def example1():
uinvite_network, ll = snafu.uinvite(
fluencydata.lists[0], # provide fluency lists
datamodel, # specify data model
fitinfo=fitinfo, # specify fit info
debug=True) # suppress print output to console when set to False
return uinvite_network
def example3():
usf_network, usf_items = snafu.load_network(
"../snet/USF_animal_subset.snet")
# Here you can specify multiple networks as a prior; the first parameter is
# a list of networks, the second parameter is a list of dictionaries that
# map indices to items in each network
usf_prior = snafu.genGraphPrior([usf_network], [usf_items])
uinvite_network, ll = snafu.uinvite(fluencydata.lists[0],
prior=(usf_prior,
fluencydata.items[0]))
return uinvite_network
def network_properties(command, root_path):
subj_props = command['data_parameters']
command = command['network_parameters']
# U-INVITE won't work with perseverations
if command['network_method'] == "U-INVITE":
removePerseverations=True
else:
removePerseverations=False
if subj_props['factor_type'] == "subject":
ids = str(subj_props['subject'])
group = False
elif subj_props['factor_type'] == "group":
ids = str(subj_props['group']) # without str() causes unicode issues for "all" :(
group = True
filedata = snafu.readX(ids, subj_props['fullpath'], category=subj_props['category'], spellfile=label_to_filepath(subj_props['spellfile'], root_path, "spellfiles"), removePerseverations=removePerseverations, group=group)
filedata.nonhierarchical()
Xs = filedata.Xs
items = filedata.items
irts = filedata.irts
numnodes = filedata.numnodes
toydata=snafu.DataModel({
'numx': len(Xs),
'trim': 1,
'jump': float(command['jump_probability']),
'jumptype': command['jump_type'],
'priming': float(command['priming_probability']),
'startX': command['first_item']})
fitinfo=snafu.Fitinfo({
'prior_method': "zeroinflatedbetabinomial",
'prior_a': 1,
'prior_b': 2,
'zibb_p': 0.5,
'startGraph': command['starting_graph'],
'goni_size': int(command['goni_windowsize']),
'goni_threshold': int(command['goni_threshold']),
'followtype': "avg",
'prune_limit': 100,
'triangle_limit': 100,
'other_limit': 100})
if command['prior']=="None":
prior=None
elif command['prior']=="USF":
usf_file_path = "/snet/USF_animal_subset.snet"
filename = root_path + usf_file_path
usf_graph, usf_items = snafu.read_graph(filename)
usf_numnodes = len(usf_items)
priordict = snafu.genGraphPrior([usf_graph], [usf_items], fitinfo=fitinfo)
prior = (priordict, usf_items)
if command['network_method']=="RW":
bestgraph = snafu.noHidden(Xs, numnodes)
elif command['network_method']=="Goni":
bestgraph = snafu.goni(Xs, numnodes, td=toydata, valid=0, fitinfo=fitinfo)
elif command['network_method']=="Chan":
bestgraph = snafu.chan(Xs, numnodes)
elif command['network_method']=="Kenett":
bestgraph = snafu.kenett(Xs, numnodes)
elif command['network_method']=="FirstEdge":
bestgraph = snafu.firstEdge(Xs, numnodes)
elif command['network_method']=="U-INVITE":
bestgraph, ll = snafu.uinvite(Xs, toydata, numnodes, fitinfo=fitinfo, debug=False, prior=prior)
nxg = nx.to_networkx_graph(bestgraph)
nxg_json = jsonGraph(nxg, items)
return graph_properties(nxg,nxg_json)
for sub in subs:
Xs, items, irtdata, numnodes = snafu.readX(
sub,
category,
filepath,
removePerseverations=True,
spellfile="spellfiles/zemla_spellfile.csv")
if method == "rw":
graph = snafu.nrw(Xs, numnodes)
if method == "goni":
graph = snafu.goni(Xs, numnodes, fitinfo=fitinfo)
if method == "chan":
graph = snafu.chan(Xs, numnodes)
if method == "kenett":
graph = snafu.kenett(Xs, numnodes)
if method == "fe":
graph = snafu.firstEdge(Xs, numnodes)
if method == "uinvite_flat":
graph, ll = snafu.uinvite(Xs, toydata, numnodes, fitinfo=fitinfo)
for i in range(len(graph)):
for j in range(len(graph)):
if i > j:
item1 = items[i]
item2 = items[j]
itempair = np.sort([item1, item2])
fo.write(sub + "," + method + "," + itempair[0] + "," +
itempair[1] + "," + str(graph[i, j]) + "\n")
fo.close()
graph = snafu.goni(Xs_flat, groupnumnodes, fitinfo=fitinfo)
# Estimate the best network using Chan
if method=="chan":
graph = snafu.chan(Xs_flat, groupnumnodes)
# Estimate the best network using Kenett
if method=="kenett":
graph = snafu.kenett(Xs_flat, groupnumnodes)
# Estimate the best network using First-Edge
if method=="fe":
graph = snafu.firstEdge(Xs_flat, groupnumnodes)
# Estimate the best network using a non-hierarchical U-INVITE
if method=="uinvite_flat":
graph, ll = snafu.uinvite(Xs_flat, toydata, groupnumnodes, fitinfo=fitinfo)
# Estimate the best network using hierarchical U-INVITE
if method=="uinvite_hierarchical":
sub_graphs, priordict = snafu.hierarchicalUinvite(Xs_hier, items, numnodes, toydata, fitinfo=fitinfo)
graph = snafu.priorToGraph(priordict, groupitems)
# convert numpy matrix to networkx graph and replace indices with semantic labels
graph = nx.to_networkx_graph(graph)
nx.relabel_nodes(graph, groupitems, copy=False)
graphs.append(graph)
header=','.join(methods)
snafu.write_graph(graphs, "human_graphs.csv",header=header)
# Generate Chan graph from data
if 'chan' in methods:
chan_graph = snafu.chan(flatdata, usf_numnodes)
# Generate Kenett graph from data
if 'kenett' in methods:
kenett_graph = snafu.kenett(flatdata, usf_numnodes)
# Generate First Edge graph from data
if 'fe' in methods:
fe_graph = snafu.firstEdge(flatdata, usf_numnodes)
# Generate non-hierarchical U-INVITE graph from data
if 'uinvite_flat' in methods:
uinvite_flat_graph, ll = snafu.uinvite(flatdata, toydata, usf_numnodes, fitinfo=fitinfo)
# Generate hierarchical U-INVITE graph from data
if 'uinvite_hierarchical' in methods:
uinvite_graphs, priordict = snafu.hierarchicalUinvite(data_hier[:ssnum], items[:ssnum], numnodes[:ssnum], toydata, fitinfo=fitinfo)
# U-INVITE paper uses an added "threshold" such that at least 2 participants must have an edge for it to be in the group network
# So rather than using the same prior as the one used during fitting, we have to generate a new one
priordict = snafu.genGraphPrior(uinvite_graphs, items[:ssnum], fitinfo=fitinfo, mincount=2)
# Generate group graph from the prior
uinvite_group_graph = snafu.priorToGraph(priordict, usf_items)
# Write data to file!
for method in methods:
if method=="rw": costlist = [snafu.costSDT(rw_graph, usf_graph), snafu.cost(rw_graph, usf_graph)]
# Generate Chan graph from data
if 'chan' in methods:
chan_graph = snafu.chan(flatdata, usf_numnodes)
# Generate Kenett graph from data
if 'kenett' in methods:
kenett_graph = snafu.kenett(flatdata, usf_numnodes)
# Generate First Edge graph from data
if 'fe' in methods:
fe_graph = snafu.firstEdge(flatdata, usf_numnodes)
# Generate non-hierarchical U-INVITE graph from data
if 'uinvite_flat' in methods:
uinvite_flat_graph, ll = snafu.uinvite(flatdata,
toydata,
usf_numnodes,
fitinfo=fitinfo)
# Generate hierarchical U-INVITE graph from data
if 'uinvite_hierarchical' in methods:
uinvite_graphs, priordict = snafu.hierarchicalUinvite(
data_hier[:ssnum],
items[:ssnum],
numnodes[:ssnum],
toydata,
fitinfo=fitinfo)
# U-INVITE paper uses an added "threshold" such that at least 2 participants must have an edge for it to be in the group network
# So rather than using the same prior as the one used during fitting, we have to generate a new one
priordict = snafu.genGraphPrior(uinvite_graphs,
items[:ssnum],
def network_properties(command, root_path):
subj_props = command['data_parameters']
command = command['network_parameters']
# U-INVITE won't work with perseverations
if command['network_method'] == "U-INVITE":
removePerseverations = True
else:
removePerseverations = False
if subj_props['factor_type'] == "subject":
Xs, items, irts, numnodes = snafu.readX(
subj_props['subject'],
subj_props['category'],
subj_props['fullpath'],
spellfile=label_to_filepath(subj_props['spellfile'], root_path,
"spellfiles"),
removePerseverations=removePerseverations)
elif subj_props['factor_type'] == "group":
Xs, items, irts, numnodes = snafu.readX(
subj_props['subject'],
subj_props['category'],
subj_props['fullpath'],
spellfile=label_to_filepath(subj_props['spellfile'], root_path,
"spellfiles"),
removePerseverations=removePerseverations,
group=subj_props['group'],
flatten=True)
toydata = snafu.Data({
'numx': len(Xs),
'trim': 1,
'jump': float(command['jump_probability']),
'jumptype': command['jump_type'],
'priming': float(command['priming_probability']),
'startX': command['first_item']
})
fitinfo = snafu.Fitinfo({
'prior_method': "betabinomial",
'prior_a': 1,
'prior_b': 1,
'startGraph': command['starting_graph'],
'goni_size': int(command['goni_windowsize']),
'goni_threshold': int(command['goni_threshold']),
'followtype': "avg",
'prune_limit': 100,
'triangle_limit': 100,
'other_limit': 100
})
if command['prior'] == "None":
prior = None
elif command['prior'] == "USF":
usf_file_path = "/snet/USF_animal_subset.snet"
filename = root_path + usf_file_path
usf_graph, usf_items = snafu.read_graph(filename)
usf_numnodes = len(usf_items)
priordict = snafu.genGraphPrior([usf_graph], [usf_items],
fitinfo=fitinfo)
prior = (priordict, usf_items)
if command['network_method'] == "RW":
bestgraph = snafu.noHidden(Xs, numnodes)
elif command['network_method'] == "Goni":
bestgraph = snafu.goni(Xs,
numnodes,
td=toydata,
valid=0,
fitinfo=fitinfo)
elif command['network_method'] == "Chan":
bestgraph = snafu.chan(Xs, numnodes)
elif command['network_method'] == "Kenett":
bestgraph = snafu.kenett(Xs, numnodes)
elif command['network_method'] == "FirstEdge":
bestgraph = snafu.firstEdge(Xs, numnodes)
elif command['network_method'] == "U-INVITE":
bestgraph, ll = snafu.uinvite(Xs,
toydata,
numnodes,
fitinfo=fitinfo,
debug=False,
prior=prior)
nxg = nx.to_networkx_graph(bestgraph)
node_degree = np.mean(dict(nxg.degree()).values())
nxg_json = jsonGraph(nxg, items)
clustering_coefficient = nx.average_clustering(nxg)
try:
aspl = nx.average_shortest_path_length(nxg)
except:
aspl = "disjointed graph"
return {
"type": "network_properties",
"node_degree": node_degree,
"clustering_coefficient": clustering_coefficient,
"aspl": aspl,
"graph": nxg_json
}
graph = snafu.goni(Xs_flat, groupnumnodes, fitinfo=fitinfo)
# Estimate the best network using Chan
if method=="chan":
graph = snafu.chan(Xs_flat, groupnumnodes)
# Estimate the best network using Kenett
if method=="kenett":
graph = snafu.kenett(Xs_flat, groupnumnodes)
# Estimate the best network using First-Edge
if method=="fe":
graph = snafu.firstEdge(Xs_flat, groupnumnodes)
# Estimate the best network using a non-hierarchical U-INVITE
if method=="uinvite_flat":
graph, ll = snafu.uinvite(Xs_flat, toydata, groupnumnodes, fitinfo=fitinfo)
# Estimate the best network using hierarchical U-INVITE
if method=="uinvite_hierarchical":
sub_graphs, priordict = snafu.hierarchicalUinvite(Xs_hier, items, numnodes, toydata, fitinfo=fitinfo)
graph = snafu.priorToGraph(priordict, groupitems)
# convert numpy matrix to networkx graph and replace indices with semantic labels
graph = nx.to_networkx_graph(graph)
nx.relabel_nodes(graph, groupitems, copy=False)
graphs.append(graph)
header=','.join(methods)
snafu.write_graph(graphs, "human_graphs.csv",header=header)