def test_HDDMTruncated_distributions(self):
params = hddm.generate.gen_rand_params()
data, params_subj = hddm.generate.gen_rand_data(subjs=4, params=params)
m = hddm.HDDMTruncated(data)
m.sample(self.iter, burn=self.burn)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['v'], pm.Normal)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['v'].parents['mu'], pm.Normal)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['v'].parents['tau'], pm.Deterministic)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['v'].parents['tau'].parents['x'], pm.Uniform)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['a'], pm.TruncatedNormal)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['a'].parents['mu'], pm.Uniform)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['a'].parents['tau'], pm.Deterministic)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['a'].parents['tau'].parents['x'], pm.Uniform)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['t'], pm.TruncatedNormal)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['t'].parents['tau'], pm.Deterministic)
assert isinstance(m.nodes_db.ix['wfpt.0']['node'].parents['t'].parents['tau'].parents['x'], pm.Uniform)
def test_HDDMTruncated_distributions(self):
params = hddm.generate.gen_rand_params()
data, params_subj = hddm.generate.gen_rand_data(subjs=4,
params=params,
size=10)
m = hddm.HDDMTruncated(data)
m.sample(self.iter, burn=self.burn)
self.assertIsInstance(m.nodes_db.loc["wfpt.0"]["node"].parents["v"],
pm.Normal)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]["node"].parents["v"].parents["mu"],
pm.Normal)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]["node"].parents["v"].parents["tau"],
pm.Deterministic,
)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]
["node"].parents["v"].parents["tau"].parents["x"],
pm.Uniform,
)
self.assertIsInstance(m.nodes_db.loc["wfpt.0"]["node"].parents["a"],
pm.TruncatedNormal)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]["node"].parents["a"].parents["mu"],
pm.Uniform)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]["node"].parents["a"].parents["tau"],
pm.Deterministic,
)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]
["node"].parents["a"].parents["tau"].parents["x"],
pm.Uniform,
)
self.assertIsInstance(m.nodes_db.loc["wfpt.0"]["node"].parents["t"],
pm.TruncatedNormal)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]["node"].parents["t"].parents["tau"],
pm.Deterministic,
)
self.assertIsInstance(
m.nodes_db.loc["wfpt.0"]
["node"].parents["t"].parents["tau"].parents["x"],
pm.Uniform,
)
def check_outlier_model(seed=None, p_outlier=0.05):
"""Estimate data which contains outliers"""
if seed is not None:
np.random.seed(seed)
#generate params and data
params_true = hddm.generate.gen_rand_params(include=())
data, temp = hddm.generate.gen_rand_data(size=500, params=params_true)
data = pd.DataFrame(data)
#generating outliers
n_outliers = int(len(data) * p_outlier)
outliers = data[:n_outliers].copy()
#fast outliers
outliers.rt[:n_outliers // 2] = np.random.rand(
n_outliers // 2) * (min(abs(data['rt'])) - 0.11) + 0.11
#slow outliers
outliers.rt[n_outliers //
2:] = np.random.rand(n_outliers - n_outliers // 2) * 2 + max(
abs(data['rt']))
outliers.response = np.random.randint(0, 2, n_outliers)
print("generating %d outliers. %f of the dataset" %
(n_outliers, float(n_outliers) / (n_outliers + len(data))))
print("%d outliers are fast" % sum(outliers.rt < min(data.rt)))
print("%d outliers are slow" % sum(outliers.rt > max(data.rt)))
#Estimating the data without outliers. this is the best estimation we could get
#from this data
hm = hddm.HDDMTruncated(data)
hm.map()
index = ['true', 'estimated']
best_estimate = hm.values
df = pd.DataFrame([params_true, hm.values], index=index,
dtype=np.float).dropna(1)
print("benchmark: MAP of clean data. This is as good as we can get")
print(df)
#combine data with outliers
data = pd.concat((data, outliers), ignore_index=True)
#estimate the data with outlier, to confirm that it is worse
hm = hddm.HDDMTruncated(data)
hm.map()
index = ['best_estimate', 'this_estimate']
df = pd.DataFrame([best_estimate, hm.values], index=index,
dtype=np.float).dropna(1)
print("MAP with outliers: This is as bas as we can get")
print(df)
#MAP with p_outlier as random variable
hm = hddm.HDDMTruncated(data, include='p_outlier')
hm.map()
df = pd.DataFrame([best_estimate, hm.values], index=index, dtype=np.float)
df.ix['best_estimate']['p_outlier'] = 0
print("MAP with random p_outlier (Estimated from the data)")
print(df.dropna(1))
#MAP with fixed p_outlier
fixed_p_outlier = 0.1
hm = hddm.HDDMTruncated(data, p_outlier=fixed_p_outlier)
hm.map()
df = pd.DataFrame([best_estimate, hm.values], index=index, dtype=np.float)
print("MAP with fixed p_outlier (%.3f) " % fixed_p_outlier)
print(df.dropna(1))
#Chi-square
hm = hddm.HDDMTruncated(data)
hm.optimize(method='chisquare')
df = pd.DataFrame([best_estimate, hm.values], index=index,
dtype=np.float).dropna(1)
print("Chisquare method")
print(df)
return data
def init_model(self, data):
self.model = hddm.HDDMTruncated(data,
group_only_nodes=['sz', 'st', 'sv'],
**self.init_kwargs)
def init_model(self, data):
self.model = hddm.HDDMTruncated(data, **self.init_kwargs)