for j, i in enumerate(ids):
with h5py.File(
os.path.join(DATA_DIR, 'nest_output', i,
'LFP_firing_rate.h5')) as f:
lfp = f['data'][:, 150:]
lfps[j] = lfp
lfps = np.array(lfps)
labels = np.array(labels)
## Get PSDs
fs, psds = calc_psds(lfps)
del lfps
## Remove cases of simulations where no neurons spiked
psds, labels = remove_zero_lfps(psds, labels)
test_psds = psds[:10000]
test_labels = labels[:10000]
training_psds = psds[10000:]
training_labels = labels[10000:]
## Hyperparams
batch_size = 100
epochs = 400
lr = 1e-3
x_train = training_psds
y_train = training_labels.copy()
x_test = test_psds
y_test = test_labels.copy()
for j, i in enumerate(ids[:10000]):
with h5py.File(
os.path.join(DATA_DIR_LARGE, 'nest_output', i,
'LFP_firing_rate.h5')) as f:
lfp = f['data'][:, 150:]
lfps[j] = lfp
lfps = np.array(lfps)
labels_large = np.array(labels_large)
## Get PSDs
fs, psds_large = calc_psds(lfps)
del lfps
## Remove cases of simulations where no neurons spiked
psds_large, labels_large = remove_zero_lfps(psds_large, labels_large)
psds_large = np.swapaxes(psds_large, 1, 2)
## Rescale labels
labels_large_rescaled = labels_large - np.array([[0.8, 3.5, 0.05]])
labels_large_rescaled /= np.array([[3.2, 4.5, 0.35]])
## CNN model
def set_up_model(x_train, lr, n_dense=128, output=3):
keras.backend.clear_session()
input_shape = (x_train.shape[1], x_train.shape[2])
inputs = keras.layers.Input(shape=input_shape)
x = keras.layers.Conv1D(20,
kernel_size=(12),
for j, i in enumerate(ids[:10000]):
with h5py.File(
os.path.join(DATA_DIR_LARGE, 'nest_output', i,
'LFP_firing_rate.h5')) as f:
lfp = f['data'][:, 150:]
lfps[j] = lfp
lfps = np.array(lfps)
labels_large = np.array(labels_large)
## Get PSDs
fs, psds_large = calc_psds(lfps)
del lfps
## Remove cases of simulations where no neurons spiked
psds_large, labels_large = remove_zero_lfps(psds_large, labels_large)
psds_large = np.swapaxes(psds_large, 1, 2)
#### Load small parameterspace LFPs
ids, labels_small = read_ids_parameters(
os.path.join(DATA_DIR_SMALL, 'id_parameters.txt'))
## Test sims
ids = ids[:10000]
labels_small = labels_small[:10000]
lfps = np.zeros((len(ids), 6, 2851), dtype=np.float32)
for j, i in enumerate(ids[:10000]):
with h5py.File(
os.path.join(DATA_DIR_SMALL, 'nest_output', i,
print('loading grid sampled data')
ids, grid_labels = read_ids_parameters(os.path.join(DATA_DIR_GRID, 'id_parameters.txt'))
lfps = np.zeros((len(ids), 6, 2851), dtype=np.float32)
for j, i in enumerate(ids):
with h5py.File(os.path.join(DATA_DIR_GRID, 'nest_output', i, 'LFP_firing_rate.h5')) as f:
lfp = f['data'][:,150:]
lfps[j] = lfp
grid_labels = np.array(grid_labels)
## Get PSDs
fs, grid_psds = calc_psds(lfps)
del lfps
## Remove cases of simulations where no neurons spiked
psds, labels = remove_zero_lfps(grid_psds, grid_labels)
### RANDOMLY SAMPLED DATA
# Load LFPs
print('loading randomly sampled data')
ids, random_labels = read_ids_parameters(os.path.join(DATA_DIR_RANDOM, 'id_parameters.txt'))
lfps = np.zeros((len(ids), 6, 2851), dtype=np.float32)
for j, i in enumerate(ids):
print('%d'%j, end='\r')
with h5py.File(os.path.join(DATA_DIR_RANDOM, 'nest_output', i, 'LFP_firing_rate.h5')) as f:
lfp = f['data'][:,150:]
lfps[j] = lfp
print('converting')
random_labels = np.array(random_labels)