# Setting for CGNN-Fourier
cdt.SETTINGS.use_Fast_MMD = True
cdt.SETTINGS.NB_RUNS = 64
# Setting for CGNN-MMD
# cdt.SETTINGS.use_Fast_MMD = False
#cdt.SETTINGS.NB_RUNS = 32
datafile = "Example_graph_confounders_numdata.csv"
skeletonfile = "Example_graph_confounders_skeleton.csv"
data = pd.read_csv(datafile)
skeleton_links = pd.read_csv(skeletonfile)
skeleton = cdt.UndirectedGraph(skeleton_links)
data = pd.DataFrame(scale(data), columns=data.columns)
GNN = cdt.causality.pairwise.GNN(backend="TensorFlow")
p_directed_graph = GNN.orient_graph_confounders(data,
skeleton,
printout=datafile +
'_printout.csv')
gnn_res = pd.DataFrame(p_directed_graph.list_edges(descending=True),
columns=['Cause', 'Effect', 'Score'])
gnn_res.to_csv(datafile + "_pairwise_predictions.csv")
CGNN_confounders = cdt.causality.graph.CGNN_confounders(backend="TensorFlow")
directed_graph = CGNN_confounders.orient_directed_graph(data, p_directed_graph)
cgnn_res = pd.DataFrame(directed_graph.list_edges(descending=True),
cdt.CGNN_SETTINGS.NB_RUNS = 16
# cdt.CGNN_SETTINGS.train_epochs = 2000
# cdt.CGNN_SETTINGS.test_epochs = 500
datafile = sys.argv[1]
skeletonfile = sys.argv[2]
cdt.CGNN_SETTINGS.asymmetry_param = float(sys.argv[3])
cdt.CGNN_SETTINGS.h_layer_dim = int(sys.argv[4])
print("Processing " + datafile + "...")
undirected_links = pd.read_csv(skeletonfile, sep='\t')
#undirected_links = pd.read_csv(skeletonfile)
umg = cdt.UndirectedGraph(undirected_links)
df_data = pd.read_csv(datafile, sep='\t')
#df_data = pd.read_csv(datafile)
#df_type = pd.read_csv(type_file)
#df_block = pd.read_csv(save_block, sep = '\t')
#type_variables = {}
#for i in range(df_type.shape[0]):
# type_variables[df_type["Node"].loc[i]] = df_type["Type"].loc[i]
type_variables = {}
for node in df_data.columns:
type_variables[node] = "Numerical"
CGNN_decomposable = cdt.causality.graph.CGNN_decomposable(backend="TensorFlow")
cdt.SETTINGS.NB_MAX_RUNS = 32
cdt.SETTINGS.nb_run_feature_selection = 16
cdt.SETTINGS.regul_param = 0.004
cdt.SETTINGS.threshold_UMG = 0.13
datafile = sys.argv[1]
umg_file = datafile + "_umg.csv"
pairwise_file = datafile + "_pairwise_predictions.csv"
print("Processing " + datafile + "...")
df_data = pd.read_csv(datafile, sep='\t')
df_umg = pd.read_csv(umg_file, index_col=False)
df_pairwise = pd.read_csv(pairwise_file, index_col=False)
umg = cdt.UndirectedGraph(df_umg)
#print(umg)
p_directed_graph = cdt.DirectedGraph(df_pairwise, skeleton=umg)
#print(p_directed_graph)
#FSGNN = cdt.independence.graph.FSGNN()
#umg = FSGNN.create_skeleton_from_data(df_data)
#umg_res = pd.DataFrame(umg.list_edges(descending=True), columns=['Cause', 'Effect', 'Score'])
#umg_res.to_csv(datafile + "_umg.csv")
#GNN = cdt.causality.pairwise.GNN(backend="TensorFlow")
#p_directed_graph = GNN.orient_graph(df_data, umg, printout=datafile + '_printout.csv')
#gnn_res = pd.DataFrame(p_directed_graph.list_edges(descending=True), columns=['Cause', 'Effect', 'Score'])
#gnn_res.to_csv(datafile + "_pairwise_predictions.csv")
#CGNN = cdt.causality.graph.CGNN(backend="TensorFlow")