def fit(self, p, samples=100):
"""Estimate the g-formula for network data under weak interference at coverage `p`
"""
marginals = []
for s in range(samples):
# Selecting and applying binary treatment
g = self.df.copy()
g[self.exposure] = np.random.binomial(n=1, p=p, size=g.shape[0])
# Back-calculate updated exposure mapping
v_vector = np.asarray(g[self.exposure])
g[self.exposure + '_sum'] = fast_exp_map(self.adj_matrix, v_vector, measure='sum')
g[self.exposure + '_mean'] = fast_exp_map(self.adj_matrix, v_vector, measure='mean')
g[self.exposure + '_mean'] = g[self.exposure + '_mean'].fillna(0) # isolates should have mean=0
g[self.exposure + '_var'] = fast_exp_map(self.adj_matrix, v_vector, measure='var')
g[self.exposure + '_var'] = g[self.exposure + '_var'].fillna(0) # isolates should have var=0
g[self.exposure + '_mean_dist'] = fast_exp_map(self.adj_matrix, v_vector, measure='mean_dist')
g[self.exposure + '_mean_dist'] = g[self.exposure + '_mean_dist'].fillna(0) # isolates have mean_dist=0
g[self.exposure + '_var_dist'] = fast_exp_map(self.adj_matrix, v_vector, measure='var_dist')
g[self.exposure + '_var_dist'] = g[self.exposure + '_var_dist'].fillna(0) # isolates have var_dist=0
if self._thresholds_any_:
create_threshold(data=g, variables=self._thresholds_variables_,
thresholds=self._thresholds_, definitions=self._thresholds_def_)
# Generating predictions for treatment plan
g[self.outcome] = np.nan
g[self.outcome] = self._outcome_model.predict(g)
marginals.append(np.mean(g[self.outcome]))
self.marginals_vector = marginals
self.marginal_outcome = np.mean(marginals)
def __init__(self, network, exposure, outcome, verbose=False):
"""Implementation of the g-formula estimator described in Sofrygin & van der Laan 2017
"""
# Background processing to convert network attribute data to pandas DataFrame
df = network_to_df(network)
if not df[exposure].value_counts().index.isin([0, 1]).all():
raise ValueError("NetworkGFormula only supports binary exposures currently")
if df[outcome].value_counts().index.isin([0, 1]).all():
self._continuous_ = False
else:
self._continuous_ = True
network = nx.convert_node_labels_to_integers(network, first_label=0, label_attribute='_original_id_')
self.network = network
self.adj_matrix = nx.adjacency_matrix(network, weight=None)
self.exposure = exposure
self.outcome = outcome
# Creating variable mapping for all variables in the network
for v in [var for var in list(df.columns) if var not in ['_original_id_', outcome]]:
v_vector = np.asarray(df[v])
df[v + '_sum'] = fast_exp_map(self.adj_matrix, v_vector, measure='sum')
df[v + '_mean'] = fast_exp_map(self.adj_matrix, v_vector, measure='mean')
df[v + '_mean'] = df[v + '_mean'].fillna(0) # isolates should have mean=0
df[v + '_var'] = fast_exp_map(self.adj_matrix, v_vector, measure='var')
df[v + '_var'] = df[v + '_var'].fillna(0) # isolates should have var=0
df[v + '_mean_dist'] = fast_exp_map(self.adj_matrix, v_vector, measure='mean_dist')
df[v + '_mean_dist'] = df[v + '_mean_dist'].fillna(0) # isolates should have mean_dist=0
df[v + '_var_dist'] = fast_exp_map(self.adj_matrix, v_vector, measure='var_dist')
df[v + '_var_dist'] = df[v + '_var_dist'].fillna(0) # isolates should have var_dist=0
# Calculating Degree
degree_data = pd.DataFrame.from_dict(dict(network.degree), orient='index').rename(columns={0: 'degree'})
self.df = pd.merge(df, degree_data, how='left', left_index=True, right_index=True)
# Output attributes
self.marginals_vector = None
self.marginal_outcome = None
# Storage for items I need later
self._outcome_model = None
self._q_model = None
self._verbose_ = verbose
self._thresholds_ = []
self._thresholds_variables_ = []
self._thresholds_def_ = []
self._thresholds_any_ = False
def test_fast_exp_map_graph3(self):
G = nx.complete_graph(5)
a = [1, 1, 1, 0, 0]
for node in G.nodes():
G.nodes[node]['A'] = a[node]
npt.assert_equal(
fast_exp_map(nx.adjacency_matrix(G, weight=None),
np.array(a),
measure='sum'), exp_map(G, 'A'))
def test_fast_exp_map_directed(self):
G = nx.DiGraph()
G.add_edges_from([(0, 1), (0, 2), (0, 3), (0, 4)])
a = [1, 0, 1, 1, 1]
for node in G.nodes():
G.nodes[node]['A'] = a[node]
npt.assert_equal(
fast_exp_map(nx.adjacency_matrix(G, weight=None),
np.array(a),
measure='sum'), exp_map(G, 'A'))
def _generate_pooled_sample(self, p, samples):
pooled_sample = []
for s in range(samples):
g = self.df.copy()
g[self.exposure] = np.random.binomial(n=1, p=p, size=g.shape[0])
g[self.exposure+'_sum'] = fast_exp_map(self.adj_matrix, np.array(g[self.exposure]), measure='sum')
g[self.exposure + '_mean'] = fast_exp_map(self.adj_matrix, np.array(g[self.exposure]), measure='mean')
g[self.exposure + '_mean'] = g[self.exposure + '_mean'].fillna(0) # isolates should have mean=0
g[self.exposure + '_var'] = fast_exp_map(self.adj_matrix, np.array(g[self.exposure]), measure='var')
g[self.exposure + '_var'] = g[self.exposure + '_var'].fillna(0) # isolates should have mean=0
g[self.exposure + '_mean_dist'] = fast_exp_map(self.adj_matrix,
np.array(g[self.exposure]), measure='mean_dist')
g[self.exposure + '_mean_dist'] = g[self.exposure + '_mean_dist'].fillna(0) # isolates should have mean=0
g[self.exposure + '_var_dist'] = fast_exp_map(self.adj_matrix,
np.array(g[self.exposure]), measure='var_dist')
g[self.exposure + '_mean_dist'] = g[self.exposure + '_mean_dist'].fillna(0) # isolates should have mean=0
if self._gs_measure_ is None:
network = self.network.copy()
a = np.array(g[self.exposure])
for n in network.nodes():
network.node[n][self.exposure] = a[n]
df = exp_map_individual(network, measure=self.exposure, max_degree=self._max_degree_).fillna(0)
for c in self._nonparam_cols_:
g[c] = df[c]
if self._thresholds_any_:
create_threshold(data=g, variables=self._thresholds_variables_,
thresholds=self._thresholds_, definitions=self._thresholds_def_)
g['_sample_id_'] = s
pooled_sample.append(g)
return pd.concat(pooled_sample, axis=0, ignore_index=True)
raise ValueError("Invalid set-up specification for " + network +
" network")
else:
raise ValueError("Invalid network name in .sh script")
# Determining if shift or absolute
shift = bool(int(shift))
if shift:
prop_treated = [-2.5, -2.0, -1.5, -1.0, -0.5, 0.5, 1.0, 1.5, 2.0, 2.5]
# Generating probabilities (true) to assign
data = network_to_df(G)
adj_matrix = nx.adjacency_matrix(G, weight=None)
data['O_mean'] = fast_exp_map(adj_matrix,
np.array(data['O']),
measure='mean')
data['G_mean'] = fast_exp_map(adj_matrix,
np.array(data['G']),
measure='mean')
prob = logistic.cdf(-1.3 - 1.5 * data['P'] + 1.5 * data['P'] * data['G'] +
0.95 * data['O_mean'] + 0.95 * data['G_mean'])
log_odds = np.log(probability_to_odds(prob))
else:
prop_treated = [
0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65,
0.7, 0.75, 0.8, 0.85, 0.9, 0.95
]
truth = truth_values(network=network,
raise ValueError("Invalid set-up specification for " + network +
" network")
else:
raise ValueError("Invalid network name in .sh script")
# Determining if shift or absolute
shift = bool(int(shift))
if shift:
prop_treated = [-2.5, -2.0, -1.5, -1.0, -0.5, 0.5, 1.0, 1.5, 2.0, 2.5]
# Generating probabilities (true) to assign
data = network_to_df(G)
adj_matrix = nx.adjacency_matrix(G, weight=None)
data['E_mean'] = fast_exp_map(adj_matrix,
np.array(data['E']),
measure='mean')
prob = logistic.cdf(-0.5 + 0.05 * (data['B'] - 30) +
0.25 * data['G'] * data['E'] + 0.05 * data['E_mean'])
log_odds = np.log(probability_to_odds(prob))
else:
prop_treated = [
0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65,
0.7, 0.75, 0.8, 0.85, 0.9, 0.95
]
truth = truth_values(network=network,
dgm=exposure,
restricted_degree=restrict,
shift=shift)