def eval_state_data(
data_dir,
num_examples=128,
batch_size=32,
tfrecords_file_format="codraw_%s_combined_state_glove.tfrecords",
split='train'):
model = state_models.SconesGPT2StateModel(num_units=64)
input_dataset = model.get_input_fn(os.path.join(data_dir),
batch_size,
tfrecords_file_format,
split=split,
shuffle=False)()
dataset_iter = iter(input_dataset)
with open(os.path.join(data_dir, 'CoDraw_1_0.json')) as f:
data_json = json.load(f)
cur_id = -1
seqs = []
gts = []
cur_seq = []
read_examples = 0
for _ in range(num_examples):
try:
examples = next(dataset_iter)[0]
except StopIteration:
break
data = examples['example_id'], examples['combined_len'], examples[
'current_scene_len'], examples['combined_vecs'], examples[
'seq_d'], examples['seq_t']
for i in range(len(data[0])):
if cur_id != int(data[0][i]):
prev_id = cur_id
cur_id = int(data[0][i])
if cur_seq:
gts.append(load_gt(prev_id, data_json, split))
seqs.append(cur_seq)
cur_seq = []
cur_step = {k: v[i].numpy() for k, v in examples.items()}
cur_seq.append(cur_step)
prev_id = cur_id
gts.append(load_gt(prev_id, data_json, split))
seqs.append(cur_seq)
return seqs, gts
f.close()
finally:
f_train.close()
f_test.close()
f_scores.close()
# Save time elapsed
f = open(model_path + model_name + "_time_elapsed.txt", 'w')
f.write(str(time_elapsed))
f.close()
if TEST:
t0 = time.time()
print 'Predicting...','\n'
real_labels_mean = load_gt(test_gt_list)
real_labels_frames = y_test
results = np.zeros((X_test.shape[0], tags.shape[0]))
predicted_labels_mean = np.zeros((num_frames_test.shape[0], 1))
predicted_labels_frames = np.zeros((y_test.shape[0], 1))
song_paths = open(test_songs_list, 'r').read().splitlines()
previous_numFrames = 0
n=0
for i in range(0, num_frames_test.shape[0]):
print song_paths[i]
num_frames=num_frames_test[i]
files = [file for file in files if os.path.isfile(os.path.join(folder, file))]
for file in files:
shutil.copyfile(os.path.join(folder, file), os.path.join(volpy_folder, file))
#%%
for name in names:
folder = os.path.join(ROOT_FOLDER, name)
s_folder = os.path.join(ROOT_FOLDER, name, 'sgpmd')
file = f'{name}.hdf5'
m = cm.load(os.path.join(folder, file))
m.save(os.path.join(s_folder, name+'.tif'))
#%%
for name in names:
folder = os.path.join(ROOT_FOLDER, name)
spatial, temporal, spikes = load_gt(folder)
#ROIs = spatial.transpose([1,2,0])
ROIs = spatial.copy()
volpy_folder = os.path.join(folder, 'volpy')
np.save(os.path.join(volpy_folder, 'ROIs_gt'), ROIs)
#%% volpy params
#for ridge_bg in [0.5, 0.1, 0.01, 0.001, 0]:
context_size = 35 # number of pixels surrounding the ROI to censor from the background PCA
flip_signal = True # Important!! Flip signal or not, True for Voltron indicator, False for others
hp_freq_pb = 1 / 3 # parameter for high-pass filter to remove photobleaching
threshold_method = 'simple' # 'simple' or 'adaptive_threshold'
min_spikes= 30 # minimal spikes to be found