states_t_bxn, outputs_t_bxn = sess.run([states_t, outputs_t],
feed_dict=feed_dict)
states_nxt = np.transpose(np.squeeze(np.asarray(states_t_bxn)))
outputs_t_bxn = np.squeeze(np.asarray(outputs_t_bxn))
r_sxt = np.dot(P_nxn, states_nxt)
for s in xrange(nspikifications):
data_e.append(r_sxt)
u_e.append(u_1xt)
outs_e.append(outputs_t_bxn)
truth_data_e = normalize_rates(data_e, E, N)
spiking_data_e = spikify_data(truth_data_e, rng, dt=FLAGS.dt,
max_firing_rate=FLAGS.max_firing_rate)
train_inds, valid_inds = get_train_n_valid_inds(E, train_percentage,
nspikifications)
data_train_truth, data_valid_truth = split_list_by_inds(truth_data_e,
train_inds,
valid_inds)
data_train_spiking, data_valid_spiking = split_list_by_inds(spiking_data_e,
train_inds,
valid_inds)
data_train_truth = nparray_and_transpose(data_train_truth)
data_valid_truth = nparray_and_transpose(data_valid_truth)
data_train_spiking = nparray_and_transpose(data_train_spiking)
data_valid_spiking = nparray_and_transpose(data_valid_spiking)
# save down the inputs used to generate this data
train_inputs_u, valid_inputs_u = split_list_by_inds(u_e,
input_times = None
rates, x0s, inputs = \
generate_data(rnn, T=T, E=E, x0s=x0s, P_sxn=P_sxn,
input_magnitude=input_magnitude,
input_times=input_times)
if FLAGS.noise_type == "poisson":
noisy_data = spikify_data(rates, rng, rnn['dt'], rnn['max_firing_rate'])
elif FLAGS.noise_type == "gaussian":
noisy_data = gaussify_data(rates, rng, rnn['dt'], rnn['max_firing_rate'])
else:
raise ValueError("Only noise types supported are poisson or gaussian")
# split into train and validation sets
train_inds, valid_inds = get_train_n_valid_inds(E, train_percentage,
nreplications)
# Split the data, inputs, labels and times into train vs. validation.
rates_train, rates_valid = \
split_list_by_inds(rates, train_inds, valid_inds)
noisy_data_train, noisy_data_valid = \
split_list_by_inds(noisy_data, train_inds, valid_inds)
input_train, inputs_valid = \
split_list_by_inds(inputs, train_inds, valid_inds)
condition_labels_train, condition_labels_valid = \
split_list_by_inds(condition_labels, train_inds, valid_inds)
input_times_train, input_times_valid = \
split_list_by_inds(input_times, train_inds, valid_inds)
# Turn rates, noisy_data, and input into numpy arrays.
rates_train = nparray_and_transpose(rates_train)
input_magnitude=0.0, input_times=None)
spikes_b = spikify_data(rates_b, rng, rnn_b['dt'], rnn_b['max_firing_rate'])
# not the best way to do this but E is small enough
rates = []
spikes = []
for trial in range(E):
if rnn_to_use[trial] == 0:
rates.append(rates_a[trial])
spikes.append(spikes_a[trial])
else:
rates.append(rates_b[trial])
spikes.append(spikes_b[trial])
# split into train and validation sets
train_inds, valid_inds = get_train_n_valid_inds(E, train_percentage,
nspikifications)
rates_train, rates_valid = split_list_by_inds(rates, train_inds, valid_inds)
spikes_train, spikes_valid = split_list_by_inds(spikes, train_inds, valid_inds)
condition_labels_train, condition_labels_valid = split_list_by_inds(
condition_labels, train_inds, valid_inds)
ext_input_train, ext_input_valid = split_list_by_inds(
ext_input, train_inds, valid_inds)
rates_train = nparray_and_transpose(rates_train)
rates_valid = nparray_and_transpose(rates_valid)
spikes_train = nparray_and_transpose(spikes_train)
spikes_valid = nparray_and_transpose(spikes_valid)
# add train_ext_input and valid_ext input
data = {'train_truth': rates_train,
