import nn_extra_student from config_rnn import defaults batch_size = 32 sample_batch_size = 1 n_samples = 4 rng = np.random.RandomState(42) rng_test = np.random.RandomState(317070) seq_len = defaults.seq_len nonlinearity = tf.nn.elu weight_norm = True train_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='train', rng=rng, dataset='fashion_mnist') test_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='test', dataset='fashion_mnist', rng=rng_test) obs_shape = train_data_iter.get_observation_size() # (seq_len, 28,28,1) print('obs shape', obs_shape) ndim = np.prod(obs_shape[1:]) corr_init = np.ones((ndim, ), dtype='float32') * 0.1 nu_init = 1000
batch_size = 16 sample_batch_size = 1 n_samples = 4 rng = np.random.RandomState(42) test_rng = np.random.RandomState(317070) seq_len = defaults.seq_len eps_corr = defaults.eps_corr mask_dims = defaults.mask_dims nonlinearity = tf.nn.elu weight_norm = True train_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, dataset='cifar10', set='train', rng=rng) test_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, dataset='cifar10', set='test', rng=test_rng) obs_shape = train_data_iter.get_observation_size() # (seq_len, 28,28,1) print('obs shape', obs_shape) ndim = np.prod(obs_shape[1:]) corr_init = np.ones((ndim, ), dtype='float32') * 0.1 optimizer = 'rmsprop' learning_rate = 0.001
from config_rnn import defaults batch_size = 64 sample_batch_size = 1 n_samples = 4 rng = np.random.RandomState(42) rng_test = np.random.RandomState(317070) seq_len = defaults.seq_len eps_corr = defaults.eps_corr mask_dims = defaults.mask_dims nonlinearity = tf.nn.elu weight_norm = True train_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='train', rng=rng) test_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='test', rng=rng_test) valid_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='test', rng=rng) test_data_iter2 = data_iter.BaseTestBatchSeqDataIterator(seq_len=seq_len, set='test', rng=rng)
import nn_extra_student from config_rnn import defaults batch_size = 32 sample_batch_size = 1 n_samples = 4 rng = np.random.RandomState(42) rng_test = np.random.RandomState(317070) seq_len = defaults.seq_len eps_corr = defaults.eps_corr mask_dims = defaults.mask_dims nonlinearity = tf.nn.elu weight_norm = True train_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='train', rng=rng, digits=[0, 2, 4, 6, 8]) test_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='test', digits=[1, 3, 5, 7, 9], rng=rng_test) valid_data_iter = data_iter.BaseExchSeqDataIterator(seq_len=seq_len, batch_size=batch_size, set='test', rng=rng_test, digits=[0, 2, 4, 6, 8]) test_data_iter2 = data_iter.BaseTestBatchSeqDataIterator(seq_len=seq_len, set='test', rng=rng_test, digits=[1, 3, 5, 7, 9]) obs_shape = train_data_iter.get_observation_size() # (seq_len, 28,28,1) print('obs shape', obs_shape) ndim = np.prod(obs_shape[1:])