import json
import numpy as np
import os.path as path
import tensorflow as tf
from python.dataset import AutoComplete, AutoCompleteFixed
from python.model import PureNLSTM
dirname = path.dirname(path.realpath(__file__))
article_dir = path.join(dirname, '..', '..', 'public')
tf.logging.set_verbosity(tf.logging.INFO)
train_dataset = AutoComplete(repeat=True)
test_dataset = AutoCompleteFixed(
"parts of north africa",
batch_size=1,
)
model = PureNLSTM(train_dataset,
name='autocomplete_nlstm_600',
embedding_size=600,
verbose=True)
for output_i, output in enumerate(model.predict(dataset=test_dataset)):
probabilities = output['probabilities']
predict_sorted = np.argsort(probabilities, axis=1)[:, ::-1]
source = test_dataset.decode_source(output['source'])
target = test_dataset.decode_target(output['target'])
predict = test_dataset.decode_target(predict_sorted)
import numpy as np
import tensorflow as tf
from tqdm import tqdm
from python.dataset import AutoComplete
from python.model import PureGRU
from python.summary import ContextAccuracy
tf.logging.set_verbosity(tf.logging.INFO)
train_dataset = AutoComplete(repeat=True, batch_size=64)
test_dataset = AutoComplete(dataset='test', repeat=False, batch_size=16)
model = PureGRU(train_dataset,
name='autocomplete_gru_600',
embedding_size=600,
verbose=True)
context_accuracy = ContextAccuracy()
for output_i, output in enumerate(
tqdm(model.predict(dataset=test_dataset),
total=test_dataset.observations)):
probabilities = output['probabilities']
predict_sorted = np.argsort(probabilities, axis=1)[:, ::-1]
source = test_dataset.decode_source(output['source'])
target = test_dataset.decode_target(output['target'])
predict = test_dataset.decode_target(predict_sorted)
context_accuracy.add(source, predict, target)
import tensorflow as tf
from python.dataset import AutoComplete
from python.model import PureLSTM
tf.random.set_random_seed(450849059) # From random.org
tf.logging.set_verbosity(tf.logging.INFO)
train_dataset = AutoComplete(repeat=True, batch_size=64)
valid_dataset = AutoComplete(repeat=True, dataset='valid', batch_size=64)
test_dataset = AutoComplete(repeat=False, dataset='test', batch_size=64)
model = PureLSTM(train_dataset,
name='autocomplete_lstm_600',
embedding_size=600,
verbose=True)
model.train(max_steps=train_dataset.batches * 2, valid_dataset=valid_dataset)
import numpy as np
import tensorflow as tf
from python.dataset import AutoComplete
from python.model import PureLSTM
tf.logging.set_verbosity(tf.logging.INFO)
train_dataset = AutoComplete(repeat=True, batch_size=64)
test_dataset = AutoComplete(dataset='test', repeat=False, batch_size=1)
model = PureLSTM(train_dataset, name='autocomplete_lstm_600',
embedding_size=600,
verbose=True)
print(model.evaluate(dataset=test_dataset))
import numpy as np
import tensorflow as tf
from python.dataset import AutoComplete
from python.model import PassThrough
tf.logging.set_verbosity(tf.logging.INFO)
train_dataset = AutoComplete(repeat=True)
valid_dataset = AutoComplete(dataset='test', repeat=False, batch_size=1)
model = PassThrough(train_dataset,
name='autocomplete_passthough',
verbose=True)
model.train(max_steps=1)
for output_i, output in enumerate(model.predict(dataset=valid_dataset)):
length = output['length']
source = valid_dataset.decode_source(output['source'])
target = valid_dataset.decode_target(output['target'])
print(f'sequence {output_i} of length {length}')
for char, word in zip(source, target):
print(f' {char} -> {word}')
break