def load_segmentation_model(modeldata):
model = HiddenMarkovModel('model')
states = {}
for s in modeldata:
if len(s['emission']) == 1:
emission = NormalDistribution(*s['emission'][0][:2])
else:
weights = np.array([w for _, _, w in s['emission']])
dists = [NormalDistribution(mu, sigma)
for mu, sigma, _ in s['emission']]
emission = GeneralMixtureModel(dists, weights=weights)
state = State(emission, name=s['name'])
states[s['name']] = state
model.add_state(state)
if 'start_prob' in s:
model.add_transition(model.start, state, s['start_prob'])
for s in modeldata:
current = states[s['name']]
for nextstate, prob in s['transition']:
model.add_transition(current, states[nextstate], prob)
model.bake()
return model
def update_hmm(self):
num_states = self.num_states
start_prob = self.start_prob
num_emissions = self.num_emissions
hmm = HiddenMarkovModel('hmm')
dist = [
DiscreteDistribution(
dict(zip(range(num_emissions), self.emissions[i])))
for i in range(num_states)
]
states = [
State(dist[i], 's' + str(i).zfill(2)) for i in range(num_states)
]
hmm.add_states(states)
for i in range(num_states):
s_i = states[i]
hmm.add_transition(hmm.start, s_i, start_prob[i])
for j in range(num_states):
s_j = states[j]
p = self.transitions[i, j]
hmm.add_transition(s_i, s_j, p)
self.hmm = hmm
self.hmm.bake()
def _buildModel(self, data):
'''
builds the model given the data to init the distributions at good point
data: 2d matrix every row is a vector of features
'''
# we want to call from_matrix(transition, dists, starts, ends)
tm = np.zeros((self.statesNumber, self.statesNumber))
indices = [(x, x) for x in range(self.statesNumber)]
indices.extend([(x, x + 1) for x in range(self.statesNumber)])
indices.pop(
) # this the item (self.statesNumber-1 , self.statesNumber) that is out of bound
indices = np.array(indices)
tm[indices[:, 0], indices[:, 1]] = 0.5
tm[self.statesNumber - 1, self.statesNumber -
1] = 0.5 # this is the end state prob, i write it alone as we may change it specificity
dists = self._initDists(data)
starts = np.zeros((self.statesNumber, ))
starts[0] = 1
ends = np.zeros((self.statesNumber, ))
ends[-1] = 0.5
self.model = HiddenMarkovModel.from_matrix(tm,
dists,
starts,
ends,
name=self.mname)
return self.model
def init():
m = 1000 # restricts number of genes, used for local testing
gc, mt, track = load_data(m)
state_range = [5, 10, 25, 50, 100]
z_range = [3, 5, 10, 20]
msequences, mlabels = df_to_sequence_list(mt.data)
gsequences, glabels = df_to_sequence_list(gc.data)
sequences = np.concatenate((msequences, gsequences), 0)
labels = np.concatenate((mlabels, glabels))
sequences = np.concatenate((sequences, -1 * sequences))
# tie positive and negative expression sequences
tied = {}
for i, label in enumerate(labels):
tied[label] = [i, i+labels.size]
state_labels = np.concatenate(((labels + '+'), (labels + '-')))
labels = np.concatenate((labels, labels))
# noise model trained on all data once
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts)
noise.freeze_distributions()
return gc, mt, sequences, labels, state_labels, tied, noise, z_range, \
state_range
def init(m, seed):
if m == -1:
m = None
gc, mt, track = load_data(m, seed)
msequences, mlabels = df_to_sequence_list(mt.data)
gsequences, glabels = df_to_sequence_list(gc.data)
sequences = np.concatenate((msequences, gsequences), 0)
labels = np.concatenate((mlabels, glabels))
sequences = np.concatenate((sequences, -1 * sequences))
# tie positive and negative expression sequences
tied = {}
for i, label in enumerate(labels):
tied[label] = [i, i+labels.size]
labels = np.concatenate(((labels + '+'), (labels + '-')))
# noise model trained on all data once
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts,
name='noise')
noise.freeze_distributions()
return sequences, labels, tied, noise
def insert_delete_main_hmm(data_matrix):
v_columns = column_clasify(data_matrix)
v_zones = create_zones(v_columns)
v_grouped_states = group_states(v_zones, 'test')
v_model = HiddenMarkovModel()
v_first_state = State(None, name='ali_start')
v_last_state = State(None, name='ali_end')
v_model.add_state(v_first_state)
v_model.add_transition(v_model.start, v_first_state, 1)
v_model.add_state(v_last_state)
add_states(v_model, v_grouped_states)
v_trans = calculate_transitions(v_first_state, v_last_state,
v_grouped_states)
apply_transitions(v_model, v_trans)
v_model.bake()
return v_model
def HiddenSpaceGenerator(cls, X, n_components):
"""This method creates more features by training a HiddenMarkovModel
on the game statistics, then returns the hidden state space of each
timestep/game as a new feature. HOWEVER, note that this doesn't
return a list of features, instead it returns a HMM that can generate
more features but can still be used for classification also.
Arguments:
X: the input features (with a series_idx)
n_components: The number of hidden state space variables to
initialize. Note that (sadly) pomegranate does not implement
continuous HMMs, so it will discretize every continuous
variable by K-means so then the outputted space will be
discrete.
"""
# the last series id, none are negative so we will get a new one
last_series_idx = -1
# restrict down, but since a temporal we need to make a list of entries
_X = []
for row in X:
if row[0] != last_series_idx:
# create a new series to train on, start with empty input
# series then add to them while the row has the same index
last_series_idx = row[0]
_X.append(np.full((0, X.shape[1] - 1), None, dtype=None))
# add the datapoint to the current series
_X[-1] = np.vstack((_X[-1], row[1:]))
# now train an HMM to the data
return HiddenMarkovModel.from_samples(MultivariateGaussianDistribution,
n_components, _X)
def load(self, file_path):
with open(file_path, 'rb') as f:
objects = pickle.load(f)
try:
self.transformer = objects['transformer']
except KeyError: # for backwards compatibility
self.transformer = objects['lda']
self.hmm = HiddenMarkovModel.from_json(objects['hmm'])
def get_variable_number_of_repeats_matcher_hmm(patterns,
copies=1,
vpaths=None):
model = get_constant_number_of_repeats_matcher_hmm(patterns, copies,
vpaths)
start_repeats_matches = State(None, name='start_repeating_pattern_match')
end_repeats_matches = State(None, name='end_repeating_pattern_match')
mat = model.dense_transition_matrix()
states = model.states
states.append(start_repeats_matches)
states.append(end_repeats_matches)
states_count = len(mat)
start_repeats_ind = states_count
end_repeats_ind = states_count + 1
mat = np.c_[mat, np.zeros(states_count), np.zeros(states_count)]
mat = np.r_[mat, [np.zeros(states_count + 2)]]
mat = np.r_[mat, [np.zeros(states_count + 2)]]
unit_ends = []
for i, state in enumerate(model.states):
if state.name.startswith('unit_end'):
unit_ends.append(i)
first_unit_start = None
for i in range(len(mat[model.start_index])):
if mat[model.start_index][i] != 0:
first_unit_start = i
mat[model.start_index][first_unit_start] = 0.0
mat[model.start_index][start_repeats_ind] = 1
mat[start_repeats_ind][first_unit_start] = 1
for unit_end in unit_ends:
next_state = None
for j in range(len(mat[unit_end])):
if mat[unit_end][j] != 0:
next_state = j
mat[unit_end][next_state] = 0.5
mat[unit_end][end_repeats_ind] = 0.5
mat[end_repeats_ind][model.end_index] = 1
starts = np.zeros(states_count + 2)
starts[model.start_index] = 1.0
ends = np.zeros(states_count + 2)
ends[model.end_index] = 1.0
state_names = [state.name for state in states]
distributions = [state.distribution for state in states]
name = 'Repeat Matcher HMM Model'
new_model = Model.from_matrix(mat,
distributions,
starts,
ends,
name=name,
state_names=state_names,
merge=None)
new_model.bake(merge=None)
return new_model
def hmm(df, emissions, n_states, algorithm):
model = HiddenMarkovModel.from_samples(
distribution=MultivariateGaussianDistribution,
n_components=n_states,
X=df[emissions].to_numpy(),
algorithm=algorithm,
verbose=True,
)
return model
def get_read_matcher_model(left_flanking_region,
right_flanking_region,
patterns,
copies=1,
vpaths=None):
model = get_suffix_matcher_hmm(left_flanking_region)
repeats_matcher = get_variable_number_of_repeats_matcher_hmm(
patterns, copies, vpaths)
right_flanking_matcher = get_prefix_matcher_hmm(right_flanking_region)
model.concatenate(repeats_matcher)
model.concatenate(right_flanking_matcher)
model.bake(merge=None)
mat = model.dense_transition_matrix()
first_repeat_matches = []
repeat_match_states = []
suffix_start = None
for i, state in enumerate(model.states):
if state.name[0] == 'M' and state.name.split('_')[-1] == '0':
first_repeat_matches.append(i)
if state.name[0] == 'M' and state.name.split('_')[-1] not in [
'prefix', 'suffix'
]:
repeat_match_states.append(i)
if state.name == 'suffix_start_suffix':
suffix_start = i
mat[model.start_index][suffix_start] = 0.3
for first_repeat_match in first_repeat_matches:
mat[model.
start_index][first_repeat_match] = 0.7 / len(first_repeat_matches)
for match_state in repeat_match_states:
to_end = 0.7 / len(repeat_match_states)
total = 1 + to_end
for next_state in range(len(mat[match_state])):
if mat[match_state][next_state] != 0:
mat[match_state][next_state] /= total
mat[match_state][model.end_index] = to_end / total
starts = np.zeros(len(model.states))
starts[model.start_index] = 1.0
ends = np.zeros(len(model.states))
ends[model.end_index] = 1.0
state_names = [state.name for state in model.states]
distributions = [state.distribution for state in model.states]
name = 'Read Matcher'
new_model = Model.from_matrix(mat,
distributions,
starts,
ends,
name=name,
state_names=state_names,
merge=None)
new_model.bake(merge=None)
return new_model
def test_sample_from_site():
dists = [
NormalDistribution(5, 1),
NormalDistribution(1, 7),
NormalDistribution(8, 2)
]
trans_mat = np.array([[0.7, 0.3, 0.0], [0.0, 0.8, 0.2], [0.0, 0.0, 0.9]])
starts = np.array([1.0, 0.0, 0.0])
ends = np.array([0.0, 0.0, 0.1])
model = HiddenMarkovModel.from_matrix(trans_mat, dists, starts, ends)
model.plot()
def gaussian_hmm(n_states, lower, upper, variance, model_id):
"""
insantiate a model with random parameters
randomly generates start and transition matrices
generates nomal distrobutions for each state from partition on sequences
"""
np.random.seed(int(time.time()))
model = HiddenMarkovModel(model_id)
# make states with distrobutions from random subsets of timepoints
x = np.linspace(lower, upper, n_states)
states = []
for i in range(n_states):
dist = \
NormalDistribution(x[i], variance)
states.append(State(dist, name=str(i)))
model.add_states(states)
# add uniform start probabilities
start_prob = 1.0 / n_states
start_probs = []
for i in range(n_states):
start_probs.append(start_prob + np.random.ranf())
start_probs = np.array(start_probs)
start_probs = start_probs / start_probs.sum()
for i, state in enumerate(states):
model.add_transition(model.start, state, start_probs[i])
# add transition probabilities proportional to probability of generating
# one state mean from another
for state1 in states:
transitions = []
for other_state in states:
transitions.append(np.exp(state1.distribution.log_probability(
other_state.distribution.parameters[0])) + np.random.ranf())
transitions = np.array(transitions)
transitions = transitions / transitions.sum()
for i, state2 in enumerate(states):
model.add_transition(state1, state2, transitions[i])
model.bake()
print 'Initialized HMM: ', model.name
return model
def init(base_dir):
print base_dir
cluster_directories = \
glob.glob(base_dir + '/*')
initial_clusterings = {}
clusterings = {}
clusterings_models = {}
for cluster_dir in cluster_directories:
try:
clustering_id = cluster_dir.split('/')[-1:][0]
# read initial clusters
initial_clusters = {}
filepath = '/'.join(cluster_dir.split('/') + ['init_assignments.txt'])
lines = (open(filepath, 'r').read().splitlines())
l = 0
while l < len(lines):
cluster_name = lines[l]
cluster_members = lines[l + 1].split('\t')
initial_clusters[cluster_name] = cluster_members
l += 4
initial_clusterings[clustering_id] = initial_clusters
# read final clusters
clusters = {}
filepath = '/'.join(cluster_dir.split('/') + ['assignments.txt'])
lines = (open(filepath, 'r').read().splitlines())
l = 0
while l < len(lines):
cluster_name = lines[l]
cluster_members = lines[l + 1].split('\t')
clusters[cluster_name] = cluster_members
l += 4
clusterings[clustering_id] = clusters
# load models
models = {}
model_files = glob.glob(cluster_dir + '/*')
for model_file in model_files:
try:
model_id = model_file.split('/')[-1:][0]
json = open(model_file).read()
models[model_id] = HiddenMarkovModel.from_json(json)
print 'model loaded from: ', model_file
except:
pass
clusterings_models[clustering_id] = models
except:
pass
return initial_clusterings, clusterings
def lambda_handler(event, context):
# TODO implement
content_object = s3.get_object(Bucket='bhargav-ml-trained-models', Key='pos_model.txt')
file_content = content_object['Body'].read().decode()
json_content = json.loads(file_content)
model = HiddenMarkovModel.from_json(json_content)
sentence = event['body'].split(' ')
output = simplify_decoding(sentence, model)
return {
'statusCode': 200,
'headers': {'Content-Type': 'text/plain', 'Access-Control-Allow-Origin': '*'},
'body': output
}
def get_vntr_matcher_hmm(self, read_length):
"""Try to load trained HMM for this VNTR
If there was no trained HMM, it will build one and store it for later usage
"""
logging.info('Using read length %s' % read_length)
copies = self.get_copies_for_hmm(read_length)
base_name = str(
self.reference_vntr.id) + '_' + str(read_length) + '.json'
stored_hmm_file = settings.TRAINED_HMMS_DIR + base_name
if settings.USE_TRAINED_HMMS and os.path.isfile(stored_hmm_file):
model = Model()
model = model.from_json(stored_hmm_file)
return model
flanking_region_size = read_length
vntr_matcher = self.build_vntr_matcher_hmm(copies,
flanking_region_size)
json_str = vntr_matcher.to_json()
with open(stored_hmm_file, 'w') as outfile:
outfile.write(json_str)
return vntr_matcher
def cluster(self):
if self.preprocessed_data is None:
print("No preprocessed_data attribute found")
return -1
if self.alg == "Kmeans":
from sklearn.cluster import KMeans
km = KMeans(n_clusters=self.K, precompute_distances=True)
km.fit(np.concatenate(
self.preprocessed_data)) #flattens all dates together
self.states = [km.predict(d) for d in self.preprocessed_data]
elif self.alg == "HMM":
from pomegranate import HiddenMarkovModel, MultivariateGaussianDistribution
distribution = MultivariateGaussianDistribution
hmm=HiddenMarkovModel().from_samples(distribution,n_components=self.K\
,X=self.preprocessed_data.copy())
self.states = [
np.array(hmm.predict(d.copy())) for d in self.preprocessed_data
]
else:
print("Unrecognised or undefined clustering algorithm.")
return -1
self.experiment_progress = 2
def create_hidden_MarkovModel(e_df, q_df, start_p_dict):
"""
Creates a Hidden Markov Model based on DataFrame
@args:
- e_df (pd.Dataframe): contains the emission probabilites
- q_df (pd.Dataframe): contains the emission probabilites
"""
model = HiddenMarkovModel(name="Example Model")
'#1: Create a dict for each key in trans. df'
model_dict = {}
for key in q_df.keys().values:
model_dict[key] = {}
'#2: Create the states'
for key in model_dict:
'#2.1.Step Add teh emission prob. to each state, , P(observation | state)'
emission_p = DiscreteDistribution(e_df[key].to_dict())
sunny_state = State(emission_p, name=key)
model_dict[key] = State(emission_p, name=key)
model.add_state(model_dict[key])
'#2.2.Step: Add the start probability for each state'
model.add_transition(model.start, model_dict[key], start_p_dict[key])
'#3.Step: Add the transition probability to each state'
for key, item in q_df.to_dict("index").items():
for item_name, value in item.items():
print(key, " , ", item_name, ": ", value)
tmp_origin = model_dict[key]
tmp_destination = model_dict[item_name]
model.add_transition(tmp_origin, tmp_destination,
q_df.loc[key, item_name])
# finally, call the .bake() method to finalize the model
model.bake()
return model
def hmm(df, num_states):
"HMM program"
# df['value']=df['value'].replace(0,np.nan) #this removes unmappable areas of chr
# df_dropna=df.dropna(subset=['value']) #this removes unmappable areas of chr (NaN is otherwise considered 0)
vals = df["value"].values
model = HiddenMarkovModel.from_samples(NormalDistribution,
X=[vals],
n_components=num_states)
states = model.predict(vals)
# Rename states to increase with mean signal
order = np.argsort(df['value'].groupby(states).mean())
states = [order[s] for s in states]
df["state"] = states
df['state'][np.isnan(df['value'])] = np.nan
return df
def run_test(arg, k):
np.random.seed(k)
exp_type = arg['type']
N = arg['N']
alpha = arg['alpha']
n_comp = arg['n_comp']
norm_params = arg['norm_params']
save_dir = arg['dir']
sequence = generator.Sequence(N, alpha, type=exp_type, params=norm_params)
labels = list(map(myutils.rename_state, sequence.path))
model = HiddenMarkovModel.from_samples(DiscreteDistribution,
n_components=n_comp,
X=[sequence.sequence],
labels=[labels],
algorithm='labeled')
return model, sequence.sequence
def generate(
genre_folder: str,
bpm: int,
beats: int,
steps: int,
onset: str,
components: int,
regex: str,
output: str,
include: bool,
):
"""
This command generates a new unique beat, based on the audio files in the given input folder.
"""
audios = util.read_audio_files(include, Path(genre_folder), regex)
sequences, samples = util.create_knowledge_base(
audios, OnsetAlgorithm(onset.lower()), beats, steps
)
# Create the model
# sequences = [add_up_ones(seq) for seq in sequences]
model: HiddenMarkovModel = HiddenMarkovModel.from_samples(
DiscreteDistribution,
n_components=components,
X=sequences,
algorithm="viterbi",
verbose=True,
name="groover",
)
# model: MarkovChain = MarkovChain.from_samples(X=sequences)
# lengths: List[int] = [len(x) for x in sequences]
sequence = model.sample(length=beats * steps)
sequence = sequences[0]
print(sequence)
# sequence = ones(sequence)
# print(len(sequence))
print(
"BPM: {}, Beats: {}, Steps:{}, Onset Algorithm: {}".format(
bpm, beats, steps, onset
)
)
# Save the beat
util.create_beat(sequence, samples, bpm, beats, steps).save(Path(output))
class HMMWrapper:
def __init__(self):
self.model = HiddenMarkovModel()
self.start = self.model.start
self.end = self.model.end
self.states_before_bake = []
self.states = None
def add_state(self, state, start_prob=0):
self.states_before_bake.append((state, start_prob))
self.model.add_state(state)
def add_transition(self, start_state, end_state, prob):
# print('adding from', start_state.name, 'to', end_state.name, prob)
self.model.add_transition(start_state, end_state, prob)
def bake(self):
starter_states_no_prob = []
free_start_prob = 1.0
for state in self.states_before_bake:
if 'none' not in state[0].name:
if not state[1]:
starter_states_no_prob.append(state)
else:
free_start_prob -= state[1]
print('asignado ' + str(state[1]) + ' a ' + state[0].name)
self.add_transition(self.start, state[0], state[1])
len_no_prob = len(starter_states_no_prob)
starter_prob = free_start_prob / len_no_prob
print(len_no_prob, starter_prob)
for state in starter_states_no_prob:
self.add_transition(self.start, state, starter_prob)
self.model.bake()
self.states = self.model.states
def make_states_from_alignment(self, first_state, last_state, seq_matrix,
name):
columns = column_clasify(seq_matrix)
zones = create_zones(columns)
grouped_states = group_states(zones, name)
add_states(self, grouped_states)
trans = calculate_transitions(first_state, last_state, grouped_states)
apply_transitions(self, trans)
def predict(self, *args, **kwargs):
return self.model.predict(*args, **kwargs)
def fit_hmm(self,
signal_arrays,
state_vectors,
distribution,
state_transition_threshold=1e-4,
**kwargs):
# We want to bunch together artefact states with their
# corresponding "clean" states.
state_vectors = [np.abs(vec) for vec in state_vectors]
# remove 'undefined' samples
# TODO: let pomegranate handle that
signal_arrays = [
arr[vec != 0] for arr, vec in zip(signal_arrays, state_vectors)
]
state_vectors = [vec[vec != 0] for vec in state_vectors]
# Pomegranate expects string labels for valid states and None for invalid states.
# labels = [[str(state) if state != 0 else None for state in vec] for vec in state_vectors]
labels = [[str(state) for state in vec] for vec in state_vectors]
# construct matching state names
# state_names = [str(state) for state in np.unique(np.concatenate(state_vectors)) if state != 0]
state_names = [
str(state) for state in np.unique(np.concatenate(state_vectors))
]
# fit HMM states to transformed signals
signals = [self.transform(arr) for arr in signal_arrays]
hmm = HiddenMarkovModel.from_samples(distribution=distribution,
n_components=len(state_names),
X=signals,
labels=labels,
algorithm='labeled',
state_names=state_names,
**kwargs)
if state_transition_threshold > 0.:
new_hmm = _sparsify_hmm(hmm, state_transition_threshold)
return new_hmm
else:
return hmm
def fit(self, data):
"""
Fits a model---learns transition and emission probabilities
Arguments:
data: list of SMILES
"""
list_data = [list(smiles) for smiles in data]
self.model = HiddenMarkovModel.from_samples(
DiscreteDistribution, n_components=self.n_components,
end_state=True, X=list_data,
init='kmeans||', verbose=self.verbose, n_jobs=self.n_jobs,
max_iterations=self.epochs,
batches_per_epoch=self.batches_per_epoch,
random_state=self.seed
)
self.fitted = True
return self
def load(cls, path):
"""
Loads saved model
Arguments:
path: path to saved .pkl file
Returns:
Loaded HMM
"""
with open(path, "rb") as f:
data = pickle.load(f)
hmm = data['model']
del data['model']
model = cls(**data)
model.model = HiddenMarkovModel.from_json(hmm)
model.fitted = True
return model
def fit(self, X, y=None):
X_processed = self._check_and_preprocess(X, True)
self.hmmmodel = HiddenMarkovModel.from_samples(
NormalDistribution,
self.n_states,
X_processed,
algorithm="baum-welch",
n_jobs=8,
verbose=self.verbose,
batches_per_epoch=20,
max_iterations=self.max_iterations)
self.hmmmodel.bake()
self.decision_scores_ = np.zeros(X.shape[0])
for i, sequence in enumerate(X_processed):
self.decision_scores_[i] = -self.hmmmodel.log_probability(sequence)
self._process_decision_scores()
def create_casas7_HMM_with_prepared_train_and_test_based_on_seq_of_activities(
train_set, list_of_persons_in_train, test_set,
list_of_persons_in_test):
'''
create a single HMM for all of persons
train_set = an ndarray that has train_set for each person separately
test_set =
'''
#concatinate train_sets and test_sets of all of people
number_of_persons = len(train_set)
final_train_set = train_set[0]
final_test_set = test_set[0]
final_train_set_labels = list_of_persons_in_train[0]
final_test_set_labels = list_of_persons_in_test[0]
#print(type(final_train_set) , type(train_set) , type(train_set[1]))
for per in range(1, number_of_persons):
final_train_set = np.concatenate((final_train_set, train_set[per]),
axis=0)
final_test_set = np.concatenate((final_test_set, test_set[per]),
axis=0)
final_train_set_labels = np.concatenate(
(final_train_set_labels, list_of_persons_in_train[per]), axis=0)
final_test_set_labels = np.concatenate(
(final_test_set_labels, list_of_persons_in_test[per]), axis=0)
#r = np.shape(final_train_set)
#for i in range(r[0]):
# print(np.shape(final_train_set[i]))
#final_train_set = np.array([[1,2,3,0,0] , [1,2,0,0,0]], dtype = np.ndarray)
#final_train_set_labels = np.array([1,2] , dtype= np.ndarray)
print(type(final_train_set[11]), np.shape(final_train_set[11]))
print(final_train_set[0:2])
model = HiddenMarkovModel.from_samples(
DiscreteDistribution,
n_components=2,
X=final_train_set,
labels=final_train_set_labels,
algorithm='labeled'
) # according to my tests :D n_components is number of hidden states
print(model)
#return 0
#test
'''predicted_labels = np.zeros_like(actual_labels)
def __init__(self,
length=None,
n_features=None,
initial=None,
match_match=0.9,
delete_insert=0.1,
flank_prob=0): #last is polymorphism dummy
super(ProfileHMM, self).__init__()
if length is not None:
n_states = 3 * length + 1
#print(self.get_emission_dists(n_states, n_features, initial)[:3])
self.model = HiddenMarkovModel.from_matrix(
transition_probabilities=self.get_transmat(
n_states, match_match, delete_insert),
distributions=self.get_emission_dists(n_states, n_features,
initial),
starts=self.get_startprob(n_states),
ends=self.get_endprob(n_states),
state_names=self.get_state_names(length))
def build_model(self):
distributions = []
for _ in range(self.hidden_size):
emission_probs = np.random.random(self.num_characters)
emission_probs = emission_probs / emission_probs.sum()
distributions.append(
DiscreteDistribution(
dict(zip(self.all_characters, emission_probs))))
trans_mat = np.random.random((self.hidden_size, self.hidden_size))
trans_mat = trans_mat / trans_mat.sum(axis=1, keepdims=1)
starts = np.random.random(self.hidden_size)
starts = starts / starts.sum()
# testing initializations
np.testing.assert_almost_equal(starts.sum(), 1)
np.testing.assert_array_almost_equal(np.ones(self.hidden_size),
trans_mat.sum(axis=1))
self.model = HiddenMarkovModel.from_matrix(trans_mat, distributions,
starts)
self.model.bake()
def create_hmm_from_sample(file_address):
#data, _ , _ = read_sequence_based_CSV_file_with_activity(file_address = file_address, has_header = True , separate_data_based_on_persons = False )
#data = read_data_from_CSV_file(dest_file = file_address, data_type = np.int , has_header = True , return_as_pandas_data_frame = False )
'''
data = np.delete(data , 2, 1)
data = np.delete(data , 2, 1)
data = np.delete(data , 0, 1)
data = np.delete(data , 0, 1)
data = np.delete(data , 0, 1)
print(np.shape(data))
'''
#print(data)
data = np.array([['a', 'b'], ['a', 'b']])
data = np.array([[np.array([1, 2, 3]),
np.array([1, 1, 1])],
[np.array([1, 1, 2]),
np.array([1, 2, 2])]])
data = [
np.array([[1, 2, 3], [1, 2, 3]], np.int32),
np.array([[1, 2, 3], [1, 2, 3]], np.int32),
np.array([[1, 2, 3], [1, 2, 3]], np.int32)
]
print(data)
#data = np.array([[['a' , 'b'] , ['a' , 'a']] , [['a' , 'b'] , ['b' , 'b']]])
#data = create_sequence_of_sensor_events_based_on_activity(address_to_read = file_address, has_header = False, address_for_save = " ", isSave = False)#read_data_from_CSV_file(dest_file = file_address, data_type = numpy.int , has_header = False , return_as_pandas_data_frame = False )
model = HiddenMarkovModel.from_samples(
MultivariateDistribution, n_components=3, X=data
) # according to my tests :D n_components is number of hidden states
#print(model)
#print(model._baum_welch_summarize())
#model.plot()
'''
print("dense_transition_matrix:" , model.dense_transition_matrix())
print("edge_count:" , model.edge_count())
print("edges:" , model.edges)
print("name:" , model.name)
print("state_count:" , model.state_count())
'''
print(model)
def oriHMMParams(self, numdists=3):
"""
Set initial parameters for the Hidden Markov Model (HMM).
"""
# GMM emissions
# 3 Hidden States:
# 0--downstream, 1--no bias, 2--upstream
if numdists == 1:
dists = [
NormalDistribution(-2.5, 7.5),
NormalDistribution(0, 7.5),
NormalDistribution(2.5, 7.5)
]
else:
var = 7.5 / (numdists - 1)
means = [[], [], []]
for i in range(numdists):
means[0].append(i * 7.5 / (numdists - 1) + 2.5)
means[1].append(i * 7.5 * (-1)**i / (numdists - 1))
means[2].append(-i * 7.5 / (numdists - 1) - 2.5)
dists = []
for i, m in enumerate(means):
tmp = []
for j in m:
tmp.append(NormalDistribution(j, var))
mixture = GeneralMixtureModel(tmp)
dists.append(mixture)
# transition matrix
A = [[0.34, 0.33, 0.33], [0.33, 0.34, 0.33], [0.33, 0.33, 0.34]]
starts = np.ones(3) / 3
hmm = HiddenMarkovModel.from_matrix(A,
dists,
starts,
state_names=['0', '1', '2'],
name='mixture{0}'.format(numdists))
return hmm
def init():
m = 1000 # restricts number of genes, used for local testing
gc, mt, track = load_data(m)
state_range = [5, 10, 25, 50, 100]
z_range = [3, 5, 10, 20]
msequences, mlabels = df_to_sequence_list(mt.data)
gsequences, glabels = df_to_sequence_list(gc.data.iloc[:m, :])
sequences = np.concatenate((msequences, gsequences))
labels = np.concatenate((mlabels, glabels))
# noise model trained on all data once
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts)
noise.freeze_distributions()
return gc, mt, sequences, labels, noise, z_range, state_range
class ModelWrapper:
def __init__(self):
self.model = HiddenMarkovModel()
def add_state(self, distribution, name):
state = State(distribution, name=name)
self.model.add_state(state)
return state
def bake(self):
self.model.bake()
def viterbi(self, seq):
return self.model.viterbi(seq)
def add_transition(self, states, next_state_data):
for state in states:
for next_data in next_state_data:
self.model.add_transition(state, next_data[0], next_data[1])
def init(m, seed):
if m == -1:
m = None
gc, mt, track = load_data(m, seed)
msequences, mlabels = df_to_sequence_list(mt.data)
gsequences, glabels = df_to_sequence_list(gc.data)
sequences = np.concatenate((msequences, gsequences), 0)
labels = np.concatenate((mlabels, glabels))
# noise model trained on all data once
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts,
name='noise')
noise.freeze_distributions()
return sequences, labels, noise
def __init__(self,
length=None,
n_features=None,
initial=None,
match_match=0.9,
delete_insert=0.1,
flank_prob=0.9999999):
super(ProfileHMM, self).__init__()
if length is not None:
n_states = 3 * length + 1
transmat = self.get_transmat(n_states, match_match, delete_insert,
length, flank_prob)
#print(transmat.shape)
#np.set_printoptions(edgeitems=10, linewidth=200)
#print(transmat.round(2))
emissions = self.get_emission_dists(n_states, n_features, initial)
self.model = HiddenMarkovModel.from_matrix(
transition_probabilities=transmat,
distributions=emissions,
starts=self.get_startprob(n_states, flank_prob),
ends=self.get_endprob(n_states, flank_prob),
state_names=self.get_state_names(length))
def init(m=None, seed=None):
gc, mt, track = load_data(m, seed)
msequences, mlabels = df_to_sequence_list(mt.data)
gsequences, glabels = df_to_sequence_list(gc.data)
sequences = np.concatenate((msequences, gsequences), 0)
labels = np.concatenate((mlabels, glabels))
# noise model trained on all data once
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts,
name='noise')
noise.freeze_distributions()
# khmm clustering over a range of k and states-per model
k_range = [10, 25, 50, 100, 200]
state_range = [5, 10, 25, 50, 100]
return sequences, labels, noise, k_range, state_range
def create_casas7_hmm(file_address, has_activity):
if has_activity:
list_of_data, list_of_persons, _ = read_sequence_based_CSV_file_with_activity(
file_address=file_address,
has_header=True,
separate_data_based_on_persons=False)
else:
list_of_data, list_of_persons = read_sequence_based_CSV_file_without_activity(
file_address=file_address,
has_header=True,
separate_data_based_on_persons=False)
model = ""
try:
model = HiddenMarkovModel.from_samples(DiscreteDistribution,
n_components=5,
X=list_of_data,
algorithm='baum-welch')
#model = HiddenMarkovModel.from_samples(DiscreteDistribution, n_components=2, X=list_of_data , labels = list_of_persons , algorithm = 'labeled' )
except KeyError:
print('there is an exception')
print(model)
#print((list_of_persons[0]))
print("np.shape(list_of_data):", np.shape(list_of_data))
#print(model._baum_welch_summarize())
model.plot()
print("dense_transition_matrix:", model.dense_transition_matrix())
print("edge_count:", model.edge_count())
print("edges:", model.edges)
print("name:", model.name)
print("state_count:", model.state_count())
#print("summarize:" , model.summarize())
print(model.thaw())
#!/usr/bin/env python2.7 # example.py: Yet Another Hidden Markov Model library # Contact: Jacob Schreiber ( [email protected] ) """ A simple example highlighting how to build a model using states, add transitions, and then run the algorithms, including showing how training on a sequence improves the probability of the sequence. """ import random from pomegranate import * from pomegranate import HiddenMarkovModel as Model random.seed(0) model = Model(name="ExampleModel") distribution = UniformDistribution(0.0, 1.0) state = State(distribution, name="uniform") state2 = State(NormalDistribution(0, 2), name="normal") silent = State(None, name="silent") model.add_state(state) model.add_state(state2) model.add_transition(state, state, 0.4) model.add_transition(state, state2, 0.4) model.add_transition(state2, state2, 0.4) model.add_transition(state2, state, 0.4) model.add_transition(model.start, state, 0.5) model.add_transition(model.start, state2, 0.5) model.add_transition(state, model.end, 0.2)
infinite model, with no extra work! This change is passed on to all the algorithms. ''' from pomegranate import * from pomegranate import HiddenMarkovModel as Model import itertools as it import numpy as np # Define the states s1 = State( NormalDistribution( 5, 2 ), name="S1" ) s2 = State( NormalDistribution( 15, 2 ), name="S2" ) s3 = State( NormalDistribution( 25, 2 ), name="S3 ") # Define the transitions model = Model( "infinite" ) model.add_transition( model.start, s1, 0.7 ) model.add_transition( model.start, s2, 0.2 ) model.add_transition( model.start, s3, 0.1 ) model.add_transition( s1, s1, 0.6 ) model.add_transition( s1, s2, 0.1 ) model.add_transition( s1, s3, 0.3 ) model.add_transition( s2, s1, 0.4 ) model.add_transition( s2, s2, 0.4 ) model.add_transition( s2, s3, 0.2 ) model.add_transition( s3, s1, 0.05 ) model.add_transition( s3, s2, 0.15 ) model.add_transition( s3, s3, 0.8 ) model.bake() sequence = [ 4.8, 5.6, 24.1, 25.8, 14.3, 26.5, 15.9, 5.5, 5.1 ]
s2 = State(b, name="M2")
s22 = State(bb, name="M22")
s222 = State(bbb, name="M222")
s2222 = State(bbbb, name="M2222")
s22222 = State(bbbbb, name="M22222")
s222222 = State(bbbbbb, name="M222222")
s3 = State(c, name="M3")
s33 = State(cc, name="M33")
s333 = State(ccc, name="M333")
s3333 = State(cccc, name="M3333")
s33333 = State(ccccc, name="M33333")
s333333 = State(cccccc, name="M333333")
hmm = HiddenMarkovModel()
hmm.add_states(s1, s11, s111, s2, s22, s222, s3, s33, s333,
s1111, s11111, s111111, s2222, s22222, s222222, s3333, s33333,
s333333)
hmm.add_transition(hmm.start, s1, 1.)
hmm.add_transition(hmm.start, s11, 1.)
hmm.add_transition(hmm.start, s111, 1.)
hmm.add_transition(hmm.start, s2, 1.)
hmm.add_transition(hmm.start, s22, 1.)
hmm.add_transition(hmm.start, s222, 1.)
hmm.add_transition(hmm.start, s3, 1.)
hmm.add_transition(hmm.start, s33, 1.)
hmm.add_transition(hmm.start, s333, 1.)
hmm.add_transition(hmm.start, s1111, 1.)
hmm.add_transition(hmm.start, s11111, 1.)
def prep(cluster_directory_root, depth, genefile):
# load data
gc, mt, track = load_data(None, 0)
genes = load(open(genefile, 'r'))
gc.data = gc.data.loc[genes, :]
data = pd.concat([gc.data, mt.data])
labels = data.index.values
original_labels = labels
pos_labels = labels + '+'
neg_labels = labels + '-'
pos_data = pd.DataFrame(data=data.as_matrix(), index=pos_labels,
columns=data.columns.values)
neg_data = pd.DataFrame(data=(data.as_matrix() * -1), index=neg_labels,
columns=data.columns.values)
data = pd.concat([data, pos_data, neg_data])
print data.index.values
generic_dir = cluster_directory_root.split('/') + (['*'] * depth)
generic_dir = ('/').join(generic_dir)
cluster_directories = \
glob.glob(generic_dir)
clusterings = {}
clusterings_models = {}
for cluster_dir in cluster_directories:
try:
clustering_id = cluster_dir.split('/')[-1:][0]
# read final clusters
clusters = {}
filepath = '/'.join(cluster_dir.split('/') + ['assignments.txt'])
lines = (open(filepath, 'r').read().splitlines())
l = 0
while l < len(lines):
cluster_name = lines[l]
cluster_members = lines[l + 1].split('\t')
if cluster_members == ['']:
cluster_members = []
clusters[cluster_name] = cluster_members
l += 4
clusterings[clustering_id] = clusters
# load models
models = {}
model_files = glob.glob(cluster_dir + '/*')
for model_file in model_files:
try:
model_id = model_file.split('/')[-1:][0]
json = open(model_file).read()
models[model_id] = HiddenMarkovModel.from_json(json)
print 'model loaded from: ', model_file
except:
pass
clusterings_models[clustering_id] = models
except:
pass
"""
background = set()
for clustering in clusterings.itervalues():
for cid, members in clustering.iteritems():
background.update(set(members))
"""
background = list(original_labels)
# data = data.loc[background, :]
# generate ranomd clusterings of the same size k as our models
random_clusterings = {}
np.random.seed(int(time.time()))
for clustering_id, clustering in clusterings.iteritems():
source = np.array(background)
random_assignments = np.random.choice(len(clustering), source.size)
random_clusters = {}
for i, cluster_id in enumerate(clustering.iterkeys()):
random_clusters[cluster_id] = \
source[np.where(random_assignments == i)[0]].tolist()
random_clusterings[clustering_id] = random_clusters
# generate random signed clustering
random_signed_clusterings = {}
pn = np.array(['+', '-'])
for clustering_id, clustering in clusterings.iteritems():
source = np.array(background)
random_assignments = np.random.choice(len(clustering), source.size)
random_clusters = {}
for i, cluster_id in enumerate(clustering.iterkeys()):
members = source[np.where(random_assignments == i)[0]].tolist()
signed_members = []
for member in members:
sign = np.random.choice(pn, 1)[0]
signed_members.append(member + sign)
random_clusters[cluster_id] = signed_members
random_signed_clusterings[clustering_id] = random_clusters
return clusterings, random_clusterings, random_signed_clusterings,\
clusterings_models, data, original_labels
def init_lr_hmm(sequences, steps, states_per_step,
force_end=False, model_id='Left-Righ HMM', seed=None):
"""
insantiate a left-right model with random parameters
randomly generates start and transition matrices
generates nomal distrobutions for each state from partition on sequences
force_end if we require sequence to end in end state
"""
# seed random number generator
if seed is not None:
np.random.seed(seed)
model = HiddenMarkovModel(model_id)
n_states = steps * states_per_step
# make distrobutions from chronological subsets of timepoints
step_size = int(math.ceil(sequences.shape[1] / float(n_states+1)))
# generate states
states = np.empty((steps, states_per_step), dtype=object)
for i in range(steps):
for j in range(states_per_step):
temp_assignment = np.arange(step_size * i, step_size * (i+1))
dist = \
NormalDistribution.from_samples(sequences[:, temp_assignment])
state_name = str(i) + '-' + str(j)
states[i, j] = State(dist, name=str(state_name))
# add states to model
model.add_states(states.flatten().tolist())
# make random transition from start -> step0
trans = np.random.ranf(states_per_step)
trans = trans / trans.sum()
for j in range(states_per_step):
model.add_transition(model.start, states[0, j], trans[j])
# make random transition from step(i) -> step(i+1)
for i in range(steps-1):
for j in range(states_per_step):
trans = np.random.ranf(states_per_step + 1)
trans = trans / trans.sum()
# self transition
model.add_transition(states[i, j], states[i, j], trans[0])
# out transition
for x in range(states_per_step):
model.add_transition(states[i, j], states[i + 1, x],
trans[x + 1])
# make random transition from stepn -> end
if force_end:
for j in range(states_per_step):
trans = np.random.ranf(2)
trans = trans / trans.sum()
# self transition
model.add_transition(states[(steps - 1), j],
states[(steps - 1), j], trans[0])
# end transition
model.add_transition(states[(steps - 1), j], model.end, trans[1])
model.bake()
print 'Initialized Left-Right HMM:', model.name, '[', \
steps, states_per_step, ']'
return model
def init_cycle_hmm(sequences, steps, states_per_step, model_id):
"""
insantiate a left-right model with random parameters
randomly generates start and transition matrices
generates nomal distrobutions for each state from partition on sequences
"""
model = HiddenMarkovModel(model_id)
n_states = steps * states_per_step
# make distrobutions from chronological subsets of timepoints
step_size = int(math.ceil(sequences.shape[1] / float(n_states+1)))
# generate states
states = np.empty((steps, states_per_step), dtype=object)
for i in range(steps):
for j in range(states_per_step):
temp_assignment = np.arange(step_size * i, step_size * (i+1))
dist = \
NormalDistribution.from_samples(sequences[:, temp_assignment])
state_name = str(i) + '-' + str(j)
states[i, j] = State(dist, name=str(state_name))
# add states to model
model.add_states(states.flatten().tolist())
# make random transition from start -> step0
trans = np.random.ranf(n_states)
trans = trans / trans.sum()
for i, state in enumerate(states.flatten().tolist()):
model.add_transition(model.start, state, trans[i])
# make random transition from step(i) -> step(i+1)
for i in range(steps-1):
for j in range(states_per_step):
trans = np.random.ranf(states_per_step + 1)
trans = trans / trans.sum()
# self transition
model.add_transition(states[i, j], states[i, j], trans[0])
# out transition
for x in range(states_per_step):
model.add_transition(states[i, j], states[i + 1, x],
trans[x + 1])
# make random transition from stepn -> step0
for j in range(states_per_step):
trans = np.random.ranf(states_per_step + 1)
trans = trans / trans.sum()
# self transition
model.add_transition(states[(steps - 1), j], states[(steps - 1), j],
trans[0])
# out transition
for x in range(states_per_step):
model.add_transition(states[(steps - 1), j], states[0, x],
trans[x + 1])
model.bake()
print 'Initialized Cyclic State HMM:', '[', \
steps, states_per_step, ']'
return model
def run(cluster_directory_root, depth, plottype):
# load data
gc, mt, track = load_data(None, 0)
data = pd.concat([gc.data, mt.data])
labels = data.index.values
pos_labels = labels + '+'
neg_labels = labels + '-'
pos_data = pd.DataFrame(data=data.as_matrix(), index=pos_labels,
columns=data.columns.values)
neg_data = pd.DataFrame(data=data.as_matrix(), index=neg_labels,
columns=data.columns.values)
data = pd.concat([data, pos_data, neg_data])
generic_dir = cluster_directory_root.split('/') + (['*'] * depth)
generic_dir = ('/').join(generic_dir)
cluster_directories = \
glob.glob(generic_dir)
clusterings = {}
clusterings_models = {}
for cluster_dir in cluster_directories:
try:
clustering_id = cluster_dir.split('/')[-1:][0]
# read final clusters
clusters = {}
filepath = '/'.join(cluster_dir.split('/') + ['assignments.txt'])
lines = (open(filepath, 'r').read().splitlines())
l = 0
while l < len(lines):
cluster_name = lines[l]
cluster_members = lines[l + 1].split('\t')
clusters[cluster_name] = cluster_members
l += 4
clusterings[clustering_id] = clusters
# load models
models = {}
model_files = glob.glob(cluster_dir + '/*')
for model_file in model_files:
try:
model_id = model_file.split('/')[-1:][0]
json = open(model_file).read()
models[model_id] = HiddenMarkovModel.from_json(json)
print 'model loaded from: ', model_file
except:
pass
clusterings_models[clustering_id] = models
except:
pass
background = set()
for clustering in clusterings.itervalues():
for cid, members in clustering.iteritems():
background.update(set(members))
background = list(background)
# data = data.loc[background, :]
# generate ranomd clusterings of the same size k as our models
random_clusterings = {}
for clustering_id, clustering in clusterings.iteritems():
source = np.array(background)
random_assignments = np.random.choice(len(clustering), source.size)
random_clusters = {}
for i, cluster_id in enumerate(clustering.iterkeys()):
random_clusters[cluster_id] = \
source[np.where(random_assignments == i)[0]].tolist()
random_clusterings[clustering_id] = random_clusters
# run dunn and davies_bouldin for clusterings and random permutations
rand_dunn = report_dunn(random_clusterings, clusterings_models, data)
savename = cluster_directory_root + 'dunn_index_random'
dump(rand_dunn, open(savename, 'w'))
rand_davies = report_davies_bouldin(random_clusterings, clusterings_models,
data)
savename = cluster_directory_root + 'davies_bouldin_index_random'
dump(rand_davies, open(savename, 'w'))
if plottype == 'none':
pass
elif plottype == 'kn_grid':
rand_dunn_df = pd.DataFrame()
rand_davies_df = pd.DataFrame()
for clustering_id, clustering in clusterings.iteritems():
cid = clustering_id.replace('k', '_'). \
replace('n', '_').split('_')
m = cid[0]
k = int(cid[1])
n = int(cid[2])
rand_dunn_df.loc[k, n] = rand_dunn[clustering_id]
rand_davies_df.loc[k, n] = rand_davies[clustering_id]
rand_davies_df = rand_davies_df.fillna(0)
rand_dunn_df = rand_dunn_df.fillna(0)
rand_dunn_df = rand_dunn_df.sort_index().sort_index(1)
rand_davies_df = rand_davies_df.sort_index().sort_index(1)
odir = cluster_directory_root
title = 'RANDOM_' + str(m) + ': Dunn Index'
HeatMap(rand_dunn_df.as_matrix(), rand_dunn_df.index.values,
rand_dunn_df.columns.values,
title=title, odir=odir)
odir = cluster_directory_root
title = 'RANDOM_' + str(m) + ': Davies-Bouldin Index'
HeatMap(rand_davies_df.as_matrix(), rand_davies_df.index.values,
rand_davies_df.columns.values,
title=title, odir=odir)
elif plottype == 'row':
rand_dunn_df = pd.Series()
rand_davies_df = pd.Series()
for clustering_id, clustering in clusterings.iteritems():
rand_dunn_df.loc[clustering_id] = rand_dunn[clustering_id]
rand_davies_df.loc[clustering_id] = rand_davies[clustering_id]
rand_davies_df = rand_davies_df.fillna(0)
rand_dunn_df = rand_dunn_df.fillna(0)
rand_dunn_df = rand_dunn_df.sort_index()
rand_davies_df = rand_davies_df.sort_index()
odir = cluster_directory_root
title = 'RANDOM' + ': Dunn Index'
HeatMap(rand_dunn_df.as_matrix().reshape(-1, 1),
rand_dunn_df.index.values,
[' '], title=title, odir=odir, cmin=0, cmax=.5)
odir = cluster_directory_root
title = 'RANDOM' + ': Davies-Bouldin Index'
HeatMap(rand_davies_df.as_matrix().reshape(-1, 1),
rand_davies_df.index.values,
[' '], title=title, odir=odir, cmin=5, cmax=10)
return clusterings, clusterings_models
X_1 = X[y == 0]
X_2 = X[y == 1]
X_3 = X[y == 2]
else:
X_1 = X[2000:4000]
X_2 = X[400:800]
X_3 = X[7000:8000]
a = MultivariateGaussianDistribution.from_samples(X_1)
b = MultivariateGaussianDistribution.from_samples(X_2)
c = MultivariateGaussianDistribution.from_samples(X_3)
s1 = State(a, name="M1")
s2 = State(b, name="M2")
s3 = State(c, name="M3")
hmm = HiddenMarkovModel()
hmm.add_states(s1, s2, s3)
hmm.add_transition(hmm.start, s1, 0.34)
hmm.add_transition(hmm.start, s3, 0.33)
hmm.add_transition(hmm.start, s2, 0.33)
hmm.add_transition(s1, s1, 0.9)
hmm.add_transition(s1, s2, 0.05)
hmm.add_transition(s1, s3, 0.05)
hmm.add_transition(s2, s1, 0.05)
hmm.add_transition(s2, s3, 0.05)
hmm.add_transition(s2, s2, 0.9)
hmm.add_transition(s3, s3, 0.9)
hmm.add_transition(s3, s2, 0.05)
cc = MultivariateGaussianDistribution.from_samples(X_33) ccc = MultivariateGaussianDistribution.from_samples(X_333) s1 = State(a, name="M1") s11 = State(aa, name="M11") s111 = State(aaa, name="M111") s2 = State(b, name="M2") s22 = State(bb, name="M22") s222 = State(bbb, name="M222") s3 = State(c, name="M3") s33 = State(cc, name="M33") s333 = State(ccc, name="M333") hmm = HiddenMarkovModel() hmm.add_states(s1, s11, s111, s2, s22, s222, s3, s33, s333) hmm.add_transition(hmm.start, s1, 0.12) hmm.add_transition(hmm.start, s11, 0.11) hmm.add_transition(hmm.start, s111, 0.11) hmm.add_transition(hmm.start, s2, 0.11) hmm.add_transition(hmm.start, s22, 0.11) hmm.add_transition(hmm.start, s222, 0.11) hmm.add_transition(hmm.start, s3, 0.11) hmm.add_transition(hmm.start, s33, 0.11) hmm.add_transition(hmm.start, s333, 0.11) hmm.add_transition(s1, s1, 0.92) hmm.add_transition(s1, s11, 0.01)
def gen_cluster_plots(cluster_directory_root, depth):
# load data
gc, mt, track = load_data(None, 0)
data = pd.concat([gc.data, mt.data])
labels = data.index.values
pos_labels = labels + '+'
neg_labels = labels + '-'
pos_data = pd.DataFrame(data=data.as_matrix(), index=pos_labels,
columns=data.columns.values)
neg_data = pd.DataFrame(data=data.as_matrix(), index=neg_labels,
columns=data.columns.values)
data = pd.concat([data, pos_data, neg_data])
generic_dir = cluster_directory_root.split('/') + (['*'] * depth)
generic_dir = ('/').join(generic_dir)
cluster_directories = \
glob.glob(generic_dir)
clusterings = {}
clusterings_models = {}
for cluster_dir in cluster_directories:
try:
clustering_id = cluster_dir.split('/')[-1:][0]
# read final clusters
clusters = {}
filepath = '/'.join(cluster_dir.split('/') + ['assignments.txt'])
lines = (open(filepath, 'r').read().splitlines())
l = 0
while l < len(lines):
cluster_name = lines[l]
cluster_members = lines[l + 1].split('\t')
clusters[cluster_name] = cluster_members
l += 4
clusterings[clustering_id] = clusters
# load models
models = {}
model_files = glob.glob(cluster_dir + '/*')
for model_file in model_files:
try:
model_id = model_file.split('/')[-1:][0]
json = open(model_file).read()
models[model_id] = HiddenMarkovModel.from_json(json)
print 'model loaded from: ', model_file
except:
pass
clusterings_models[clustering_id] = models
except:
pass
background = set()
for clustering in clusterings.itervalues():
for cid, members in clustering.iteritems():
background.update(set(members))
background = list(background)
# data = data.loc[background, :]
# generate ranomd clusterings of the same size k as our models
for clustering_id, clustering in clusterings.iteritems():
for model_id, members in clustering.iteritems():
sequences = data.loc[members, :]
pltdir = '/'.join(cluster_directory_root.split('/') + ['plots'])
# make line plots directory
if not os.path.isdir(pltdir + '/line'):
print "Creating directory...", pltdir
os.mkdir(pltdir + '/line')
savename = pltdir + '/line/' + model_id + '_lineplot'
plt_title = model_id + ' Line Plot'
ax = sequences.T.plot(legend=False, rot=2)
ax.set_title(plt_title)
ax.set_xlabel('Timepoint')
ax.set_ylabel('Normalized Expression')
print 'Saving: ', savename
fig = ax.get_figure()
fig.savefig(savename)
fig.clear()
# make autocorr plots directory
if not os.path.isdir(pltdir + '/autocorr'):
print "Creating directory...", pltdir
os.mkdir(pltdir + '/autocorr')
savename = pltdir + '/autocorr/' + model_id + '_autocorr'
plt_title = model_id + ' Autocorr Plot'
for seq in sequences.index:
ax = autocorrelation_plot(sequences.loc[seq])
ax.set_title(plt_title)
print 'Saving: ', savename
fig = ax.get_figure()
fig.savefig(savename)
fig.clear()
# make lag plots directory
if not os.path.isdir(pltdir + '/lag'):
print "Creating directory...", pltdir
os.mkdir(pltdir + '/lag')
from pylab import *
NUM_COLORS = len(members)
cm = get_cmap('gist_rainbow')
colors = []
for i in range(NUM_COLORS):
colors.append(cm(1.*i/NUM_COLORS))
savename = pltdir + '/lag/' + model_id + '_lagplot'
plt_title = model_id + ' Lag Plot'
for i, seq in enumerate(sequences.index):
ax = lag_plot(sequences.loc[seq], c=colors[i])
ax.set_title(plt_title)
print 'Saving: ', savename
fig = ax.get_figure()
fig.savefig(savename)
fig.clear()
"""
def gen_model(sequences, labels, algorithm, initialization, restarts, n, k,
out_dir, base_id, tied):
if initialization == 'rand':
init_method = init_gaussian_hmm
init_args = {'n_states': n[0]}
if init_lr_hmm == 'lr':
init_method = init_lr_hmm
s = n[0]
sps = n[1]
init_args = {'steps': s, 'state_per_step': sps, 'force_end': True}
if initialization == 'cycle':
init_method = init_cycle_hmm
s = n[0]
sps = n[1]
init_args = {'steps': s, 'state_per_step': sps}
best = 0
best_score = -1e1000
# genes/metabolites will be assigned to noise model if other models
# fail to model it better
noise_dist = [NormalDistribution(0, 1)]
noise_trans = np.array([[1]])
starts = np.array([1])
noise = HiddenMarkovModel.from_matrix(noise_trans, noise_dist, starts,
name='noise')
noise.freeze_distributions()
np.random.seed(int(time.time()))
randassigns = []
for x in range(restarts):
randassigns.append(np.random.randint(k, size=labels.size))
for x in range(restarts):
randassign = randassigns[x]
assignments = {}
for i in range(k):
model_id = str(i)
assignments[model_id] = \
np.where(randassign == i)[0].tolist()
in_model = assignments[model_id]
print assignments
# gen model for number of restarts
for x in range(restarts):
try:
collection_id = base_id + '_' + str(x)
odir = '/'.join(out_dir.split('/') + [collection_id])
print 'Learning: ', collection_id
# generate random initial assignments
# initialize models on random assignments
randassign = randassigns[x]
assignments = {}
models = {}
for i in range(k):
model_id = str(i)
assignments[model_id] = \
np.where(randassign == i)[0].tolist()
in_model = assignments[model_id]
models[model_id] = \
init_method(sequences[in_model, :], model_id=model_id,
**init_args)
# add noise model
models['noise'] = noise
assignments['noise'] = []
# all are un-fixed
fixed = {}
for model_id, model in models.iteritems():
fixed[model_id] = []
# perform clustering
models, assignments, c = cluster(models=models,
sequences=sequences,
assignments=assignments,
algorithm=algorithm,
fixed=fixed, tied=tied,
labels=labels,
odir=odir)
score = total_log_prob(models, sequences, assignments)
if best_score < score:
best_score = score
best = collection_id
bestfile = '/'.join(out_dir.split('/') + ['best'])
with open(bestfile, 'w') as f:
print >> f, collection_id
f.close()
except:
error_file = odir.split('/') + ['errors.txt']
error_file = '/'.join(error_file)
f = open(error_file, 'a')
print >> f, 'error computing parameters for: ', collection_id
print >> f, "Unexpected error:", sys.exc_info()[0]
f.close()
return best
def init_gaussian_hmm(sequences, n_states, model_id, seed=None):
"""
insantiate a model with random parameters
randomly generates start and transition matrices
generates nomal distrobutions for each state from partition on sequences
"""
"""
# make random transition probability matrix
# scale each row to sum to 1
trans = np.random.ranf((n_states, n_states))
for i in range(n_states):
trans[i, :] = trans[i, :] / trans[i, :].sum()
# make distrobutions from random subsets of timepoints
x = int(math.ceil(sequences.shape[1] / float(n_states)))
# x = math.min(3, x)
dists = []
for i in range(n_states):
temp_assignment = np.random.choice(sequences.shape[1], x)
dists.append(NormalDistribution.from_samples
(sequences[:, temp_assignment]))
# random start probabilities
# scale to sum to 1
starts = np.random.ranf(n_states)
starts = starts / sum(starts)
model = HiddenMarkovModel.from_matrix(trans, dists, starts, name=model_id)
"""
# seed random numer generator
if seed is not None:
np.random.seed(seed)
model = HiddenMarkovModel(model_id)
# make states with distrobutions from random subsets of timepoints
x = int(math.ceil(sequences.shape[1] / float(n_states)))
states = []
for i in range(n_states):
temp_assignment = np.random.choice(sequences.shape[1], x)
dist = \
NormalDistribution.from_samples(sequences[:, temp_assignment])
states.append(State(dist, name=str(i)))
model.add_states(states)
# add random start probabilities
start_probs = np.random.ranf(n_states)
start_probs = start_probs / start_probs.sum()
for i, state in enumerate(states):
model.add_transition(model.start, state, start_probs[i])
# add random transition probabilites out of each state
for state1 in states:
transitions = np.random.ranf(n_states)
transitions = transitions / transitions.sum()
for i, state2 in enumerate(states):
model.add_transition(state1, state2, transitions[i])
model.bake()
print 'Initialized HMM: ', model.name
return model
# [email protected] """ Example rainy-sunny HMM using yahmm. Example drawn from the wikipedia HMM article: http://en.wikipedia.org/wiki/Hidden_Markov_model describing what Bob likes to do on rainy or sunny days. """ from pomegranate import * from pomegranate import HiddenMarkovModel as Model import random import math random.seed(0) model = Model( name="Rainy-Sunny" ) # Emission probabilities rainy = State( DiscreteDistribution({ 'walk': 0.1, 'shop': 0.4, 'clean': 0.5 }), name='Rainy' ) sunny = State( DiscreteDistribution({ 'walk': 0.6, 'shop': 0.3, 'clean': 0.1 }), name='Sunny' ) model.add_transition( model.start, rainy, 0.6 ) model.add_transition( model.start, sunny, 0.4 ) # Transition matrix, with 0.05 subtracted from each probability to add to # the probability of exiting the hmm model.add_transition( rainy, rainy, 0.65 ) model.add_transition( rainy, sunny, 0.25 ) model.add_transition( sunny, rainy, 0.35 ) model.add_transition( sunny, sunny, 0.55 )
