def train_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(30, 30, len(word2idx))
rnn.fit(sentences,
learning_rate=1e-4,
show_fig=True,
activation=T.nnet.relu,
epochs=2000)
rnn.save('RNN_D30_M30_epochs2000_relu.npz')
def train_poetry():
# students: tanh didn't work but you should try it
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(50, 50, len(word2idx))
rnn.fit(sentences,
learning_rate=10e-5,
show_fig=True,
activation=T.nnet.relu,
epochs=2000)
rnn.save('RRNN_D50_M50_epochs2000_relu.npz')
def train_poetry():
sentences, word2idx = get_robert_frost()
# embedding size = 50, and hidden neurons = 50
rnn = SimpleRNN(50, 50, len(word2idx))
rnn.fit(sentences,
learning_rate=10e-5,
show_fig=False,
activation=T.nnet.relu,
epochs=200)
rnn.save('RRNN_D50_M50_epochs200_relu.npz')
def train_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(50, 50, len(word2idx))
rnn.fit(sentences,
leraning_rate=1e-4,
show_fig=True,
activation=T.nnet.relu,
epochs=2000)
rnn.save('./rnn_class/RRNN_D50_M50_epochs2000_relu.npz')
def train_poetry(epochs=200, learning_rate=10e-5, mu=0.9, show_fig=True):
# students: tanh didn't work but you should try it
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(30, 30, len(word2idx))
rnn.fit(sentences,
learning_rate=learning_rate,
mu=mu,
show_fig=show_fig,
activation=T.nnet.relu,
epochs=epochs)
rnn.save('RNN_D30_M30_epochs{}_relu.npz'.format(epochs))
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load("RRNN_D30_M30_epochs1000_relu.npz", activation=T.nnet.relu)
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi/=pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu.npz', T.nnet.relu)
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu.npz', T.nnet.relu)
# determine initial state distribution for starting sentences
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def train_poetry():
sentences, word2idx = get_robert_frost()
D = 30
M = 50
epochs = 500
srn = SRN(D, M, len(word2idx))
srn.fit(sentences,
learning_rate=10e-5,
show_fig=True,
activation=T.nnet.relu,
epochs=epochs)
srn.save('RNN_D%d_M%d_epochs%d_relu.npz' % (D, M, epochs))
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs200_relu.npz', T.nnet.relu)
# Create the initial word distribution.
V = len(word2idx)
pi = np.zeros(V) # creates a vector of length V.
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu.npz', T.nnet.relu)
# determine initial state distribution for starting sentences
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry(session, savefile):
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load(savefile, tf.nn.relu, session)
# determine initial state distribution for starting sentences
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
srn = SRN.load('RNN_D30_M30_epochs300_relu.npz', T.nnet.relu)
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
srn.generate(pi, word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs200_relu.npz', T.nnet.relu)
V = len(word2idx)
pi = np.zeros(V)
# create the word distribution
for sentence in sentences:
# get first word's frequency count
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_potery():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu_my.npz',
T.nnet.relu) # sprawdzic czy dziala tez bez
#rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu_my', T.nnet.relu)
V = len(word2idx)
pi = np.zeros(V)
for sentence in sentences:
pi[sentence[0]] += 1
pi /= pi.sum()
rnn.generate(pi, word2idx)
def generate_poetry_batches(epochs=500):
"""generate_poetry_batches"""
sentences, word2idx = get_robert_frost()
idx2word = {v:k for k, v in word2idx.items()}
# total number of sentences
n_sentences = len(sentences)
# total number of words in corpus
n_total = sum((len(sentence) + 1) for sentence in sentences)
for i in range(epochs):
for j in range(n_sentences):
if np.random.random() < 0.1:
print("generating an END to START")
input_sequence = [0] + sentences[j]
# [1] is the end token
output_sequence = sentences[j] + [1]
else:
# input sequence is from the start to 2nd last word of X
# so that the last word can be predicted
input_sequence = [0] + sentences[j][:-1]
output_sequence = sentences[j]
input_seq_sentence = ''
output_seq_sentence = ''
for word_idx in input_sequence:
input_seq_sentence += idx2word[word_idx] + " "
for word_idx in output_sequence:
output_seq_sentence += idx2word[word_idx] + " "
print("input_seq_sentence: ", input_seq_sentence)
print("output_seq_sentence: ", output_seq_sentence)
print("n_total: ", n_total)
# has to be calculated manually
n_total += len(output_sequence)
keypressed = input('Press q to quit: ')
if keypressed == 'q':
break
keypressed = input('Press q to quit: ')
if keypressed == 'q':
break
def train_poetry(session, dims, savefile):
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(dims, dims, len(word2idx), tf.nn.relu, session)
rnn.fit(sentences, epochs=17, show_fig=True)
rnn.save(savefile)
def generate_potery():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load(
'RRNN_D50_M50_epochs2000_relu_my.npz') # sprawdzic czy dziala tez bez
#rnn = SimpleRNN.load('RNN_D30_M30_epochs2000_relu_my', T.nnet.relu)
rnn.generate(word2idx)
def train_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(30, 30, len(word2idx))
rnn.fit(sentences, learning_rate=1e-4, show_fig=True, activation=T.nnet.relu, epochs=2000)
rnn.save('RNN_D30_M30_epochs2000_relu.npz')
def train_poetry():
# students: tanh didn't work but you should try it
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(50, 50, len(word2idx))
def train_poetry():
# students: tanh didn't work but you should try it
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN(30, 30, len(word2idx))
rnn.fit(sentences, learning_rate=10e-5, show_fig=True, activation=T.nnet.relu, epochs=2000)
rnn.save('RNN_D30_M30_epochs2000_relu.npz')
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RRNN_D50_M50_epochs2000_relu.npz', T.nnet.relu)
rnn.generate(word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load("RNN_D30_M30_epochs400_relu.npz",
activation=T.nnet.relu)
rnn.generate(word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rnn = SimpleRNN.load('RRNN_D50_M50_epochs2000_relu.npz', T.nnet.relu)
rnn.generate(word2idx)
def generate_poetry():
sentences, word2idx = get_robert_frost()
rrnn = RRNN.load('RRNN_D30_M30_epochs500_relu.npz', T.nnet.relu)
rrnn.generate(word2idx)
