Example about causaldiscoverytoolbox HOT 7 CLOSED

Hi,
It seems like your pytorch version is not 0.4. Could you update it and try again?
Best,
Diviyan

from causaldiscoverytoolbox.

Awesome, that sorted the first one out! On the next tab, I have this issue.

Thanks for taking the time.

# Orient the edges of the graph
from cdt.causality.graph import CGNN
Cgnn = CGNN()
start_time = time.time()
dgraph = Cgnn.predict(data, graph=ugraph, nb_runs=12, nb_max_runs=20, train_epochs=1500, test_epochs=1000)
print("--- Execution time : %4.4s seconds ---" % (time.time() - start_time))

# Plot the output graph
nx.draw_networkx(dgraph, font_size=8) # The plot function allows for quick visualization of the graph.
plt.show()
# Print output results : 
pd.DataFrame(list(dgraph.edges(data='weight')), columns=['Cause', 'Effect', 'Score'])

The pairwise GNN model is computed on each edge of the UMG to initialize the model and start CGNN with a DAG

TypeError Traceback (most recent call last)
in ()
3 Cgnn = CGNN()
4 start_time = time.time()
----> 5 dgraph = Cgnn.predict(data, graph=ugraph, nb_runs=12, nb_max_runs=20, train_epochs=1500, test_epochs=1000)
6 print("--- Execution time : %4.4s seconds ---" % (time.time() - start_time))
7

~/anaconda/envs/py36/lib/python3.6/site-packages/cdt/causality/graph/model.py in predict(self, df_data, graph, **kwargs)
31 return self.orient_directed_graph(df_data, graph, **kwargs)
32 elif type(graph) == nx.Graph:
---> 33 return self.orient_undirected_graph(df_data, graph, **kwargs)
34 else:
35 print('Unknown Graph type')

~/anaconda/envs/py36/lib/python3.6/site-packages/cdt/causality/graph/CGNN.py in orient_undirected_graph(self, data, umg, nh, nb_runs, nb_jobs, gpu, lr, train_epochs, test_epochs, verbose, nb_max_runs)
257 og = gnn.orient_graph(data, umg, nb_runs=nb_runs, nb_max_runs=nb_max_runs,
258 nb_jobs=nb_jobs, train_epochs=train_epochs,
--> 259 test_epochs=test_epochs, verbose=verbose, gpu=gpu) # Pairwise method
260 # print(nx.adj_matrix(og).todense().shape)
261

~/anaconda/envs/py36/lib/python3.6/site-packages/cdt/causality/pairwise/model.py in orient_graph(self, df_data, graph, printout, nb_runs, **kwargs)
91
92 elif type(graph) == nx.Graph:
---> 93 edges = list(graph.edges)
94 output = nx.DiGraph()
95

TypeError: 'method' object is not iterable

from causaldiscoverytoolbox.

Oh yes, I had this bug corrected, could you update your toolbox to master?
Best,
Diviyan

from causaldiscoverytoolbox.

Sorry I got one problem after the update. - Running the same code.

The pairwise GNN model is computed on each edge of the UMG to initialize the model and start CGNN with a DAG

AttributeError Traceback (most recent call last)
in ()
3 Cgnn = CGNN()
4 start_time = time.time()
----> 5 dgraph = Cgnn.predict(data, graph=ugraph, nb_runs=5, nb_max_runs=6, train_epochs=15, test_epochs=8)
6 print("--- Execution time : %4.4s seconds ---" % (time.time() - start_time))
7

/Volumes/extra/FirmAI/Causal Inference/CausalDiscoveryToolbox-master/examples/cdt/causality/graph/model.py in predict(self, df_data, graph, **kwargs)
31 return self.orient_directed_graph(df_data, graph, **kwargs)
32 elif type(graph) == nx.Graph:
---> 33 return self.orient_undirected_graph(df_data, graph, **kwargs)
34 else:
35 print('Unknown Graph type')

/Volumes/extra/FirmAI/Causal Inference/CausalDiscoveryToolbox-master/examples/cdt/causality/graph/CGNN.py in orient_undirected_graph(self, data, umg, nh, nb_runs, nb_jobs, gpu, lr, train_epochs, test_epochs, verbose, nb_max_runs)
269 # print(nx.adj_matrix(og).todense().shape)
270 # print(list(og.edges()))
--> 271 dag = dagify_min_edge(og)
272 # print(nx.adj_matrix(dag).todense().shape)
273

/Volumes/extra/FirmAI/Causal Inference/CausalDiscoveryToolbox-master/examples/cdt/utils/graph_utils.py in dagify_min_edge(g)
14 """
15 while not nx.is_directed_acyclic_graph(g):
---> 16 cycle = nx.simple_cycles(g).next()
17 scores = []
18 edges = []

AttributeError: 'generator' object has no attribute 'next'

from causaldiscoverytoolbox.

It might be a python 3 thing, I will jsut change it locally. Thank, one last question - more personal - would you recommend any packages for automated causal effects from observational data after I have done the casual discovery? I have had a look at the following, https://github.com/laurencium/Causalinference, and https://github.com/akelleh/causality. If you are unsure then please just ignore the question, thanks so much for your help :)

from causaldiscoverytoolbox.

Hi again, glad that some errors are sorted out. Please keep me updated about the generator error.
Concerning causal inference, I've got no great experience in python packages. You could look into the "IDA" algorithm (implemented in the pcalg R package).

We plan to bring tools for causal effect evaluation in this toolbox, but it will be at a later date =)
Best,
Diviyan

from causaldiscoverytoolbox.

I'll be closing this issue, don't hesitate to open it if a bug pops again.
Best,
Diviyan

from causaldiscoverytoolbox.