered is True:
nstart = G.add_internal_vertex_id(nstart, "vertex", "vertex")
df = pagerank_wrapper.pagerank(G, alpha, personalization, max_iter, tol,
nstart)
if G.renumbered:
df = G.unrenumber(df, "vertex")
return df
def pagerank(input_graph,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-5,
nstart=None,
load_balance=True):
"""
Find the PageRank values for each vertex in a graph using multiple GPUs.
cuGraph computes an approximation of the Pagerank using the power method.
The input graph must contain edge list as dask-cudf dataframe with
one partition per GPU.
Parameters
----------
graph : cugraph.DiGraph
cuGraph graph descriptor, should contain the connectivity information
as dask cudf edge list dataframe(edge weights are not used for this
algorithm). Undirected Graph not currently supported.
alpha : float
The damping factor alpha represents the probability to follow an
outgoing edge, standard value is 0.85.
Thus, 1.0-alpha is the probability to “teleport” to a random vertex.
Alpha should be greater than 0.0 and strictly lower than 1.0.
personalization : cudf.Dataframe
GPU Dataframe containing the personalization information.
personalization['vertex'] : cudf.Series
Subset of vertices of graph for personalization
personalization['values'] : cudf.Series
Personalization values for vertices
max_iter : int
The maximum number of iterations before an answer is returned.
If this value is lower or equal to 0 cuGraph will use the default
value, which is 30.
tolerance : float
Set the tolerance the approximation, this parameter should be a small
magnitude value.
The lower the tolerance the better the approximation. If this value is
0.0f, cuGraph will use the default value which is 1.0E-5.
Setting too small a tolerance can lead to non-convergence due to
numerical roundoff. Usually values between 0.01 and 0.00001 are
acceptable.
nstart : not supported
initial guess for pagerank
load_balance : bool
Set as True to perform load_balancing after global sorting of
dask-cudf DataFrame. This ensures that the data is uniformly
distributed among multiple GPUs to avoid over-loading.
Returns
-------
PageRank : cudf.DataFrame
GPU data frame containing two cudf.Series of size V: the vertex
identifiers and the corresponding PageRank values.
df['vertex'] : cudf.Series
Contains the vertex identifiers
df['pagerank'] : cudf.Series
Contains the PageRank score
Examples
--------
>>> import cugraph.dask as dcg
>>> Comms.initialize()
>>> chunksize = dcg.get_chunksize(input_data_path)
>>> ddf = dask_cudf.read_csv(input_data_path, chunksize=chunksize,
delimiter=' ',
names=['src', 'dst', 'value'],
dtype=['int32', 'int32', 'float32'])
>>> dg = cugraph.DiGraph()
>>> dg.from_dask_cudf_edgelist(ddf, source='src', destination='dst',
edge_attr='value')
>>> pr = dcg.pagerank(dg)
>>> Comms.destroy()
"""
from cugraph.structure.graph import null_check
nstart = None
client = default_client()
if (input_graph.local_data is not None
and input_graph.local_data['by'] == 'dst'):
data = input_graph.local_data['data']
else:
data = get_local_data(input_graph, by='dst', load_balance=load_balance)
if personalization is not None:
null_check(personalization["vertex"])
null_check(personalization["values"])
if input_graph.renumbered is True:
personalization = input_graph.add_internal_vertex_id(
personalization, "vertex", "vertex").compute()
result = dict([(data.worker_info[wf[0]]["rank"],
client.submit(call_pagerank,
Comms.get_session_id(),
wf[1],
data.local_data,
alpha,
max_iter,
tol,
personalization,
nstart,
workers=[wf[0]]))
for idx, wf in enumerate(data.worker_to_parts.items())])
wait(result)
if input_graph.renumbered:
return input_graph.unrenumber(result[0].result(), 'vertex').compute()
return result[0].result()
def pagerank(input_graph,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-5,
nstart=None):
"""
Find the PageRank values for each vertex in a graph using multiple GPUs.
cuGraph computes an approximation of the Pagerank using the power method.
The input graph must contain edge list as dask-cudf dataframe with
one partition per GPU.
Parameters
----------
graph : cugraph.DiGraph
cuGraph graph descriptor, should contain the connectivity information
as dask cudf edge list dataframe(edge weights are not used for this
algorithm). Undirected Graph not currently supported.
alpha : float
The damping factor alpha represents the probability to follow an
outgoing edge, standard value is 0.85.
Thus, 1.0-alpha is the probability to “teleport” to a random vertex.
Alpha should be greater than 0.0 and strictly lower than 1.0.
personalization : cudf.Dataframe
GPU Dataframe containing the personalization information.
Currently not supported.
personalization['vertex'] : cudf.Series
Subset of vertices of graph for personalization
personalization['values'] : cudf.Series
Personalization values for vertices
max_iter : int
The maximum number of iterations before an answer is returned.
If this value is lower or equal to 0 cuGraph will use the default
value, which is 30.
tolerance : float
Set the tolerance the approximation, this parameter should be a small
magnitude value.
The lower the tolerance the better the approximation. If this value is
0.0f, cuGraph will use the default value which is 1.0E-5.
Setting too small a tolerance can lead to non-convergence due to
numerical roundoff. Usually values between 0.01 and 0.00001 are
acceptable.
nstart : not supported
initial guess for pagerank
Returns
-------
PageRank : dask_cudf.DataFrame
GPU data frame containing two dask_cudf.Series of size V: the
vertex identifiers and the corresponding PageRank values.
ddf['vertex'] : dask_cudf.Series
Contains the vertex identifiers
ddf['pagerank'] : dask_cudf.Series
Contains the PageRank score
Examples
--------
>>> import cugraph.dask as dcg
>>> ... Init a DASK Cluster
>> see https://docs.rapids.ai/api/cugraph/stable/dask-cugraph.html
>>> chunksize = dcg.get_chunksize(input_data_path)
>>> ddf = dask_cudf.read_csv(input_data_path, chunksize=chunksize,
delimiter=' ',
names=['src', 'dst', 'value'],
dtype=['int32', 'int32', 'float32'])
>>> dg = cugraph.DiGraph()
>>> dg.from_dask_cudf_edgelist(ddf, source='src', destination='dst',
edge_attr='value')
>>> pr = dcg.pagerank(dg)
"""
from cugraph.structure.graph import null_check
nstart = None
client = default_client()
input_graph.compute_renumber_edge_list(transposed=True)
ddf = input_graph.edgelist.edgelist_df
vertex_partition_offsets = get_vertex_partition_offsets(input_graph)
num_verts = vertex_partition_offsets.iloc[-1]
num_edges = len(ddf)
data = get_distributed_data(ddf)
if personalization is not None:
null_check(personalization["vertex"])
null_check(personalization["values"])
if input_graph.renumbered is True:
personalization = input_graph.add_internal_vertex_id(
personalization, "vertex", "vertex")
p_data = get_distributed_data(personalization)
result = [
client.submit(call_pagerank,
Comms.get_session_id(),
wf[1],
num_verts,
num_edges,
vertex_partition_offsets,
alpha,
max_iter,
tol,
p_data.worker_to_parts[wf[0]][0],
nstart,
workers=[wf[0]])
for idx, wf in enumerate(data.worker_to_parts.items())
]
else:
result = [
client.submit(call_pagerank,
Comms.get_session_id(),
wf[1],
num_verts,
num_edges,
vertex_partition_offsets,
alpha,
max_iter,
tol,
personalization,
nstart,
workers=[wf[0]])
for idx, wf in enumerate(data.worker_to_parts.items())
]
wait(result)
ddf = dask_cudf.from_delayed(result)
if input_graph.renumbered:
return input_graph.unrenumber(ddf, 'vertex')
return ddf
def jaccard_w(input_graph, weights, vertex_pair=None):
"""
Compute the weighted Jaccard similarity between each pair of vertices
connected by an edge, or between arbitrary pairs of vertices specified by
the user. Jaccard similarity is defined between two sets as the ratio of
the volume of their intersection divided by the volume of their union. In
the context of graphs, the neighborhood of a vertex is seen as a set. The
Jaccard similarity weight of each edge represents the strength of
connection between vertices based on the relative similarity of their
neighbors. If first is specified but second is not, or vice versa, an
exception will be thrown.
Parameters
----------
graph : cugraph.Graph
cuGraph graph descriptor, should contain the connectivity information
as an edge list (edge weights are not used for this algorithm). The
adjacency list will be computed if not already present.
weights : cudf.Series
Specifies the weights to be used for each vertex.
vertex_pair : cudf.DataFrame
A GPU dataframe consisting of two columns representing pairs of
vertices. If provided, the jaccard coefficient is computed for the
given vertex pairs, else, it is computed for all vertex pairs.
Returns
-------
df : cudf.DataFrame
GPU data frame of size E (the default) or the size of the given pairs
(first, second) containing the Jaccard weights. The ordering is
relative to the adjacency list, or that given by the specified vertex
pairs.
df['source'] : cudf.Series
The source vertex ID
df['destination'] : cudf.Series
The destination vertex ID
df['jaccard_coeff'] : cudf.Series
The computed weighted Jaccard coefficient between the source and
destination vertices.
Examples
--------
>>> M = cudf.read_csv('datasets/karate.csv', delimiter=' ',
>>> dtype=['int32', 'int32', 'float32'], header=None)
>>> sources = cudf.Series(M['0'])
>>> destinations = cudf.Series(M['1'])
>>> weights = cudf.Series(numpy.ones(
>>> max(sources.max(),destinations.max())+1, dtype=numpy.float32))
>>> G = cugraph.Graph()
>>> G.add_edge_list(sources, destinations, None)
>>> df = cugraph.jaccard_w(G, weights)
"""
if type(input_graph) is not Graph:
raise Exception("input graph must be undirected")
if (type(vertex_pair) == cudf.DataFrame):
null_check(vertex_pair[vertex_pair.columns[0]])
null_check(vertex_pair[vertex_pair.columns[1]])
elif vertex_pair is None:
pass
else:
raise ValueError("vertex_pair must be a cudf dataframe")
df = jaccard_wrapper.jaccard(input_graph, weights, vertex_pair)
return df
