def findKeyAttrs(samples, protect_attr='', result_attr='class'):
"""
Args:
samples(pandas DataFrame):
protect_attr(string || Array<string>):
Return:
key_attrs(list<string>): a list of key attributes that directly influence the decision
"""
from pycausal.pycausal import pycausal as pc
pc = pc()
pc.start_vm()
from pycausal import search as s
# pc import must keep the above order
# choose a causal mining algorithm
causal = 'fges'
if causal == 'bayes':
### use bayes Est to find the key attributes
### somewhat slow, extract more key attributes
graph = s.bayesEst(samples, depth=0, alpha=0.05, verbose=True)
else:
## OR use Fast Greedy Equivalence Search
## faster than bayes, get less key attributes
graph = s.tetradrunner()
graph.getAlgorithmParameters(algoId='fges', scoreId='bdeu')
graph.run(algoId='fges',
dfs=samples,
scoreId='bdeu',
priorKnowledge=None,
dataType='discrete',
structurePrior=0.5,
samplePrior=0.5,
maxDegree=5,
faithfulnessAssumed=True,
verbose=False)
# graph.getNodes()
key_attrs = []
print('edges', graph.getEdges())
for edge in graph.getEdges():
if 'class' in edge:
# extract attr name from the edge
# remove --> or --o or --- and white space
attr = re.sub(r'-+>?o?|{}|\s+'.format(result_attr), '', edge)
key_attrs.append(attr)
# remove protect attrs
if type(protect_attr) is not str:
# if protect attr is a list
for a in protect_attr:
if a in key_attrs:
key_attrs.remove(a)
elif protect_attr in key_attrs:
# if protect attr is a string
key_attrs.remove(protect_attr)
print('key attributes', key_attrs)
return key_attrs
def py_causal(
sp_pro_deg_v, knowledge_path, bp=False,
dataType="discrete", algoId="fges",
scoreId="disc-bic-score", structurePrior=1.0, samplePrior=1.0, maxDegree=20,
faithfulnessAssumed=True, symmetricFirstStep=True, verbose=True,
numberResampling=100, percentResampleSize=90, resamplingEnsemble=1,
addOriginalDataset=True, resamplingWithReplacement=True):
"""Notes for py_causal.
Args:
sp_sga_deg: 2-D df, index are sample IDs, columns are SGAs and DGEs names.
df's value is 0 or 1.
bp: bool, using bootstrap. Default is False.
Returns:
node_l: A list of contain nodes in network.
edge_l: A 2-D df contains two columns, source node to target node,
didn't contain edge without direction.
bic: fges output, contain nodes, edges, network score.
"""
#avoid changing input file
sp_pro_deg = deepcopy(sp_pro_deg_v)
# connect to java
from pycausal.pycausal import pycausal as pc
pc = pc()
pc.start_vm(java_max_heap_size="1000M")
# generate knowledge
from pycausal import prior as p
prior = p.knowledgeFromFile(knowledge_path)
# search
from pycausal import search as s
tetrad = s.tetradrunner()
if bp == True:
tetrad.run(
dfs=sp_pro_deg, priorKnowledge=prior,
dataType=dataType, algoId=algoId,
scoreId=scoreId, structurePrior=structurePrior, samplePrior=samplePrior, maxDegree=maxDegree,
faithfulnessAssumed=faithfulnessAssumed, symmetricFirstStep=symmetricFirstStep, verbose=verbose,
numberResampling=numberResampling, percentResampleSize=percentResampleSize, resamplingEnsemble=resamplingEnsemble,
addOriginalDataset=addOriginalDataset, resamplingWithReplacement=resamplingWithReplacement)
else:
tetrad.run(
dfs=sp_pro_deg, priorKnowledge=prior,
dataType=dataType, algoId=algoId,
scoreId=scoreId, structurePrior=structurePrior, samplePrior=samplePrior, maxDegree=maxDegree,
faithfulnessAssumed=faithfulnessAssumed, symmetricFirstStep=symmetricFirstStep, verbose=verbose)
node_l = tetrad.getNodes()
edge_l = tetrad.getEdges()
bic = tetrad.getTetradGraph()
return node_l, edge_l, bic
def init_causal_graph_dot_src(self, df, forbidden_edges, required_edges):
p = pc()
p.start_vm()
tetrad = s.tetradrunner()
prior = pr.knowledge(forbiddirect = forbidden_edges, requiredirect = required_edges)
tetrad.run(algoId = 'fges', dfs = df, priorKnowledge = prior, scoreId = 'sem-bic', dataType = 'continuous', penaltyDiscount = 2, maxDegree = -1, faithfulnessAssumed = True, verbose = True)
dot_src = p.tetradGraphToDot(tetrad.getTetradGraph())
#p.stop_vm()
self.edges = tetrad.getEdges()
self.nodes = tetrad.getNodes()
dot_src = self.trim_init_src_string(dot_src)
self.dot_src_lines = self.dot_src_to_lines(dot_src)
self.dot_src = self.lines_to_dot_src(self.dot_src_lines)
self.init_dot_src = self.dot_src
self.uncolored_dot_src = self.init_dot_src
def learn_fci(self, df, tabu_edges):
"""This function is used to learn model using FCI"""
from pycausal.pycausal import pycausal as pc
from pycausal import search as s
from pycausal import prior as p
pc = pc()
pc.start_vm()
forbid = [list(i) for i in tabu_edges]
prior = p.knowledge(forbiddirect=forbid)
tetrad = s.tetradrunner()
tetrad.getAlgorithmParameters(algoId='fci', testId='fisher-z-test')
tetrad.run(algoId='fci',
dfs=df,
testId='fisher-z-test',
depth=-1,
maxPathLength=-1,
completeRuleSetUsed=False,
verbose=False)
edges = tetrad.getEdges()
dot_str = pc.tetradGraphToDot(tetrad.getTetradGraph())
graph = pydot.graph_from_dot_data(dot_str)
# graph[0].write_pdf(fname)
pc.stop_vm()
return edges
#!/usr/local/bin/python
import os
import pandas as pd
import pydot
from IPython.display import SVG
data_dir = os.path.join(os.getcwd(), 'data', 'audiology.txt')
df = pd.read_table(data_dir, sep="\t")
from pycausal.pycausal import pycausal as pc
pc = pc()
pc.start_vm(java_max_heap_size='100M')
from pycausal import search as s
tetrad = s.tetradrunner()
tetrad.run(algoId='fges',
dfs=df,
scoreId='bdeu',
dataType='discrete',
structurePrior=1.0,
samplePrior=1.0,
maxDegree=3,
faithfulnessAssumed=True,
verbose=True)
tetrad.getNodes()
tetrad.getEdges()
dot = tetrad.getDot()
dot.write_svg('fges-discrete.svg')
#!/usr/local/bin/python
import os
import pandas as pd
import pydot
from IPython.display import SVG
data_dir = os.path.join(os.getcwd(), 'data', 'audiology.txt')
df = pd.read_table(data_dir, sep="\t")
from pycausal.pycausal import pycausal as pc
pc = pc()
pc.start_vm(java_max_heap_size = '100M')
from pycausal import search as s
tetrad = s.tetradrunner()
tetrad.run(algoId = 'fges', dfs = df, scoreId = 'bdeu', dataType = 'discrete',
structurePrior = 1.0, samplePrior = 1.0, maxDegree = 3, faithfulnessAssumed = True, verbose = True)
tetrad.getNodes()
tetrad.getEdges()
dot_str = pc.tetradGraphToDot(tetrad.getTetradGraph())
graphs = pydot.graph_from_dot_data(dot_str)
graphs[0].write_svg('fges-discrete.svg')
pc.stop_vm()
def fges_stem(file_path, sys_iter, SGA_l, A_D):
BIC_l = [float(0)]
SGA = pd.DataFrame(SGA_l)
SGA.columns = ['cause gene name']
A_D_i = A_D
for i in range(sys_iter):
print(i)
file_l = os.listdir(file_path + '/Output/run%i' % i)
while 'completeMatrixn.csv' not in file_l:
df_name = file_path + '/Output/run%i/completeMatrix.csv' % i
df = pd.read_csv(df_name, header=0, index_col=None)
from pycausal.pycausal import pycausal as pc
pc = pc()
pc.start_vm(java_max_heap_size='6400M')
from pycausal import prior as p
# get knowledge from knowledge file
#prior = p.knowledgeFromFile(file_path + '/Input/Knowledge')
# get knowledge from DEG and SGA list
DEG_l = [x for x in df.columns if x not in SGA_l]
A_D_i = A_D_i[DEG_l]
forbid = create_knowledge(SGA, SGA_l, A_D_i)
temporal = [SGA_l, p.ForbiddenWithin(DEG_l)]
prior = p.knowledge(forbiddirect=forbid, addtemporal=temporal)
from pycausal import search as s
tetrad = s.tetradrunner()
tetrad.getAlgorithmParameters(algoId='fges', scoreId='bdeu-score')
tetrad.run(
algoId='fges',
dfs=df,
scoreId='bdeu-score',
priorKnowledge=prior,
dataType='discrete',
structurePrior=1.0,
samplePrior=1.0,
maxDegree=100,
faithfulnessAssumed=True,
verbose=True,
symmetricFirstStep=True
) # , numberResampling=10, resamplingEnsemble=1, addOriginalDataset=True)
# save edges.csv
node_l = tetrad.getNodes()
edge_l = tetrad.getEdges()
#edge_split_l = []
#for edge in edge_l:
#if '---' in edge:
#edge_n = edge.split(' ')
#if np.sum(df[edge.split(' ')[0]]) > np.sum(df[edge.split(' ')[2]]):
# edge_n.reverse()
#else:
# edge_n = edge_n
#edge_split_l.append(edge_n)
#else:
#edge_split_l.append(edge.split(' '))
#edge_split_l = [edge.split(' ') for edge in edge_l if '---' not in edge]
edge_split_l = [edge.split(' ') for edge in edge_l]
edge_df = pd.DataFrame(edge_split_l).iloc[:, [0, 2]]
edge_df.to_csv(file_path + '/Output/run%i/Edge.csv' % i,
index=False,
header=False)
# save completeMatrixn.csv
new_df = df.loc[:, node_l]
new_df.to_csv(file_path + '/Output/run%i/completeMatrixn.csv' % i,
index=False,
header=True)
# save BIC.txt
print(tetrad.getTetradGraph(),
file=open(file_path + '/Output/run%i/BIC.txt' % i, 'a'))
file_l = os.listdir(file_path + '/Output/run%i' % i)
else:
# save BIC which used to verify convergency
with open(file_path + '/Output/run%i/BIC.txt' % i, 'r') as BIC_txt:
for line in BIC_txt:
if 'BIC: -' in line:
BIC_l.append(float(line[5:-1]))
j = i + 1
mk_dir(file_path + '/Output/run%d' % j)
next_file_l = os.listdir(file_path + '/Output/run%i' % j)
while 'completeMatrix.csv' not in next_file_l:
exe_path = './MCMC/inferSGAInNetwork_TDI.exe'
m_path = ' -m ' + file_path + '/Output/run%i/completeMatrixn.csv' % i
i_path = ' -i ' + file_path + '/Input/S_A0.csv'
e_path = ' -e ' + file_path + '/Output/run%i/Edge.csv' % i
o_path = ' -o ' + file_path + '/Output/run%d/ -x 50' % j
combine = exe_path + m_path + i_path + e_path + o_path
os.system(combine)
time.sleep(20)
next_file_l = os.listdir(file_path + '/Output/run%i' % j)
else:
pd.DataFrame(BIC_l).to_csv(file_path + '/Output/BIC.csv',
index=False,
header=False)