def train(self, train_data, test_data):
sequences = train_data + test_data
# number of symbols excluding the stop symbol
num_symbols = count_unique_symbols(sequences)
longest_sequence_len = longest_sequence_length(sequences)
# keeps track of occurences of symbols at different positions: [position][symbol]
self.table = [[0]*(num_symbols) for x in range(longest_sequence_len)]
# increment counters for all symbols at all position, using all sequences in the test data
for x in range(len(sequences)):
for y in range(len(sequences[x])):
self.table[y][sequences[x][y]] += 1
# increment the stop symbol count
#normalize
for x in range(len(self.table)):
sumval = sum(self.table[x])
for y in range(len(self.table[x])):
if sumval != 0:
self.table[x][y] = Decimal(self.table[x][y]) / Decimal(sumval)
else:
self.table[x][y] = Decimal(1) / Decimal(num_symbols)
def train(self, train_data, test_data):
sequences = train_data + test_data
self.num_symbols = count_unique_symbols(sequences)
ll_bound = 10.0
self.model = self.randommodel(self.num_states, self.num_symbols)
prev = -1.0
ll = -1.0
while prev == -1.0 or ll - prev > ll_bound:
prev = ll
self.model = self.iterateEM(self.model, sequences)
probs = self.computeprobabilities(self.model, sequences)
ll = self.loglikelihood(probs)
def train(self, train_data, test_data):
self.train_data = train_data
self.num_symbols = utilities.count_unique_symbols(train_data)
self.i = 0