def test_bs_PCA_fit():
img_path_placeholder, ending_points, SESS \
= _get_cate_seed0_ending_points_sess(
64,
'/mnt/fs4/chengxuz/brainscore_model_caches/cate_aug/res18/exp_seed0_bn_wd/checkpoint-505505')
identifier = 'test-cate-bn-wd-pca'
activations_model = TensorflowSlimWrapper(
identifier=identifier, labels_offset=0,
endpoints=ending_points, inputs=img_path_placeholder,
session=SESS)
LayerPCA.hook(activations_model, n_components=1000)
#benchmark = bench.load('dicarlo.Majaj2015.IT-pls')
benchmark = bench.load('dicarlo.Majaj2015.V4-pls')
all_layers = ['encode_%i' % i for i in range(1, 10)]
model_scores = LayerScores(
identifier,
activations_model=activations_model,
)
score = model_scores(
benchmark=benchmark,
layers=all_layers,
)
print(score)
def test_bs_fit():
img_path_placeholder, ending_points, SESS \
= _get_cate_seed0_ending_points_sess(64)
#identifier = 'test-la-seed0'
#identifier = 'test-la-seed0-2'
#identifier = 'test-cate-seed1'
#identifier = 'test-cate-seed0-mask-2'
#identifier = 'test-cate-seed0-mask-midvar'
#identifier = 'test-cate-seed0-mask-midvar-smwd'
#identifier = 'test-cate-seed0-mask-midvar-params'
#identifier = 'test-cate-seed0-mask-midvar-faster'
#identifier = 'test-cate-seed0-mask-midvar-faster-debug'
#identifier = 'test-cate-seed0-mask-midvar-faster-debug-2'
#identifier = 'test-cate-seed0-mask-final-2'
#identifier = 'test-cate-seed0-mask-final-e12'
#identifier = 'test-cate-seed0-mask-final-e1-nolap'
#identifier = 'test-cate-seed0-mask-final-e1-nolap-3'
identifier = 'test-cate-seed0-mask-final-all-test-2'
activations_model = TensorflowSlimWrapper(
identifier=identifier, labels_offset=0,
endpoints=ending_points, inputs=img_path_placeholder,
session=SESS)
#all_params = []
#for each_ls in [0.1, 0.01]:
# for each_ld in [0.1, 0.01]:
# all_params.append(json.dumps([[], {'ls': each_ls, 'ld': each_ld}]))
#all_params = [
# json.dumps([[], {'ls': 0.1, 'ld': 0.1}]),
# json.dumps([[], {'ls': 0.01, 'ld': 0.01}])]
all_params = [
json.dumps([[], {
'with_corr_loss': True,
'with_lap_loss': False,
'adam_eps': 0.1,
'max_epochs': 50,
'decay_rate': 35,
}]),
#json.dumps([[], {'max_epochs': 50, 'decay_rate': 35, 'ls': 0.01, 'ld': 0.01}]),
#json.dumps([[], {'max_epochs': 60, 'decay_rate': 45}]),
]
layer_and_params = []
#all_layers = ['encode_%i' % i for i in range(1, 10)]
all_layers = ['encode_1']
for layer in all_layers:
_model_scores = ParamScores(
identifier,
activations_model=activations_model)
_score = _model_scores(
#benchmark_builder=DicarloMajaj2015ITLowMidVarMask,
benchmark_builder=DicarloMajaj2015ITMaskParams,
params=all_params,
layer=layer,
)
print(_score)
def _build_activations_model(self, batch_size):
self.activations_model = TensorflowSlimWrapper(
identifier=self.identifier,
labels_offset=0,
endpoints=self.ending_points,
inputs=self.img_path_placeholder,
session=self.SESS,
batch_size=batch_size)
return self.activations_model
def init(identifier, preprocessing_type, image_size, net_name=None, labels_offset=1, batch_size=64,
model_ctr_kwargs=None):
import tensorflow as tf
from nets import nets_factory
tf.compat.v1.reset_default_graph()
placeholder = tf.compat.v1.placeholder(dtype=tf.string, shape=[batch_size])
preprocess = TFSlimModel._init_preprocessing(placeholder, preprocessing_type, image_size=image_size)
net_name = net_name or identifier
model_ctr = nets_factory.get_network_fn(net_name, num_classes=labels_offset + 1000, is_training=False)
logits, endpoints = model_ctr(preprocess, **(model_ctr_kwargs or {}))
if 'Logits' in endpoints: # unify capitalization
endpoints['logits'] = endpoints['Logits']
del endpoints['Logits']
session = tf.compat.v1.Session()
TFSlimModel._restore_imagenet_weights(identifier, session)
wrapper = TensorflowSlimWrapper(identifier=identifier, endpoints=endpoints, inputs=placeholder, session=session,
batch_size=batch_size, labels_offset=labels_offset)
wrapper.image_size = image_size
return wrapper