uhlerlab / causaldag Goto Github PK

Hi there,
I follow the DCI tutorial and got a different result as shown in the figure:
Should I do anything else to obtain the result in the tutorial?

Problem 1: Here is the error message that I got when I try to import the graphical_model_learning package and run its gsp function.

Traceback (most recent call last):
File "C:...\new.py", line 1, in
import graphical_model_learning
File "C:...\venv\lib\site-packages\graphical_model_learning_init_.py", line 3, in
from .samplers import *
ModuleNotFoundError: No module named 'graphical_model_learning.samplers'

How should I fix that?

Problem 2: I would like to use the non-greedy version of the sparest permutation algorithm. It was available in the previous version of causaldag (i.e., 0.1a160). But is it gone for good? Or would it be re-implemented in a later version? Thanks!

Hello,
I am trying to run the tutorial for DCI for gene expression data: https://github.com/uhlerlab/causaldag/blob/master/docs/dci_tutorial.markdown

However, when I run d_dag = dci(X1, X2, alpha_ug = 0.001, alpha_skeleton=0.5, alpha_orient=0.001, max_set_size=3, verbose=1) as specified by the tutorial, I get NameError: name 'difference_ug' is not defined. Do you know what may be causing this error? I am using the same input files as the tutorial and installed DCI into my jupyter notebook using !pip install --user causaldag.

Thanks!

Best,
Elaine

In the code provided in the readme for Simple Example, the line
iv = dag.optimal_intervention(cpdag=cpdag) seems to cause an error due to missing function optimal_intervention().

Upon further investigation, I tried replacing this function with with dag.greedy_optimal_single_node_intervention() or dag.greedy_optimal_fully_orienting_interventions(), another error arises:

Traceback (most recent call last):
icpdag = dag.interventional_cpdag([iv], cpdag=cpdag)
"causaldag/causaldag/classes/dag.py", line 1660, in interventional_cpdag
cut_edges.update({(i, j) for i, j in self._arcs if len({i, j} & set(iv_nodes)) == 1})
"causaldag/causaldag/classes/dag.py", line 1660, in
cut_edges.update({(i, j) for i, j in self._arcs if len({i, j} & set(iv_nodes)) == 1})
TypeError: unhashable type: 'list'

I tried the sample code from the causaldag documentation. It didn't work because both partial_correlation_test and partial_correlation_suffstat are not in causaldag.utils.ci_tests. So I ran the following instead.

import causaldag as cd
from causaldag.utils.ci_tests import MemoizedCI_Tester, gauss_ci
from causaldag import gauss_ci_suffstat
import random
import numpy as np
random.seed(1212)
np.random.seed(12131)
nnodes = 7
d = cd.rand.directed_erdos(nnodes, exp_nbrs=4)
g = cd.rand.rand_weights(d)
samples = g.sample(1000)
suffstat = gauss_ci_suffstat(samples)
ci_tester = MemoizedCI_Tester(gauss_ci, suffstat, alpha=1e-3)
est_dag = cd.sparsest_permutation(set(range(nnodes)), ci_tester)
print(est_dag)

Here is the error message:
Traceback (most recent call last):
File "C:...\Test.py", line 14, in
est_dag = cd.sparsest_permutation(set(range(nnodes)), ci_tester, progress=True)
File "C:...\Python\Python37\lib\site-packages\causaldag\structure_learning\dag\gsp.py", line 152, in sparsest_permutation
dag = perm2dag(list(perm), ci_tester)
File "C:...\Python\Python37\lib\site-packages\causaldag\structure_learning\dag\gsp.py", line 107, in perm2dag
is_ci = ci_tester.is_ci(pi_i, pi_j, mb)
File "C:...\Python\Python37\lib\site-packages\causaldag\utils\ci_tests\ci_tester.py", line 60, in is_ci
test_results = self.ci_test(self.suffstat, i, j, cond_set=cond_set, **self.kwargs)
TypeError: 'module' object is not callable

I see the same kind of error when I call cd.gsp instead. Please advice. Thank!

Thanks for this very promising package. I am trying to apply this tool on single cell RNA seq data, of once cell type that has been sorted out. I have treatment vs control samples. However on following this tutorial, I encounter the following error. I do not have any NAs in the data set. Please how could I fix this?

!pip install --user causaldag
from causaldag import dci, dci_stability_selection
from causaldag import get_directed_and_undirected_edges
import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import networkx as nx
from sklearn.preprocessing import StandardScaler

# Matplotlib command for Jupyter notebooks only
%matplotlib inline


path_in = '/home/power/docker_scratch'
gene_exp_diff_1 = pd.read_csv(os.path.join(path_in, 'log_norm_plus1_counts_control.csv'), index_col=0)
gene_exp_diff_2 = pd.read_csv(os.path.join(path_in, 'log_norm_plus1_counts_treated.csv'), index_col=0)
gene_names = gene_exp_diff_1.columns.values

scaler = StandardScaler(with_std=False)
X1 = scaler.fit_transform(gene_exp_diff_1.values)
X2 = scaler.fit_transform(gene_exp_diff_2.values)

d_dag = dci(X1, X2, alpha_ug = 0.001, alpha_skeleton=0.5, alpha_orient=0.001, max_set_size=3, verbose=1)

ValueError: On entry to DLASCL parameter number 4 had an illegal value

Hi!

It appears like the function DAG.vstructures() is wrongly documented in

https://causaldag.readthedocs.io/en/latest/classes/generated/causaldag.classes.dag.DAG.vstructs.html#causaldag.classes.dag.DAG.vstructs

In the documentation it appears as DAG.vstructs(), which returns an error when called.

Besides that, thanks for an awesome and useful package!

Hello,
Is code available for the DCI module?

Thanks, Sandra

import numpy as np
from networkx import barabasi_albert_graph
from causaldag import DAG

def directed_random_graph(nnodes, random_graph_model):
edges = random_graph_model(nnodes).edges
random_permutation = np.arange(nnodes)
np.random.shuffle(random_permutation)
arcs = []
for edge in edges:
node1, node2 = edge
node1_position = np.where(random_permutation == node1)[0][0]
node2_position = np.where(random_permutation == node2)[0][0]
if node1_position < node2_position:
source = node1
endpoint = node2
else:
source = node2
endpoint = node1
arcs.append((source, endpoint))
return DAG(nodes=set(range(nnodes)), arcs=arcs)

def directed_barbasi(nnodes, nattach):
random_graph_model = lambda nnodes: barabasi_albert_graph(nnodes, nattach)
return directed_random_graph(nnodes, random_graph_model)

Hello,

I would need the causaldag module to run the DeCAMFounder implementation from https://github.com/uhlerlab/decamfound.

However it seems that the causaldag module implementation has changed a lot since the development of decamfound (in particular the source code has been divided into different modules).
So I cannot use the latest causaldag module installed with the command pip3 install causaldag.

Would you know which version of the causaldag module was used to develop the decamfound source code?

Thank you very much for your help!