def SelectLabeled(self, labeled_data_x, labeled_data_y, unlabeled_data_x):
# just append train data to labeled data
labeled_x = np.concatenate((self.init_labeled_data_x, labeled_data_x)) \
if len(labeled_data_x) > 0 else self.init_labeled_data_x
labeled_y = np.concatenate((self.init_labeled_data_y, labeled_data_y)) \
if len(labeled_data_x) > 0 else self.init_labeled_data_y
#
# create model to predict with confidence and credibility
model = ClassifierAdapter(
DecisionTreeClassifier(random_state=config.random_state,
min_samples_leaf=config.min_samples_leaf))
nc = ClassifierNc(model, MarginErrFunc())
model_tcp = TcpClassifier(nc, smoothing=True)
model_tcp.fit(labeled_x, labeled_y)
s = model_tcp.predict_conf(unlabeled_data_x)
#
# selection method
labeled_ind = [
i for i, a in enumerate(s)
if a[1] > config.confidence and a[2] > config.credibility
]
unlabeled_ind = [
i for i, a in enumerate(s)
if a[1] < config.confidence or a[2] < config.credibility
]
labeled_unlabeled_x, labeled_unlabeled_y, unlabeled_data_x = \
np.take(unlabeled_data_x, labeled_ind, axis=0), \
np.take(s.T, labeled_ind), np.take(unlabeled_data_x, unlabeled_ind, axis=0)
return labeled_unlabeled_x, labeled_unlabeled_y, unlabeled_data_x
def test_tcp_classification_svm(self):
# -----------------------------------------------------------------------------
# Setup training, calibration and test indices
# -----------------------------------------------------------------------------
data = load_iris()
idx = np.random.permutation(data.target.size)
train = idx[:int(idx.size / 2)]
test = idx[int(idx.size / 2):]
# -----------------------------------------------------------------------------
# Train and calibrate
# -----------------------------------------------------------------------------
tcp = TcpClassifier(
ClassifierNc(ClassifierAdapter(SVC(probability=True)),
MarginErrFunc()))
tcp.fit(data.data[train, :], data.target[train])
# -----------------------------------------------------------------------------
# Predict
# -----------------------------------------------------------------------------
prediction = tcp.predict(data.data[test, :], significance=0.1)
header = np.array(["c0", "c1", "c2", "Truth"])
table = np.vstack([prediction.T, data.target[test]]).T
df = pd.DataFrame(np.vstack([header, table]))
print(df)
def fit(self, X, Y, lengths=None, init_prob=None, tran_prob=None):
"""Fits the model on observables X and respective hidden sequences Y.
Parameters
----------
X : numpy array (n_samples, n_features)
Individual observations.
Y : numpy array (n_samples,)
Individual observations of hidden states.
lengths : list of integer
Lengths of sequences in X and Y. If None, X and Y are assumed to
be a single sequence. The sum of lengths should be n_samples.
init_prob : dict (Default: None)
The item corresponding to the i-th key is the
probability of the hidden process to start in the
i-th state.
If default (=None), it is estimated from data.
tran_prob : dict (Default: None)
The item corresponding to the i-th key of the dictionary
is a dictionary itself, which, for the j-th key,
indicates the probability of transitioning from the
i-th to the j-th state.
If default (=None), it is estimated from data.
"""
self.train_x = X
self.train_y = Y
if lengths is None:
lengths = [len(Y)]
# CP model
self.cp = TcpClassifier(self.ncm, smoothing=self.smooth)
# Initial and transition probabilities.
if not init_prob:
init_prob = self._estimate_initial_prob(Y, lengths)
if not tran_prob:
tran_prob = self._estimate_transition_prob(Y, lengths)
self.init_prob = init_prob
self.tran_prob = tran_prob
from sklearn.datasets import load_iris
from nonconformist.base import ClassifierAdapter
from nonconformist.cp import TcpClassifier
from nonconformist.nc import ClassifierNc, MarginErrFunc
# -----------------------------------------------------------------------------
# Setup training, calibration and test indices
# -----------------------------------------------------------------------------
data = load_iris()
idx = np.random.permutation(data.target.size)
train = idx[:int(idx.size / 2)]
test = idx[int(idx.size / 2):]
# -----------------------------------------------------------------------------
# Train and calibrate
# -----------------------------------------------------------------------------
tcp = TcpClassifier(ClassifierNc(ClassifierAdapter(SVC(probability=True)),
MarginErrFunc()))
tcp.fit(data.data[train, :], data.target[train])
# -----------------------------------------------------------------------------
# Predict
# -----------------------------------------------------------------------------
prediction = tcp.predict(data.data[test, :], significance=0.1)
header = np.array(['c0','c1','c2','Truth'])
table = np.vstack([prediction.T, data.target[test]]).T
df = pd.DataFrame(np.vstack([header, table]))
print(df)
import sys
sys.path.append('/Users/staffan/git/peptid_studie/experiments/src') # Nonconformist
from nonconformist.cp import TcpClassifier
from nonconformist.nc import NcFactory
iris = load_iris()
idx = np.random.permutation(iris.target.size)
# Divide the data into training set and test set
idx_train, idx_test = idx[:100], idx[100:]
model = SVC(probability=True) # Create the underlying model
nc = NcFactory.create_nc(model) # Create a default nonconformity function
tcp = TcpClassifier(nc) # Create a transductive conformal classifier
# Fit the TCP using the proper training set
tcp.fit(iris.data[idx_train, :], iris.target[idx_train])
# Produce predictions for the test set
predictions = tcp.predict(iris.data[idx_test, :])
#
targets = np.array(iris.target[idx_test], copy=True)
targets.shape = (len(targets),1)
output = np.hstack((targets, predictions))
np.savetxt('resources/multiclass.csv', output, delimiter=',')
from nonconformist.evaluation import class_mean_errors
# -----------------------------------------------------------------------------
# Setup training, calibration and test indices
# -----------------------------------------------------------------------------
data = load_iris()
idx = np.random.permutation(data.target.size)
train = idx[:int(idx.size / 2)]
test = idx[int(idx.size / 2):]
# -----------------------------------------------------------------------------
# Train and calibrate TCP
# -----------------------------------------------------------------------------
tcp = TcpClassifier(
ClassifierNc(ClassifierAdapter(SVC(probability=True, gamma='scale')),
MarginErrFunc()))
tcp.fit(data.data[train, :], data.target[train])
# -----------------------------------------------------------------------------
# Predict
# -----------------------------------------------------------------------------
prediction = tcp.predict(data.data[test, :], significance=0.1)
header = np.array(['c0', 'c1', 'c2', 'Truth'])
table = np.vstack([prediction.T, data.target[test]]).T
df = pd.DataFrame(np.vstack([header, table]))
print('TCP')
print('---')
print(df)
class CPHMM(BaseEstimator):
def __init__(self, ncm, n_states, smooth=True):
"""Initialise a CP-HMM model.
Parameters
----------
ncm : nonconformist.BaseScorer
Nonconformity measure to use.
n_states : int
Number of hidden states.
smooth : bool
If True, smooth CP is used, which achieves exact validity.
Otherwise, standard CP is used, guaranteeing error smaller or
equal to the significance level.
"""
self.ncm = ncm
self.n_states = n_states
self.smooth = smooth
def fit(self, X, Y, lengths=None, init_prob=None, tran_prob=None):
"""Fits the model on observables X and respective hidden sequences Y.
Parameters
----------
X : numpy array (n_samples, n_features)
Individual observations.
Y : numpy array (n_samples,)
Individual observations of hidden states.
lengths : list of integer
Lengths of sequences in X and Y. If None, X and Y are assumed to
be a single sequence. The sum of lengths should be n_samples.
init_prob : dict (Default: None)
The item corresponding to the i-th key is the
probability of the hidden process to start in the
i-th state.
If default (=None), it is estimated from data.
tran_prob : dict (Default: None)
The item corresponding to the i-th key of the dictionary
is a dictionary itself, which, for the j-th key,
indicates the probability of transitioning from the
i-th to the j-th state.
If default (=None), it is estimated from data.
"""
self.train_x = X
self.train_y = Y
if lengths is None:
lengths = [len(Y)]
# CP model
self.cp = TcpClassifier(self.ncm, smoothing=self.smooth)
# Initial and transition probabilities.
if not init_prob:
init_prob = self._estimate_initial_prob(Y, lengths)
if not tran_prob:
tran_prob = self._estimate_transition_prob(Y, lengths)
self.init_prob = init_prob
self.tran_prob = tran_prob
def predict(self, x, e):
"""Return a CP-HMM prediction region.
Uses CP-HMM to output a prediction region (i.e.: a set of candidate
hidden sequences) for the observed sequence x.
NOTE: If any of the elements of the sequence has an empty
prediction set, then no predictions are returned.
Parameters
----------
x : list
Observed sequence.
e : float in [0,1]
Significance level.
"""
# Reduce significance level as required.
e /= float(len(x))
# Find candidates.
candidates = self._hidden_candidates(x, e)
# Generate paths.
paths = self._generate_paths(candidates, self.tran_prob,
self.init_prob)
return paths
def _generate_paths(self, candidates, trans_prob, init_prob):
"""Generate and score paths.
Accepts a list of list of candidate. Each list of
candidate contains potential true hidden states to
compose a path.
The function produces all possible paths, and scores them
w.r.t. the transition and initial probabilities.
It returns the paths in a list, sorted by scores:
from the most likely to the least likely.
Parameters
----------
candidates : list of list
The i-th list it contains represents a set of state
candidates for the i-th element of the sequence.
trans_prob : dictionary
The keys of this dictionary are tuples in the form
(i, j). The element (i, j) is associated with the
transition probability from state i to state j.
init_prob : dictionary
The keys are numbers i (as many as the states).
init_prob[i] contains the probability of a sequence
starting in state i.
"""
paths = list(itertools.product(*candidates))
scores = []
for p in paths:
p = list(map(int, p)) # So we can use them as indexes
s = init_prob[p[0]]
for i in range(len(p) - 1):
s *= trans_prob[(p[i], p[i + 1])]
scores.append(s)
paths_sorted = [x[1] for x in sorted(zip(scores, paths), reverse=True)]
return paths_sorted
def _hidden_candidates(self, x, e):
"""Uses CP-HMM to predict, for each element of the observed sequence, a
list of candidate states. Thanks to CP's validity guarantee, the true
hidden states is within the list of candidates with probability 1-e.
Parameters
----------
x : list
Observed sequence.
e : float in [0,1]
Significance level.
"""
# Flatten sequences, "train" CP
X = self.train_x.flatten().reshape(-1, 1)
Y = self.train_y.flatten()
self.cp.fit(X, Y)
# For each element of the observed sequence x
# determine a set of candidate states.
y_candidates = []
for i in range(len(x)):
candidates_bool = self.cp.predict(x[i].reshape(-1, 1), e)[0]
candidates = self.cp.classes[candidates_bool]
y_candidates.append(candidates)
return y_candidates
def _estimate_initial_prob(self, Y, lengths):
"""Returns an frequentist estimate of the initial probabilities over
the observed hidden states. Assumes that the hidden states are
specified by sequential numbers 0, 1, ... numbers).
Parameters
----------
Y : numpy array (n_samples,)
Individual observations of hidden states.
lengths : list of integer
Lengths of sequences in X and Y. If None, X and Y are assumed to
be a single sequence. The sum of lengths should be n_samples.
"""
if not len(Y):
return []
ip = np.array([.0] * self.n_states)
for i, j in iter_from_X_lengths(Y, lengths):
ip[Y[i]] += 1
ip /= sum(ip)
# To dictionary
ip = dict(list(zip(list(range(len(ip))), ip)))
return ip
def _estimate_transition_prob(self, Y, lengths):
"""Returns an frequentist estimate of the transition probabilities over
the observed hidden states. Assumes that the hidden states are
specified by sequential numbers 0, 1, ... .
Parameters
----------
Y : numpy array (n_samples,)
Individual observations of hidden states.
lengths : list of integer
Lengths of sequences in X and Y. If None, X and Y are assumed to
be a single sequence. The sum of lengths should be n_samples.
"""
if not len(Y):
return np.empty()
tp = np.zeros((self.n_states, self.n_states))
for i, j in iter_from_X_lengths(Y, lengths):
for k in range(i, j - 1):
tp[Y[k], Y[k + 1]] += 1
# Missing values, normalise
# NOTE: here is made the assumption that states are integers
# 0, 1, ..., self.n_states-1.
for y in range(self.n_states):
if sum(tp[y, :]) == 0:
tp[y, :] = 1.0
tp[y, :] /= sum(tp[y, :])
# To dictionary
tran_prob = {}
for i in range(len(tp)):
for j in range(len(tp[0])):
tran_prob[(i, j)] = tp[i][j]
return tran_prob