def load_model(load_path: dict, device: torch.device) -> FullModel:
"""
Instantiate the model from saved dictionary of pretrained model.
Args:
num_graph_layers: Number of layers to use in the model.
num_classes: Number of classes in the dataset.
pretrain_load_path: Use pretrained weights.
Returns:
The instantiated model with the requested parameters.
"""
ckpt = torch.load(f=load_path, map_location=device)
model_params_dict = ckpt["model_params_dict"]
model_params_dict['graph_layers_sizes'] = model_params_dict[
'graph_layers_sizes'][1:]
model = FullModel(
model_type=model_params_dict['model_type'],
num_classes=model_params_dict['num_classes'],
num_node_features=model_params_dict['num_node_features'],
graph_layers_sizes=model_params_dict['graph_layers_sizes'],
text_embed_dim=model_params_dict['text_embed_dim'],
text_output_dim=model_params_dict['text_output_dim'],
linear_layers_sizes=model_params_dict['linear_layers_sizes'],
dropout_rate=model_params_dict['dropout_rate'],
vocab_size=model_params_dict['vocab_size'])
model.load_state_dict(state_dict=ckpt["model_state_dict"])
# transfer model to cpu or gpu
model = model.to(device=device)
return model, model_params_dict
def test_ipv6(self):
model = FullModel()
model.ipv6 = '10::'
assert ipaddress.ip_address(u'10::') == model.ipv6
with pytest.raises(ValidationError):
model.ipv6 = '10.0.0.0'
def test_ipv4(self):
model = FullModel()
model.ipv4 = '10.0.0.0'
assert ipaddress.ip_address(u'10.0.0.0') == model.ipv4
with pytest.raises(ValidationError):
model.ipv4 = '1::1'
def test_init(self):
model = FullModel()
new0 = URLField()
new1 = URLField()
assert 1 == len(new0.default_validators)
assert 1 == len(new1.default_validators)
new0 = ASNField()
new1 = ASNField()
assert 0 == len(new0.default_validators)
assert 0 == len(new1.default_validators)
new0 = IPAddressField()
new1 = IPAddressField()
assert_ip_validator(new0)
assert_ip_validator(new1)
new0 = IPPrefixField()
new1 = IPPrefixField()
assert_ip_validator(new0)
assert_ip_validator(new1)
new0 = MacAddressField()
new1 = MacAddressField()
assert 1 == len(new0.default_validators)
assert 1 == len(new1.default_validators)
def test_asn(self):
model = FullModel()
model.asn = 42
assert 42 == model.asn
model.full_clean()
with pytest.raises(ValidationError):
model.asn = 'invalid'
model.full_clean()
def test_mac(self):
model = FullModel()
model.mac = 'Ff:00:00:12:34:56'
model.full_clean()
assert 'Ff:00:00:12:34:56' == model.mac
with pytest.raises(ValidationError):
model.mac = 'invalid'
model.full_clean()
def section_8_1():
env = DynaMaze(FIG_8_2_INIT_POS, FIG_8_2_GOAL_POS_L, FIG_8_2_GRID_SHAPE, FIG_8_2_WALL)
alg = TabularQ(FullModel(env), SEC_8_1_ALP, DYNA_MAZE_GAMMA)
alg.seed(0)
alg.rand_sam_one_step_pla(SEC_8_1_N_STEPS, decay=True)
V = alg.get_V()
plt.title('Section 8.1 - tabular Q (1-step random sample, dyna maze)')
sns.heatmap(to_arr(V), cbar_kws={'label': 'max(Q(s, a))'})
save_plot('section8.1')
plt.show()
def load_model_and_optimizer(opt, num_GPU=None, reload_model=True):
model = FullModel.FullVisionModel(opt)
optimizer = []
for idx, layer in enumerate(model.encoder):
optimizer.append(torch.optim.Adam(layer.parameters(), lr=opt.learning_rate))
model, num_GPU = model_utils.distribute_over_GPUs(opt, model, num_GPU=num_GPU)
model, optimizer = model_utils.reload_weights(
opt, model, optimizer, reload_model=reload_model
)
return model, optimizer
if not args.time_range_points == 100:
points_name += '_tp{}'.format(args.time_range_points)
if not args.points == 500:
points_name += '_p{}'.format(args.points)
points_file_name = '_n{}'.format(args.points)
amp_dir = '/data/lhcb/users/bjoern/Analyses/B2DPi/data/various_studies/ks_cpv/amplitudes'
output_dir = '/data/lhcb/users/bjoern/Analyses/B2DPi/data/various_studies/ks_cpv/simple_full_studies'
output_dir = '{}/setup_{}_{}{}_n_{}'.format(output_dir, args.model, args.setup,
points_name, args.n_kl)
subprocess.call(['mkdir', '-p', output_dir])
# A default model
fm_default = FullModel(
"{}/{}_KS_default{}.pickle".format(amp_dir, args.model, points_file_name),
"{}/{}_KL_default{}.pickle".format(amp_dir, args.model, points_file_name),
eps=PhysicalParameters().default_eps,
parameters=PhysicalParameters(),
binning_file="../python/input/KsPiPi_optimal.pickle")
# extract the phase space coordinates from default model
s12 = fm_default.s12
s13 = fm_default.s13
quantile_suffixes = ['']
if args.quantiles == 5:
quantile_suffixes = [
'_q{:1.3f}'.format(q) for q in np.linspace(0.1, 0.9, 5)
]
if args.quantiles == 9:
quantile_suffixes = [
'_q{:1.3f}'.format(q) for q in np.linspace(0.1, 0.9, 9)
### Create model
CRNN_model = CRNN(IN_SIZE, HIDDEN_SIZE, KERNEL_SIZE, STRIDE,
GRU_NUM_LAYERS)
attn_layer = AttnMech(HIDDEN_SIZE * NUM_DIRS)
apply_attn = ApplyAttn(HIDDEN_SIZE * 2, NUM_CLASSES)
### Download ready models
# checkpoint = torch.load('crnn_final', map_location=device)
# CRNN_model.load_state_dict(checkpoint['model_state_dict'])
# checkpoint = torch.load('attn_final', map_location=device)
# attn_layer.load_state_dict(checkpoint['model_state_dict'])
# checkpoint = torch.load('apply_attn_final', map_location=device)
# apply_attn.load_state_dict(checkpoint['model_state_dict'])
full_model = FullModel(CRNN_model, attn_layer, apply_attn)
print(full_model.to(device))
print(count_parameters(full_model))
#wandb.init()
#wandb.watch(full_model)
### Create optimizer
opt = torch.optim.Adam(full_model.parameters(), weight_decay=1e-5)
### Train_val loop
for n in range(NUM_EPOCHS):
train_epoch(full_model,
opt,
train_loader,
melspec_train,
GRU_NUM_LAYERS,
#feat_std = checkpoint['std']
#acc = checkpoint['acc']
#epoch = checkpoint['epoch']
#print('==> Loading LSTM from epoch {}, acc={:2.3f}'.format(epoch, acc))
if use_cuda:
# Classifier
classifier.cuda()
classifier = torch.nn.DataParallel(classifier, device_ids=range(torch.cuda.device_count()))
# LSTM
#lstm.cuda()
#lstm = torch.nn.DataParallel(lstm, device_ids=range(torch.cuda.device_count()))
cudnn.benchmark = True
elif args.mode == 'full':
cnn_path = os.path.join(args.model_dir, '{}_{}.t7'.format(args.load_modality, args.method))
full_model = FullModel(cnn_path, args.lstm_path)
if False:#use_cuda and args.mode != 'profile':
net.cuda()
net = torch.nn.DataParallel(net, device_ids=range(torch.cuda.device_count()))
cudnn.benchmark = True
if args.mode != 'profile':
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=0.9, weight_decay=1e-5)
# Training
def profile(dataloader):
print('\n==> Profiling..')
t0 = time.time()
for batch_idx, (inputs, targets) in enumerate(dataloader):
self.yields_init = True
if __name__ == "__main__":
print("Testing {}".format(os.path.basename(__file__)))
from models import DefaultModel, DefaultCombinedDHModel
from models import FullModel
from PhysicalParameters import PhysicalParameters
from GammaFitter import GammaFitter
from GammaFitterXiSetup import GammaFitterXiSetup
dm = DefaultModel("../amplitude_calculations/output/KS_default.pickle")
dcm = DefaultCombinedDHModel(
"../amplitude_calculations/output/KS_default.pickle")
fm = FullModel("../amplitude_calculations/output/KS_default.pickle",
"../amplitude_calculations/output/KL_default.pickle",
PhysicalParameters().default_eps, PhysicalParameters(True))
bs = BiasStudy(fm,
param_sets=np.array(
[[uf.deg_to_rad(x), 0.1,
uf.deg_to_rad(130)] for x in [80]]),
KL_amplitude_files=[
"../amplitude_calculations/output/KL_seed_1.pickle",
"../amplitude_calculations/output/KL_seed_2.pickle"
])
gf = GammaFitter()
gf_xi = GammaFitterXiSetup()
bs.init_yields()
# bs.load_yields("tmp_n.pickle")
from io import BytesIO
import base64
import requests
from PIL import Image
from flask import Flask, request, render_template, flash, redirect
from flask_bootstrap import Bootstrap
from models import FullModel, PredictionError
app = Flask(__name__)
bootstrap = Bootstrap(app)
app.secret_key = b'_5#y2L"F4Q8z\n\xec]/'
app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024
model = FullModel("age_model_latest.state", n_tta_transforms=3)
@app.route('/', methods=['GET', 'POST'])
def hello():
name = request.args.get("name", "World")
return render_template("index.html", name=name)
@app.route('/result', methods=['POST'])
def predict():
image: Image.Image = None
url = request.form.get('url')
snapshot_base64 = request.form.get('snapshot')
if snapshot_base64:
data = base64.b64decode(snapshot_base64.split(',')[-1])
def test_blank(self):
model = FullModel()
model.full_clean()
def test_ipprefix(self):
model = FullModel()
model.prefix = '10.0.0.0/8'
with pytest.raises(ValidationError):
model.prefix = 'invalid'
def full_model():
amplitude_file = '{}/EvtGen_KS_default.pickle'.format(amp_dir)
amplitude_file2= '{}/EvtGen_KL_seed_3.pickle'.format(amp_dir)
binning_file = '../python/input/KsPiPi_optimal.pickle'
fm = FullModel(amplitude_file, amplitude_file2, binning_file=binning_file)
n = fm.predict_yields([0.2, 0.1, 0.5], 2000)
def get_gaus(self, name):
if self.gaus_fits.has_key(name):
return self.gaus_fits[name]
else:
print "Unknown key:", name
return None
if __name__ == "__main__":
from models import DefaultModel
from models import FullModel
from PhysicalParameters import PhysicalParameters
dm = DefaultModel("../amplitude_calculations/output/KS_default.pickle")
fm = FullModel("../amplitude_calculations/output/KS_default.pickle",
"../amplitude_calculations/output/KL_default.pickle", 0,
PhysicalParameters(False))
ts = ToyStudy(
dm, fm,
[uf.deg_to_rad(85), 0.1, uf.deg_to_rad(100)], 4000, 10)
test = np.random.normal(0, 1, size=1000)
print test.mean()
print test.std()
# print test
print ts.fit_gaussian(test)
print ts.run()
ts.seed = 3
print ts.run()
ts.seed = 1
