def test_container_get_commands_empty(self):
reg = Registry()
reg.register(FakeApp2())
commands = reg.get_commands()
self.assertFalse(len(commands) > 0)
def test_container_get_commands(self):
reg = Registry()
reg.register(FakeApp1())
reg.register(FakeApp2())
commands = reg.get_commands()
self.assertTrue(len(commands) > 0)
def test_build_registry(self):
reg = Registry()
reg.register(FakeApp1())
reg.register(FakeApp2())
reg.register(FakeApp4())
# check if list of apps is not empty list
self.assertIsNot(reg.get_apps(), [])
self.assertEqual(3, len(reg.get_apps()))
self.assertEqual(3, len(reg.get_routes()))
def test_container_error(self):
# test if application not implement container abstract class
with self.assertRaises(ContainerError) as container_error:
reg = Registry()
reg.register(FakeApp3())
exception = container_error.exception
self.assertEqual('FakeApp3', exception.app_name)
self.assertEqual(
'ContainerError: Cannot use FakeApp3 as container application',
exception.message)
def setUp(self):
self._subscriber_registry = Registry()
self._commandProcessor = CommandProcessor(
registry=self._subscriber_registry)
import torch
import torch.nn as nn
from core.registry import Registry
from core.trainer import _initialize_weights
from .ml_heads import build_ml_head
from .layers import LocalBlock
MODEL_REGISTRY = Registry("MODEL_TYPE")
MODEL_REGISTRY.__doc__ = """
Registry for Mammo models.
"""
BACKBONE_REGISTRY = Registry("BACKBONE_TYPE")
BACKBONE_REGISTRY.__doc__ = """
Registry for Mammo cls backbone.
"""
def build_model(cfg):
"""Build the whole model architecture
"""
model_name = cfg.MODEL.NAME
model = MODEL_REGISTRY.get(model_name)(cfg)
return model
@MODEL_REGISTRY.register()
class ResNet_v0(nn.Module):
"""Original ResNet
"""
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from core.registry import Registry
ML_HEAD_REGISTRY = Registry("ML_HEAD")
ML_HEAD_REGISTRY.__doc__ = """
Registry for metric learning heads.
"""
def build_ml_head(cfg, in_channels):
head_name = cfg.MODEL.ML_HEAD.NAME
out_channels = cfg.MODEL.NUM_CLASSES
s = cfg.MODEL.ML_HEAD.SCALER
m = cfg.MODEL.ML_HEAD.MARGIN
num_centers = cfg.MODEL.ML_HEAD.NUM_CENTERS
head = ML_HEAD_REGISTRY.get(head_name)(in_channels, out_channels, s, m,
num_centers)
return head
@ML_HEAD_REGISTRY.register()
class ArcFace(nn.Module):
"""
This module implements ArcFace.
"""
def __init__(self, in_channels, out_channels, s=32., m=0.5, num_centers=1):
super(ArcFace, self).__init__()
from core.buffer import Buffer
from core.test_buffer import TestBuffer
from core.registry import Registry
########## App setup ##########
app = Application()
r = app.router
########## Instantiate globals ##########
env_type = os.environ.get('ENV_TYPE')
if env_type == "production":
buffer = Buffer()
else:
buffer = TestBuffer()
registry = Registry(buffer=buffer)
########## Error handling ##########
def get_required_param(json, param):
if json is None:
logger.info("Request is not a valid json")
raise InvalidUsage("Request is not a valid json")
value = json.get(param, None)
if (value is None) or (value == '') or (value == []):
logger.info("A required request parameter '{}' had value {}".format(
param, value))
raise InvalidUsage(
"A required request parameter '{}' was not provided".format(param))
return value
def __init__(
self,
random_seed: Optional[int] = None,
force_cache_refresh: bool = False,
nomorphi: bool = False,
planner: str = "full",
) -> None:
"""
:param random_seed: seed for random number generation, for repeatability
:param force_cache_refresh:
:param nomorphi: don't load Omorphi for morphological generation. This removes the dependency on Omorphi,
so allows easier setup, but means that no morphological inflection will be performed on the output,
which is generally a very bad thing for the full pipeline
"""
# New registry and result importer
self.registry = Registry()
# DataSets
DATA_ROOT = Path(__file__).parent.absolute() / ".." / "data"
self.datasets = [
"cphi",
"health_cost",
"health_funding",
]
for dataset in self.datasets:
cache_path: Path = (DATA_ROOT / "{}.cache".format(dataset)).absolute()
if not cache_path.exists():
raise IOError("No cached dataset found at {}. Datasets must be generated before startup.")
self.registry.register("{}-data".format(dataset), DataFrameStore(str(cache_path)))
# Resources
self.resources = [
CPHIEnglishResource(),
CPHIFinnishResource(),
CPHICroatianResource(),
CPHIRussianResource(),
CPHIEstonianResource(),
CPHISlovenianResource(),
ENVEnglishResource(),
ENVFinnishResource(),
HealthCostEnglishResource(),
HealthCostFinnishResource(),
HealthFundingEnglishResource(),
HealthFundingFinnishResource(),
]
# Templates
self.registry.register("templates", self._load_templates())
# Slot Realizers:
realizers: List[SlotRealizerComponent] = []
for resource in self.resources:
for realizer in resource.slot_realizer_components():
realizers.append(realizer(self.registry))
self.registry.register("slot-realizers", realizers)
# Language metadata
self.registry.register("conjunctions", CONJUNCTIONS)
self.registry.register("errors", ERRORS)
# PRNG seed
self._set_seed(seed_val=random_seed)
def _get_components(headline=False, planner="full"):
# Put together the list of components
# This varies depending on whether it's for headlines and which language we are doing stuff in
yield EUMessageGenerator(expand=True)
yield EUImportanceSelector()
if planner == "random":
yield EURandomHeadlineDocumentPlanner() if headline else EURandomBodyDocumentPlanner()
elif planner == "score":
yield EUScoreHeadlineDocumentPlanner() if headline else EUScoreBodyDocumentPlanner()
elif planner == "earlystop":
yield EUEarlyStopHeadlineDocumentPlanner() if headline else EUEarlyStopBodyDocumentPlanner()
elif planner == "topicsim":
yield EUTopicSimHeadlineDocumentPlanner() if headline else EUTopicSimBodyDocumentPlanner()
elif planner == "contextsim":
yield EUContextSimHeadlineDocumentPlanner() if headline else EUContextSimBodyDocumentPlanner()
elif planner == "neuralsim":
if headline:
yield EUHeadlineDocumentPlanner()
else:
yield TemplateAttacher()
yield EUNeuralSimBodyDocumentPlanner()
yield EmbeddingRemover()
elif planner == "full":
yield EUHeadlineDocumentPlanner() if headline else EUBodyDocumentPlanner()
else:
raise ValueError("INCORRECT PLANNER SETTING")
yield TemplateSelector()
yield Aggregator()
yield SlotRealizer()
yield LanguageSplitComponent(
{
"en": EnglishEUDateRealizer(),
"fi": FinnishEUDateRealizer(),
"hr": CroatianEUDateRealizer(),
"de": GermanEUDateRealizer(),
"ru": RussianEUDateRealizer(),
"ee": EstonianEUDateRealizer(),
"sl": SlovenianEUDateRealizer(),
}
)
yield EUEntityNameResolver()
yield EUNumberRealizer()
yield MorphologicalRealizer(
{
"en": EnglishUralicNLPMorphologicalRealizer(),
"fi": FinnishUralicNLPMorphologicalRealizer(),
"hr": CroatianSimpleMorphologicalRealizer(),
"ru": RussianMorphologicalRealizer(),
"ee": EstonianUralicNLPMorphologicalRealizer(),
"sl": SlovenianSimpleMorphologicalRealizer(),
}
)
yield HeadlineHTMLSurfaceRealizer() if headline else BodyHTMLSurfaceRealizer()
log.info("Configuring Body NLG Pipeline (planner = {})".format(planner))
self.body_pipeline = NLGPipeline(self.registry, *_get_components(planner=planner))
self.headline_pipeline = NLGPipeline(self.registry, *_get_components(headline=True, planner=planner))
from apps.hello.routes import app as HelloApp
from apps.ping.routes import app as PingApp
from apps.mongo.routes import app as MongoApp
from core.registry import Registry
from core.exceptions.application import ContainerError
try:
"""
You should register all of your apps here.
"""
apps = Registry()
apps.register(HelloApp)
apps.register(PingApp)
apps.register(MongoApp)
except ContainerError, e:
"""
I cannot use logging here. If i use default logging here,
it will reset all tornado's logging configurations, so i just
use simple print statement.
"""
print e.message
# -*- coding: utf-8 -*-
from core.registry import Registry
# classic models used as backbones
BACKBONES = Registry('backbones')
# modules used for extracting features
FEATURE_EXTRACTORS = Registry('feature_extractors')
LOSSES = Registry('losses')
# whole models used as detectors
DETECTORS = Registry('detectors')
# bbox coder(transform bbox format)
BBOX_CODERS = Registry('bbox_coder')
# similarity_calc
SIMILARITY_CALCS = Registry('similarity_calcs')
# matcher
MATCHERS = Registry('matchers')
# anchor_generators
ANCHOR_GENERATORS = Registry('anchor_generators')
# dataset and dataloaders
DATASETS = Registry('datasets')
DATALOADERS = Registry('dataloaders')
from core.registry import Registry
from models.tf_keras_model import TFKerasModel
from models.tf_v1_model import TFV1Model
model_registry = Registry('Model')
from models.q_model import *
from models.ac_model import *
from models.q_model_keras import *
from models.ac_model_keras import *
from models.custom_model import *