def __init__(self, model_file=None, components=None):
if os.path.exists(model_file):
self.model = joblib.load(model_file)
else:
alu_file = 'Alu_sequence.pkl'
if os.path.exists(alu_file):
locis = joblib.load(alu_file)
else:
locis = read_sequence('hg19_Alu.bed', 0)
locis = random.sample(locis, 100000)
for l in tqdm(locis):
l.init_seq()
l.decode_seq()
locis = list(filter(lambda l: l.seq is not None, locis))
joblib.dump(locis, alu_file)
print('Alu Loaded')
locis = locis[0:5000]
model = MultinomialHMM(n_components=components,
verbose=True,
n_iter=50)
x = np.concatenate(list(map(attrgetter('seq'), locis)))
x = np.reshape(x, [x.shape[0], 1])
length = list(map(attrgetter('length'), locis))
model.fit(x, length)
self.model = model
joblib.dump(self.model, model_file)
def main():
rand_p_matrix = np.random.rand(4, 4)
rand_b_matrix = np.random.rand(4, 3)
print("\nGernerating p matrix...............")
p_matrix = normalization(rand_p_matrix)
print(p_matrix)
print("\nGernerating b matrix...............")
b_matrix = normalization(rand_b_matrix)
print(b_matrix)
# Generate 1000 observations
O, _ = generate_observation(1000, p_matrix, b_matrix)
# training the selection of number of states
aic = []
bic = []
likelihood = []
m = 3
print("\nTraining the HMM for selection of number of states........")
for n in range(2, 30):
observations = LabelEncoder().fit_transform(O)
model = MultinomialHMM(n_components=n, random_state=200263453)
model.fit(np.atleast_2d(observations))
logL = model.score(np.atleast_2d(observations))
p = compute_p(n, m)
a = AIC(logL, p)
b = BIC(logL, observations, p)
likelihood.append(logL)
aic.append(a)
bic.append(b)
plot(aic, 'AIC')
plot(bic, 'BIC')
plot(likelihood, 'Log likelihood')
def fit_hmm_learn(X, n_states):
samples = np.concatenate(X)
lengths = [len(x) for x in X]
hmm_learn_model = MultinomialHMM(n_components=n_states)
hmm_learn_model.fit(samples, lengths)
# Label data using hmmlearn model
return hmm_learn_model.predict(samples, lengths)
def train_hmm():
"""
HMM for sequence learning.
"""
print "Loading training data..."
train_sequence, num_classes = get_sequence("./train_data/*")
print "Build HMM..."
model = MultinomialHMM(n_components=2)
print "Train HMM..."
model.fit([train_sequence])
def train_hmm():
"""
HMM for sequence learning.
"""
print "Loading training data..."
train_sequence, num_classes = get_sequence("./train_data/*")
print "Build HMM..."
model = MultinomialHMM(n_components=2)
print "Train HMM..."
model.fit([train_sequence])
class BKT:
"""
Implements the Bayesian Knowledge Tracing model. This only
implements the Viterbi and EM algorithms. These may be used
together to implement an Intelligent Tutoring System.
"""
def __init__(self, observed):
"""
Initializes the object and sets the internal state.
Args:
observed: array-like, shape (n_samples, n_features)
"""
self.observed = np.array(observed)
if len(self.observed.shape) == 1:
self.observed = self.observed.reshape(-1, 1)
# TODO: Check other parameters to this constructor
self.model = MultinomialHMM(n_components=2, n_iter=100)
def fit(self) -> None:
"""
Fits the model to the observed states. Uses the EM algorithm
to estimate model parameters.
"""
self.model.fit(self.observed)
def get_model_params(self) -> tuple:
"""
Returns the model parameters. This must be run only after
calling the `fit` function.
Returns:
(A, pi, B): The start probabilities, the transition
probabilities, and the emission probabilities.
"""
return np.round_(self.model.startprob_, 2), np.round_(self.model.transmat_, 2), \
np.round_(self.model.emissionprob_, 2)
def predict(self, sequence) -> np.array:
"""
Returns the most likely hidden state sequence corresponding to
`sequence`.
Args:
sequence: List of observable states
Returns:
state_sequence: Array
"""
return self.model.predict(sequence)
def test_DiscreteHMM_decode(cases: str) -> None:
np.random.seed(12346)
cases = int(cases)
i = 1
N_decimal = 4
while i < cases:
tol=1e-3
n_samples = np.random.randint(10, 50)
hidden_states = np.random.randint(3, 6)
# symbols is the number of unqiue observation types.
symbols = np.random.randint(4, 9)
X = []
lengths = []
for _ in range(n_samples):
# the actual length is seq_length + 1
seq_length = symbols
this_x = np.random.choice(range(symbols), size=seq_length, replace=False)
X.append(this_x)
lengths.append(seq_length)
max_iter = 100
hmm_gold = MultinomialHMM(n_components=hidden_states, n_iter=100, tol=tol)
X_gold = np.concatenate(X).reshape((-1,1))
hmm_gold.fit(X_gold, lengths)
gold_A = hmm_gold.transmat_
gold_B = hmm_gold.emissionprob_
gold_pi = hmm_gold.startprob_
gold_logprob, gold_state_sequence = hmm_gold.decode(X_gold, lengths)
hmm_mine = DiscreteHMM(hidden_states=hidden_states,
symbols=symbols,
A=gold_A,
B=gold_B,
pi=gold_pi)
mine_logprob_list = []
mine_state_sequence = []
for this_x in X:
this_mine_logprob, this_mine_state_sequence = hmm_mine.decode(this_x)
mine_logprob_list.append(this_mine_logprob)
mine_state_sequence.append(this_mine_state_sequence)
mine_state_sequence = np.concatenate(mine_state_sequence)
mine_logprob = sum(mine_logprob_list)
assert_almost_equal(mine_logprob, gold_logprob, decimal=N_decimal)
assert_almost_equal(mine_state_sequence, gold_state_sequence, decimal=N_decimal)
i+=1
print('Successfully testing the function of computing decodes in discrete HMM!')
def main():
rand_p_matrix = np.random.rand(4, 4)
rand_b_matrix = np.random.rand(4, 3)
print("\nGernerating p matrix...............")
p_matrix = normalization(rand_p_matrix)
print(p_matrix)
print("\nGernerating b matrix...............")
b_matrix = normalization(rand_b_matrix)
print(b_matrix)
# Generate 1000 observations
O, Q = generate_observation(1000, p_matrix, b_matrix)
O_seq = [1, 2, 3, 3, 1, 2, 3, 3, 1, 2, 3, 3]
pi = (1, 0, 0, 0)
print("\nThe Orginal Observation Sequence O: {}".format(O[:12]))
print("The probability 𝑝(𝑂|𝜆) is {} with O: {}".format(
forward(O_seq, p_matrix, b_matrix, pi)[-1].sum(), O_seq))
print("\nThe Orginal Sequence Q: {}".format(Q[:12]))
print("The Most Probable Sequence Q: {} with O: {}".format(
list(viterbi(O_seq, p_matrix, b_matrix, pi)), O_seq))
obersvations = LabelEncoder().fit_transform(O)
model = MultinomialHMM(n_components=4)
model.fit(np.atleast_2d(obersvations))
est_pi = model.startprob_
est_p = model.transmat_
est_b = model.emissionprob_
print("\nThe estimated transition matrix P:\n {}".format(est_p))
print("\nThe estimated event matrix B:\n {}".format(est_b))
print("\nThe estimated start probability pi:\n {}".format(est_pi))
_, p = chisquare(p_matrix, est_p, axis=None)
print("\np-value of transition matrix P: {}".format(p))
_, p = chisquare(b_matrix, est_b, axis=None)
print("p-value of event matrix B: {}".format(p))
_, p = chisquare(pi, est_pi, axis=None)
print("p-value of start probability pi: {}".format(p))
def run_hmm_model(input_df, n_unique, A_df, Eta, n_iter = 10000,
tol=1e-2, verbose = False, params = 'e', init_params = ''):
'''
Runs the hmm model and returns the predicted results, score and model
input_df : The dataframe of keypresses
n_unique : number of unqique chars
A_df : Dataframe of trasnmission matrix
Eta : Emissions matrix
n_iter : Max number of iterations for hmm
tol : The value to stop the hmm model if score does not improve by more than this
verbose : Whether or not to print out
params : Parameters to tune
init_params : Paramters to initialize
'''
# Propotion of characters starting words in english
char_counts = get_char_counts()
# Construct model
hmm = MultinomialHMM(n_components=n_unique, startprob_prior=np.append(0, char_counts.values),
transmat_prior=A_df.values, algorithm='viterbi',
random_state=None, n_iter=n_iter, tol=tol,
verbose=verbose, params=params, init_params=init_params)
# Set values
hmm.emissionprob_ = Eta
hmm.transmat_ = A_df.values
hmm.startprob_ = np.append(0, char_counts.values)
# Feed in the clusters as the expected output
model_input = input_df['cluster'].values
# Reshape
if len(model_input.shape) == 1:
model_input = model_input.reshape((len(model_input), 1))
# Fit the model
hmm = hmm.fit(model_input)
# Score model
score, results = hmm.decode(model_input)
return score, results, hmm
def predict(self, day_to_predict):
# Get records of 30 days before day_to_predict
previous_thirty_days = get_previous_month(self.time_series, day_to_predict)
binary_crime_sequence = previous_thirty_days['Violent Crime Committed?'].values.tolist()
# Unsupervised HMM can't account for string of identical emissions.
# If we see such a string, just predict the same emission for the following day.
if binary_crime_sequence == [1]*30:
return True
if binary_crime_sequence == [0]*30:
return False
votes = []
# Train nine HMMs. They are initialized randomly, so we take "votes" from nine HMMs.
# Why 9? Odd numbers preclude ties.
# And nine is a decent tradeoff between performance and getting bad results by chance
for _ in range(3):
# Train HMM
model = MultinomialHMM(n_components=3, n_iter=10000)
model.fit([np.array(binary_crime_sequence)])
# Determine the most likely state of the last day in the sequence
last_state_probs = model.predict_proba(binary_crime_sequence)[-1]
current_state = self.get_most_likely(last_state_probs)
# Determine the most likely state of the day we're trying to predict
transition_probs = model.transmat_[current_state]
next_state = self.get_most_likely(transition_probs)
# Determine the most likely emission (crime/no crime) from a day in that state
emissions = model.emissionprob_[next_state]
vote = self.get_most_likely(emissions)
# Record this HMM's vote
votes.append(vote)
# Votes are 1 for crime, 0 for no crime. Return True if majority votes for crime.
return sum(votes) > 1
def test_HMM():
np.random.seed(12345)
np.set_printoptions(precision=5, suppress=True)
P = default_hmm()
ls, obs = P["latent_states"], P["obs_types"]
# generate a new sequence
O = generate_training_data(P, n_steps=30, n_examples=25)
tol = 1e-5
n_runs = 5
best, best_theirs = (-np.inf, []), (-np.inf, [])
for _ in range(n_runs):
hmm = MultinomialHMM()
A_, B_, pi_ = hmm.fit(O, ls, obs, tol=tol, verbose=True)
theirs = MHMM(
tol=tol,
verbose=True,
n_iter=int(1e9),
transmat_prior=1,
startprob_prior=1,
algorithm="viterbi",
n_components=len(ls),
)
O_flat = O.reshape(1, -1).flatten().reshape(-1, 1)
theirs = theirs.fit(O_flat, lengths=[O.shape[1]] * O.shape[0])
hmm2 = MultinomialHMM(A=A_, B=B_, pi=pi_)
like = np.sum([hmm2.log_likelihood(obs) for obs in O])
like_theirs = theirs.score(O_flat, lengths=[O.shape[1]] * O.shape[0])
if like > best[0]:
best = (like, {"A": A_, "B": B_, "pi": pi_})
if like_theirs > best_theirs[0]:
best_theirs = (
like_theirs,
{
"A": theirs.transmat_,
"B": theirs.emissionprob_,
"pi": theirs.startprob_,
},
)
print("Final log likelihood of sequence: {:.5f}".format(best[0]))
print("Final log likelihood of sequence (theirs): {:.5f}".format(
best_theirs[0]))
plot_matrices(P, best, best_theirs)
class HMM_Learner:
def __init__(self, M):
self.con = MultinomialHMM(n_components=M)
self.incon = MultinomialHMM(n_components=M)
self.daID = {
'ass': 0,
'bck': 1,
'be.neg': 2,
'be.pos': 3,
'el.ass': 4,
'el.inf': 5,
'el.sug': 6,
'el.und': 7,
'fra': 8,
'inf': 9,
'off': 10,
'oth': 11,
'stl': 12,
'sug': 13,
'und': 14
}
self.da_choose_n = itertools.combinations([
'ass', 'bck', 'be.neg', 'be.pos', 'el.ass', 'el.inf', 'el.sug',
'el.und', 'fra', 'inf', 'off', 'oth', 'stl', 'sug', 'und'
], 4)
def addRandomAllSequence(self, X, lengths):
da_keys = self.daID.keys()
random.shuffle(da_keys)
X1 = [[self.daID[x.lower().strip()]] for x in da_keys]
X.append(X1)
lengths.append(len(X1))
def trainHMMs(self, topics, sequences, labels):
try:
self.con = pickle.load(open('HMM_consistent.model', 'rb'))
self.incon = pickle.load(open('HMM_inconsistet.model', 'rb'))
except:
X_con = []
l_con = []
X_incon = []
l_incon = []
for t in topics:
try:
temp = sequences[t]
temp = labels[t]
except:
continue
if sequences[t]:
X1 = [[self.daID[da.lower().strip()]]
for da in sequences[t]]
if 'weak' in labels[t].lower():
X_incon.append(X1)
l_incon.append(len(sequences[t]))
else:
X_con.append(X1)
l_con.append(len(sequences[t]))
# Add two complete random sequence to support Multinomial in HMMs
self.addRandomAllSequence(X_incon, l_incon)
self.addRandomAllSequence(X_con, l_con)
self.con.fit(np.concatenate(X_con), l_con)
self.incon.fit(np.concatenate(X_incon), l_incon)
pickle.dump(self.con, open('HMM_consistent.model', 'wb'))
pickle.dump(self.incon, open('HMM_inconsistet.model', 'wb'))
def testHMMs(self, topics, sequences):
prediction = {}
for t in topics:
try:
temp = sequences[t]
except:
continue
if sequences[t]:
X1 = [[self.daID[da.lower().strip()]] for da in sequences[t]]
c = self.con.score(np.concatenate([X1]), [len(sequences[t])])
i = self.incon.score(np.concatenate([X1]), [len(sequences[t])])
prediction[t] = (c, i)
return prediction
def generateLabelSequence(self, sequence):
MIN_VAL = -10000000
topics = []
sequences = {}
isConsistent = False
max_score_da_seq = ''
max_score = MIN_VAL
for n in xrange(2, 5):
da_choose_n = [
p for p in itertools.product([
'ass', 'bck', 'be.neg', 'be.pos', 'el.ass', 'el.inf',
'el.sug', 'el.und', 'fra', 'inf', 'off', 'oth', 'stl',
'sug', 'und'
],
repeat=n)
]
#print da_choose_n
for s in da_choose_n:
temp_sequence = copy.deepcopy(sequence)
temp_sequence.extend(list(s))
topics.append(str(s))
sequences[str(s)] = temp_sequence
scores = self.testHMMs(topics, sequences)
for t in topics:
if scores[t][0] > max_score:
max_score = scores[t][0]
max_score_da_seq = t
if scores[t][0] > scores[t][1]:
isConsistent = True
max_score = scores[t][0]
max_score_da_seq = t
if isConsistent:
break
print(max_score, max_score_da_seq)
return max_score_da_seq, isConsistent
### preparing the testdata as the list of numbers
## task -3
dummy_data = []
for i in s:
if i == ' ':
dummy_data.append(26)
else:
dummy_data.append(ord(i) - ord('A'))
training_data = np.array(dummy_data)
training_data = training_data.reshape((training_data.shape[0], 1))
### hmm model
hmm_model = MultinomialHMM(n_components=2, n_iter=500, tol=0.01, verbose=False)
hmm_model.fit(training_data)
print(hmm_model.monitor_)
print("Tranisition probalitity of this model is \n")
print(hmm_model.transmat_)
print("\n")
print("Emission probalitity of this model is \n")
print(np.transpose(hmm_model.emissionprob_))
print("\n")
## the seven most probable characters
transition_prob1 = transition_prob
emission_prob1 = np.transpose(hmm_model.emissionprob_)
print("For this trained model, the seven most likely charcters are\n")
print("For state 0, the seven most likely characters are\n")
class TOE_HMM_CHARS:
def __init__(self, N=2, maxIters = 200):
self._N = N
self._M = COMPONENTS
self._maxIters = maxIters
self._syms = []
def loadBrownSymsSeq(self, T):
taggedWordsIter = brown.tagged_words()
retIdx = 0
symSequence = []
for wrd, tag in taggedWordsIter:
if wrd:
for c in wrd:
val = ord(c)
symSequence.append(val)
retIdx += 1
if retIdx >= T:
break
self._syms = symSequence
self._syms = np.concatenate((self._syms, np.arange(256))).tolist()
return symSequence
def textSeqToSymSeq(self, txtSeqArr):
symSequence =[]
for wrd in txtSeqArr:
if wrd:
for c in wrd:
val = ord(c)
symSequence.append(val)
return symSequence
def initHMM(self):
self._hmm = MultinomialHMM(n_components=self._N, n_iter=self._maxIters,
verbose=True, params='ste', init_params='ste')
# n_features (int) Number of possible symbols emitted by the model (in the samples).
# monitor_ (ConvergenceMonitor) Monitor object used to check the convergence of EM.
# transmat_ (array, shape (n_components, n_components)) Matrix of transition probabilities between states.
# startprob_ (array, shape (n_components, )) Initial state occupation distribution.
# emissionprob_ (array, shape (n_components, n_features)) Probability of emitting a given symbol when in each state.
def trainHMM(self):
self._hmm.fit(np.array(self._syms).reshape(-1, 1))
def testTxt(self, txtSeqArr):
testSymsArr = self.textSeqToSymSeq(txtSeqArr)
score = self._hmm.score(testSymsArr)
return score
def testSyms(self, symsArr):
score = self._hmm.score(symsArr)
return score
def trainedLambda(self):
A = self._hmm.transmat_
B = self._hmm.emissionprob_
pi = self._hmm.startprob_
return (A, B, pi)
def persistHMM(self, filename):
import pickle
s = pickle.dumps(self)
with open(filename, 'w') as f:
f.write(s)
@staticmethod
def loadHMM(filename):
import pickle
with open(filename, 'r') as f:
s = f.read()
model = pickle.loads(s)
return model
def pickRandomSeq(self, length = 100):
symSequence = [random.randint(0, 255) for idx in range(length)]
return symSequence
def pickOrderedSeq(self, length = 100):
symSequence = []
taggedWordsIter = brown.tagged_words()
maxIdx = len(taggedWordsIter)
maxMinIdx = maxIdx - length
minIdx = random.randint(0, maxMinIdx)
count = 0
idx = minIdx
while True:
(wrd, tag) = taggedWordsIter[idx]
if wrd:
for c in wrd:
val = ord(c)
symSequence.append(val)
count += 1
idx += 1
if count >= length:
break
return symSequence
def printHMM(self):
print("A = %s" % str(self._hmm.transmat_))
print("B = %s" % str(self._hmm.emissionprob_))
print("PI = %s" % str(self._hmm.startprob_))
print("Verify A = %s" % np.sum(self._hmm.transmat_, axis=1))
print("Verify B = %s" % np.sum(self._hmm.emissionprob_, axis=1))
print("Verify PI = %s" % np.sum(self._hmm.startprob_, axis=0))
def histo(self):
retHisto = dict((x, self._syms.count(x)) for x in range(256))
return retHisto
n_states = 2
n_emissions = len(possible_emissions)
# Training data
X = np.array([
random.sample(possible_emissions, len(possible_emissions)), [1, 1, 2, 1],
[6, 5, 5, 4, 7, 7]
])
lengths = [len(row) for row in X]
X = np.atleast_2d(np.concatenate(X))
# Create randomly initialized model
model = MultinomialHMM(n_components=2, n_iter=100)
# Train on data
model.fit(X.T, lengths=lengths)
# Trained parameters
transition_matrix = model.transmat_
emission_matrix = model.emissionprob_
initial_stat_probability = model.startprob_
# Observed sequence
observed = np.array([1, 1, 1])
# Get helmet
helmet = 0
# Sequence emissions
helping_emission = (helmet, 1, 0)
backstabbing_emission = (helmet, 0, 1)
def train(self, data, labels, tp=None):
labels = np.array(labels)
for i in range(self.nb_class):
print "Class", i
ind = np.where(labels == i)
digit_data = np.array(data)[ind]
self.fit_encode_class(digit_data, i)
sks, lengths = self.transform_encode_class(digit_data, i)
if not tp:
model = MultinomialHMM(n_components=self.nb_components,
n_iter=self.max_iter,
tol=self.tol,
verbose=True,
params='ste',
init_params='e')
init = 1. / self.nb_components
model.startprob_ = np.full(self.nb_components, init)
model.transmat_ = np.full((self.nb_components, self.nb_components),
init)
else:
model = model = MultinomialHMM(n_components=self.nb_components,
n_iter=self.max_iter,
tol=self.tol,
verbose=True,
params='ste')
# Number of distinct centroids
num_obs = len(np.unique(np.concatenate(sks)))
model.emissionprob_ = np.zeros((self.nb_components, num_obs))
hist = {}
curr = 0
bucket_len = num_obs / self.nb_components
for j in range(self.nb_components):
if j == self.nb_components - 1 and curr + bucket_len < num_obs:
offset = num_obs - curr - bucket_len
for k in range(curr, curr + bucket_len + offset):
if not j in hist:
hist[j] = []
hist[j].append(k)
model.emissionprob_[j, k] = 1
curr += bucket_len + offset
else:
for k in range(curr, curr + bucket_len):
if not j in hist:
hist[j] = []
hist[j].append(k)
model.emissionprob_[j, k] = 1
curr += bucket_len
model.startprob_ = np.zeros(self.nb_components)
# always ends by penup
model.startprob_[-1] = 1
model.transmat_ = np.zeros((self.nb_components, self.nb_components))
state_occ_count = np.zeros(self.nb_components)
for example in digit_data:
j = 0
prevobs = 0
for obs in example:
le = self.les[i]
val = le.transform(obs)
if j == 0:
prevobs = val
j += 1
continue
prevobs_state = None
obs_state = None
for k in range(self.nb_components):
if (prevobs_state != None and obs_state != None):
break
if prevobs in hist[k]:
prevobs_state = k
if val in hist[k]:
obs_state = k
state_occ_count[prevobs_state] += 1
model.transmat_[prevobs_state, obs_state] += 1
prevobs = val
j += 1
for j in range(self.nb_components):
for k in range(self.nb_components):
model.transmat_[j, k] = model.transmat_[j, k] / state_occ_count[j]
model.fit(sks, lengths)
self.models[i] = model
def convert(string): # mapping function, map A->0 , B->1 , C->3 ...
output = []
for character in string:
if character is " ":
number = 26
else:
number = ord(character) - 65
output.append(number)
return (output)
data2 = convert(data) #Convert the data from stream of chacters to stream of numbers
DD = np.array(data2)
Data_arr = DD.reshape((DD.shape[0],1))
model = MultinomialHMM(n_components=2,n_iter=200, tol=0.01, verbose=False)
print("Training started")
model.fit(Data_arr)
print("Training Done")
print("Model = ",model.monitor_)
print("The transition prob of this trained model : ")
print(model.transmat_)
emiso = np.transpose(model.emissionprob_)
print("\nThe emmision prob of this trained model : ")
print(" State-0 State-1")
print(emiso)
seven_most_probabe(emiso) #printing the 7 most likely characters
print("Stationary probbabilities : ",model. get_stationary_distribution())
print("So seeing the emission probabilities we can say that State 1 is Consonant and State 0 is Vowel")
print("\nTask - 4")
model_nat = MultinomialHMM(n_components=2)
model_nat.transmat_ = trans_prob
if train_data['return'][i] < 0.0 and analysis_data['v'][i] == 2:
emission_probability[1][2] += 1
if train_data['return'][i] < 0.0 and analysis_data['v'][i] == 3:
emission_probability[1][3] += 1
emission_probability[0] /= sum(1 for e in train_data['return'] if e >= 0.0)
emission_probability[1] /= sum(1 for e in train_data['return'] if e < 0.0)
#print(emission_probability)
hmm = MultinomialHMM(n_components=n_states)
hmm.startprob = start_probability
hmm.transmat = transition_probability
hmm.emissionprob = emission_probability
bob_says = np.array([[0, 2, 1, 1, 2, 0]]).T
hmm = hmm.fit(bob_says)
logprob, alice_hears = hmm.decode(bob_says, algorithm="viterbi")
print("Bob says:", ", ".join(map(lambda x: observations[x], bob_says)))
print("Alice hears:", ", ".join(map(lambda x: states[x], alice_hears)))
'''
law_data['hmm_states'] = hmm.predict(rets)
panel = Figure_Util.Figure()
panel.draw(law_data, title='close', subplots=['hmm_states'], figsize=(20, 10))
'''
db.disconnect()
def test_DiscreteHMM_fit(cases: str) -> None:
np.random.seed(12346)
cases = int(cases)
i = 1
N_decimal = 4
max_iter = 100
tol=1e-3
while i < cases:
n_samples = np.random.randint(10, 50)
hidden_states = np.random.randint(3, 6)
# symbols is the number of unqiue observation types.
symbols = np.random.randint(4, 9)
X = []
lengths = []
for _ in range(n_samples):
# the actual length is seq_length + 1
seq_length = symbols
this_x = np.random.choice(range(symbols), size=seq_length, replace=False)
X.append(this_x)
lengths.append(seq_length)
A = np.full((hidden_states, hidden_states),1/hidden_states)
B = []
for _ in range(hidden_states):
this_B = np.random.dirichlet(np.ones(symbols),size=1)[0]
B.append(this_B)
B = np.array(B)
pi = np.ones(hidden_states)
pi = pi/hidden_states
hmm_gold = MultinomialHMM(n_components=hidden_states,
startprob_prior=1,
transmat_prior=1,
init_params='',
n_iter=max_iter,
tol=tol)
hmm_gold.transmat_ = A
hmm_gold.emissionprob_ = B
hmm_gold.startprob_ = pi
X_gold = np.concatenate(X).reshape((-1,1))
hmm_gold.fit(X_gold, lengths)
gold_A = hmm_gold.transmat_
gold_B = hmm_gold.emissionprob_
gold_pi = hmm_gold.startprob_
hmm_mine = DiscreteHMM(hidden_states=hidden_states,
symbols=symbols,
A=A,
B=B,
pi=pi,
tol=tol,
max_iter=max_iter)
hmm_mine.fit(X)
mine_A = hmm_mine.A
mine_B = hmm_mine.B
mine_pi = hmm_mine.pi
assert_almost_equal(mine_pi, gold_pi, decimal=N_decimal)
assert_almost_equal(mine_A, gold_A, decimal=N_decimal)
assert_almost_equal(mine_B, gold_B, decimal=N_decimal)
i+=1
print('Successfully testing the function of estimating parameters in discrete HMM!')
def computeHMM(dataset, alphabet, num_matchstates=9):
num_sequences = len(dataset)
best_score = None
best_model = None
alphabet = list(alphabet)
residue_mapper = {alphabet[j]: j for j in range(0, len(alphabet))}
#one begin, one end, num_matchstates + 1 insert states, num_matchstates match states, num_matchstates deletion states.
num_states = 3 + 3 * num_matchstates
concat_dataset = np.concatenate([[[residue_mapper[x]] for x in y]
for y in dataset])
dataset_lengths = [len(x) for x in dataset]
for x in range(0, 10):
transition_matrix = np.zeros((num_states, num_states))
emission_matrix = np.zeros((num_states, len(alphabet)))
#first num_matchstates + 2 are the matchstates (including beginning and end, though those two are mute
#first do B, then M_1,...,M_m
#B goes to either I_0 or M_1.
b_row = ProfileHMM.compute_random_row(2)
transition_matrix[0][1] = b_row[0]
transition_matrix[0][2] = b_row[1]
for i in range(1, num_matchstates + 1):
#go to either match state, insertion state, or delete state.
m_row = ProfileHMM.compute_random_row(3)
#next match state
transition_matrix[i][i + 1] = m_row[0]
#insert state
transition_matrix[i][i + num_matchstates + 2] = m_row[1]
#deletion state
print('i: %d' % i)
transition_matrix[i][i + 2 * num_matchstates + 2] = m_row[2]
emission_matrix[i] = ProfileHMM.compute_random_row(
len(alphabet))
#now we do the insertion states.
for i in range(num_matchstates + 2, 2 * num_matchstates + 3):
#either go to self, or next match state.
row = ProfileHMM.compute_random_row(2)
transition_matrix[i][i] = row[0]
transition_matrix[i][i - (num_matchstates + 1)] = row[1]
emission_matrix[i] = ProfileHMM.compute_random_row(
len(alphabet))
#now do deletion states. In the loop, do all but the last one
for i in range(2 * num_matchstates + 3, 3 * num_matchstates + 2):
row = ProfileHMM.compute_random_row(2)
transition_matrix[i][i] = row[0]
transition_matrix[i][i - 2 * num_matchstates - 1] = row[1]
model = MultinomialHMM(num_states, params="ets")
model.n_features = len(alphabet)
start_prob = np.zeros(num_states)
start_prob[0] = 1.0
print('start prob array')
print(start_prob)
model.startprob_ = start_prob
model.transmat_ = transition_matrix
model.emissionprob_ = emission_matrix
try:
model.fit(concat_dataset, dataset_lengths)
except ValueError:
pdb.set_trace()
print('model')
print(model)
"""
for row in range(0, len(model.emissionprob_)):
for col in range(0, len(model.emissionprob_[row])):
count = model.emissionprob_[row][col]*num_sequences
model.emissionprob_[row][col] = (count + 0.01)/(num_sequences + len(alphabet)*0.01)
"""
print('emission probabilities')
print(model.emissionprob_)
score = model.score(concat_dataset, dataset_lengths)
if x == 0:
best_score = score
best_model = model
elif score > best_score:
best_score = score
best_model = model
return best_model
model = MultinomialHMM(n_components=3,
random_state=42,
params='e',
init_params='e')
model.startprob_ = [0.16, 0.04, 0.8]
model.transmat_ = [[0.67, 0.13, 0.2], [0, 0.5, 0.5], [0, 0, 1]]
'''
startprob_:
V:
D:
J:
trasmat_:
from each row, row become column (the probability of row become column, each row sum to 1, not column)
'''
model.fit(X, length)
emission = model.emissionprob_
model.n_features
model.transmat_
model.startprob_
model.get_stationary_distribution()
# let's test
train_score = []
for i in cdr3_train_index:
test = string2matrix_plain(cdr3[i]).astype(np.int)
score = model.score(test)
train_score.append(score)
train_score = np.array(train_score)
test_score = []
for i in cdr3_test_index:
def main(params):
DEBUG = params['DEBUG']
dataset = params['dataset']
nh_part = params['nh_part']
nh_chords = params['nh_chords']
num_gen = params['num_gen']
##################################################################
# DATA PROCESSING
# Songs indices
song_indices = [
43, 85, 133, 183, 225, 265, 309, 349, 413, 471, 519, 560, 590, 628,
670, 712, 764, 792, 836, 872, 918, 966, 1018, 1049, 1091, 1142, 1174,
1222, 1266, 1278, 1304, 1340, 1372, 1416, 1456, 1484, 1536, 1576, 1632,
1683, 1707, 1752, 1805, 1857, 1891, 1911
]
# Chords mapping
chord_names = [
'C;Em', 'A#;F', 'Dm;Em', 'Dm;G', 'Dm;C', 'Am;Em', 'F;C', 'F;G', 'Dm;F',
'C;C', 'C;E', 'Am;G', 'F;Em', 'F;F', 'G;G', 'Am;Am', 'Dm;Dm', 'C;A#',
'Em;F', 'C;G', 'G#;A#', 'F;Am', 'G#;Fm', 'Am;Gm', 'F;E', 'Dm;Am',
'Em;Em', 'G#;G#', 'Em;Am', 'C;Am', 'F;Dm', 'G#;G', 'F;A#', 'Am;G#',
'C;D', 'G;Am', 'Am;C', 'Am;A#', 'A#;G', 'Am;F', 'A#;Am', 'E;Am',
'Dm;E', 'A;G', 'Am;Dm', 'Em;Dm', 'C;F#m', 'Am;D', 'G#;Em', 'C;Dm',
'C;F', 'G;C', 'A#;A#', 'Am;Caug', 'Fm;G', 'A;A'
]
# Import .mat file
dataset_root = os.path.join('data', dataset)
mat_path = os.path.join(dataset_root, 'data.mat')
data_mat = sio.loadmat(mat_path)
chords_per_part = 2
chords_per_bar = 4
num_chords = 56
num_parts = 4
sub_sampling_ratio_parts = chords_per_bar / chords_per_part
# Get parts
parts_data_ = (np.dot(np.transpose(data_mat["feats"][-num_parts:]),
np.asarray(range(num_parts))).astype(int)).reshape(
-1, 1)
# Group by bar
parts_data = parts_data_[::sub_sampling_ratio_parts]
# Parts with position in bar. Used condition chords generation
parts_bar_data = post_processing_parts(parts_data,
sub_sampling_ratio_parts)
# Get chords transitions
chords_data = (np.dot(np.transpose(data_mat["feats"][:-num_parts]),
np.asarray(range(num_chords))).astype(int)).reshape(
-1, 1)
#################################
# Group by song
parts_length = []
chords_length = []
start_ind = 0
for end_ind in song_indices:
chords_length.append(end_ind - start_ind + 1)
start_ind = end_ind + 1
parts_length = [e / 2 for e in chords_length]
##################################################################
##################################################################
# PARTS
# Compute HMM for part modeling
hmm_part = MultinomialHMM(n_components=nh_part, n_iter=20)
hmm_part.fit(parts_data, parts_length)
# def plot_mat(matrix, name):
# fig = plt.figure()
# ax = fig.add_subplot(1,1,1)
# ax.set_aspect('equal')
# plt.imshow(matrix, interpolation='nearest', cmap=plt.cm.ocean)
# plt.colorbar()
# plt.savefig(name, format='pdf')
# plot_mat(hmm_part.transmat_, 'part_transmat.pdf')
# plot_mat(np.reshape(hmm_part.startprob_, [-1, 1]), 'part_startprob.pdf')
# plot_mat(hmm_part.emissionprob_, 'part_emissionprob.pdf')
##################################################################
##################################################################
# CHORDS
hmm_chords = MultinomialHMM_prod(n_components=nh_chords, n_iter=20)
hmm_chords.fit(chords_data, chords_length)
# plot_mat(hmm_chords.transmat_, 'chords_transmat.pdf')
# plot_mat(np.reshape(hmm_chords.startprob_, [-1, 1]), 'chords_startprob.pdf')
# plot_mat(hmm_chords.emissionprob_, 'chords_emissionprob.pdf')
##################################################################
#################################
# GENERATION
# Sample sequence
for n in range(num_gen):
gen_part_sequence_, _ = hmm_part.sample(params["gen_seq_length"])
gen_part_sequence = post_processing_parts(gen_part_sequence_,
sub_sampling_ratio_parts)
# Compute conditioning on parts
p_chords_given_partBar = build_proba(chords_data, parts_bar_data)
gen_chord_sequence, _ = hmm_chords.sampling_prod_hmm(
p_chords_given_partBar, gen_part_sequence)
######## T E S T ################
# Independent HMM ?
# gen_chord_sequence, _ = hmm_chords.sampling(n_samples=44)
##################################
if params["DEBUG"]:
with open("results_chords/" + str(n), 'wb') as f:
for count, (part, chord) in enumerate(
zip(gen_part_sequence, gen_chord_sequence)):
if count % 2 == 0:
f.write(
str(part / 2) + " ; " + chord_names[chord[0]] +
"\n")
else:
f.write(" ; " + chord_names[chord[0]] + "\n")
if count % 8 == 7:
f.write("\n")
gen_part_sequence = [e / 2 for e in gen_part_sequence]
return gen_part_sequence, gen_chord_sequence, num_chords, num_parts
def train_syllable_hmm(song_corpus, n_iterations=50):
hmm = MultinomialHMM(3)
hmm.transmat_ = np.array([[0, 0, 1], [1, 0, 0], [0, .01, .99]])
hmm.n_iter = n_iterations
hmm.fit([np.concatenate(song_corpus)])
return hmm
discrete_obs, delta_hws, delta_fas = [], [], []
for idx in mice:
d = _data_on_mouse(data, idx, smoothing_time_radius,
smoothing_amplitude_radius, smoothing_tolerance,
sampling_interval, bins)
discrete_obs.append(d[0])
delta_hws.append(d[1])
delta_fas.append(d[2])
X = np.array(discrete_obs)
model = MultinomialHMM(n_components=n_components)
predictions = []
for i in range(7):
held_out_X = np.vstack((X[:i], X[i + 1:]))
model.fit(held_out_X)
predictions.append(model.decode(X[i].reshape(X[i].shape[0], 1)))
f, axarr = plt.subplots(7, 1)
yranges = np.arange(n_components + 1, dtype=float) / n_components
colors = plt.cm.rainbow(np.linspace(0, 1, n_components))
for i in range(7):
states, indices = _axvspan_maker(predictions[i][1])
for s, idxs in zip(states, indices):
axarr[i].axvspan(idxs[0],
idxs[1],
ymin=yranges[s],
ymax=yranges[s + 1],
color=colors[s])
plt.show()
target_test = np.hstack([d["label"] for d in dataset[split_pos:]])
# Train scaler
scaler = StandardScaler()
scaler.fit(feature_train)
feature_train = scaler.transform(feature_train)
# Train random forest classifier
clf = RandomForestClassifier()
clf.fit(feature_train, target_train)
# Train HMM
pred_probs = clf.predict_proba(feature_train)[:, 1]
pred_labels = np.array([map_pred(x) for x in pred_probs], dtype=np.int64)
hmm = MultinomialHMM(n_components=2,
startprob_prior=np.array([0.5, 0.5]),
transmat_prior=np.array([
[0.8, 0.2],
[0.2, 0.8],
]))
hmm.fit(pred_labels.reshape(-1, 1))
# Evaluation of the entire procedure
predict_results = infer(feature_test, scaler, clf, hmm)
print(classification_report(target_test, predict_results))
# Save models
pickle.dump(scaler, open(path.join(project_dir, "models/scaler.pkl"), "wb"))
pickle.dump(clf, open(path.join(project_dir, "models/clf.pkl"), "wb"))
pickle.dump(hmm, open(path.join(project_dir, "models/hmm.pkl"), "wb"))
for idx in mice:
d = _data_on_mouse(data, idx, smoothing_time_radius,
smoothing_amplitude_radius, smoothing_tolerance,
sampling_interval, bins)
discrete_obs.append(d[0])
delta_hws.append(d[1])
delta_fas.append(d[2])
X = np.array(discrete_obs)
model = MultinomialHMM(n_components = n_components)
predictions = []
for i in range(7):
held_out_X = np.vstack((X[:i], X[i+1:]))
model.fit(held_out_X)
predictions.append(model.decode(X[i].reshape(X[i].shape[0], 1)))
f, axarr = plt.subplots(7, 1)
yranges = np.arange(n_components+1, dtype=float)/n_components
colors = plt.cm.rainbow(np.linspace(0, 1, n_components))
for i in range(7):
states, indices = _axvspan_maker(predictions[i][1])
for s, idxs in zip(states, indices):
axarr[i].axvspan(idxs[0], idxs[1], ymin=yranges[s], ymax=yranges[s+1], color=colors[s])
plt.show()
# healthy_model = MultinomialHMM(n_components = n_components)
# healthy_model.fit(dos)
# hs_preds = healthy_model.predict(dos.reshape(len(dos), 1))
def train_on_X(X):
X_train_hmm, X_train_lengths = transform_X_for_hmm(X)
clf = MultinomialHMM(n_components=n_components, n_iter=n_iter)
clf.fit(X_train_hmm, lengths=X_train_lengths)
return clf
final_array = []
count_array = []
for x in input_sequences:
count = 0
for y in x:
count += 1
final_array.append(y)
count_array.append(count)
data = np.loadtxt('train.csv' , delimiter=',')
sample_vector = np.array(final_array)
sequence_lengths = np.array(count_array)
num_components = 3
model = MultinomialHMM(n_components=num_components , n_iter = 1000)
model.fit(sample_vector , lengths = sequence_lengths)
#------------------------------------------------------------------------------------------------------------------------
print("Second Phase")
validating_sequences = []
# for idx in train.index:
count =0
idx = 0
flag1 = True
flag2 = True
while flag1:
temp_list = []
flag2 = True
high = high + 1
elif percent >= .50:
highMid = highMid + 1
elif percent >= .25:
lowMid = lowMid + 1
else:
low = low + 1
matrix[1, 0] = low / len(wins)
matrix[1, 1] = lowMid / len(wins)
matrix[1, 2] = highMid / len(wins)
matrix[1, 3] = high / len(wins)
return matrix
# Load Data
filename = 'data.csv'
X = np.loadtxt(filename, delimiter=',')
player1 = X[:, 0]
player2 = X[:, 1]
record = X[:, 2]
print "stateProbs(record)", stateProbs(record)
print "eProbs(player1, record", eProbs(player1, record)
clf = MultinomialHMM(n_components=2)
clf.transmat_ = stateProbs(record)
clf.emissionprob_ = eProbs(player1, record)
print "here"
clf.fit(clf.transmat_, clf.emissionprob_)
clf.predict(player1)
class TOE_HMM:
def __init__(self, N=2, maxIters = 200):
self._N = N
self._M = len(WHITE_LIST)
self._pi = self.randProbMat(1,N)[0]
self._A = self.equiProbMat(N, N)
self._B = self.equiProbMat(N, self._M)
self._maxIters = maxIters
self._syms = []
def randProbMat(self, M, N):
ret = np.random.rand(M,N)
ret = ret/ret.sum(axis=1)[:,None]
return ret
def equiProbMat(self, M, N):
ret = np.ones((M,N), dtype=float)
ret = ret/ret.sum(axis=1)[:,None]
return ret
def loadBrownSymsSeq(self, T):
taggedWordsIter = brown.tagged_words()
retIdx = 0
iterIdx = 0
symSequence = []
for wrd, tag in taggedWordsIter:
if retIdx >= T:
break
if tag in WHITE_LIST:
val = WHITE_LIST.index(tag)
symSequence.append(val)
retIdx += 1
iterIdx += 1
self._syms = symSequence
return symSequence
def textSeqToSymSeq(self, txtSeqArr):
tags = nltk.pos_tag(txtSeqArr) #PerceptronTagger
tags = [t[1] for t in tags]
tags = [WHITE_LIST.index(t) for t in tags if t in WHITE_LIST]
return tags
def initHMM(self):
# self._hmm = MultinomialHMM(n_components=self._N, startprob_prior=None, transmat_prior=None,
# algorithm='viterbi', random_state=None, n_iter=self._maxIters, tol=0.01,
# verbose=True, params='ste', init_params='ste')
self._hmm = MultinomialHMM(n_components=self._N, n_iter=self._maxIters,
verbose=True, params='ste', init_params='ste')
# self._hmm.emissionprob_ = self._B
# n_features (int) Number of possible symbols emitted by the model (in the samples).
# monitor_ (ConvergenceMonitor) Monitor object used to check the convergence of EM.
# transmat_ (array, shape (n_components, n_components)) Matrix of transition probabilities between states.
# startprob_ (array, shape (n_components, )) Initial state occupation distribution.
# emissionprob_ (array, shape (n_components, n_features)) Probability of emitting a given symbol when in each state.
def trainHMM(self):
self._hmm.fit(np.array(self._syms).reshape(-1, 1))
def testTxt(self, txtSeqArr):
testSymsArr = self.textSeqToSymSeq(txtSeqArr)
score = self._hmm.score(testSymsArr)
return score
def testSyms(self, symsArr):
# testSymsArr = self.textSeqToSymSeq(txtSeqArr)
score = self._hmm.score(symsArr)
return score
def persistHMM(self, filename):
import pickle
s = pickle.dumps(self)
with open(filename, 'w') as f:
f.write(s)
@staticmethod
def loadHMM(filename):
import pickle
with open(filename, 'r') as f:
s = f.read()
model = pickle.loads(s)
return model
def pickRandomSeq(self, length = 100):
symSequence = []
taggedWordsIter = brown.tagged_words()
maxIdx = len(taggedWordsIter)
import random
idx = 0
while idx < length:
wrdIdx = random.randint(0, maxIdx)
(wrd, tag) = taggedWordsIter[wrdIdx]
if tag in WHITE_LIST:
val = WHITE_LIST.index(tag)
symSequence.append(val)
idx += 1
return symSequence
def pickOrderedSeq(self, length = 100):
symSequence = []
taggedWordsIter = brown.tagged_words()
maxIdx = len(taggedWordsIter)
maxMinIdx = maxIdx - length
import random
minIdx = random.randint(0, maxMinIdx)
count = 0
for idx in range(minIdx, maxIdx):
(wrd, tag) = taggedWordsIter[idx]
if tag in WHITE_LIST:
val = WHITE_LIST.index(tag)
symSequence.append(val)
idx += 1
count += 1
if count >= length:
break
return symSequence
def printHMM(self):
print("A = %s" % str(self._hmm.transmat_))
print("B = %s" % str(self._hmm.emissionprob_))
print("PI = %s" % str(self._hmm.startprob_))
print("Verify A = %s" % np.sum(self._hmm.transmat_, axis=1))
print("Verify B = %s" % np.sum(self._hmm.emissionprob_, axis=1))
print("Verify PI = %s" % np.sum(self._hmm.startprob_, axis=0))
def histo(self):
retHisto = dict((x, self._syms.count(x)) for x in range(len(WHITE_LIST)))
retHisto = dict((WHITE_LIST[k], val) for k, val in retHisto.items())
return retHisto
