def __init__(self,
arch,
num_classes,
dims,
resnet_variant='alpha',
optimizer='adam',
lr=0.001,
beta1=0.9,
beta2=0.999,
weight_decay=0.0001,
momentum=0.9,
schedule='none',
steps=[100, 150],
step_factor=0.1,
k=5):
super(ComplexModel, self).__init__()
self.save_hyperparameters()
self.encoder, self.processor, self.decoder = load_complex_network(
arch,
num_classes,
resnet_variant,
)
self.real_to_complex = RealToComplex()
self.complex_to_real = ComplexToReal()
size = get_encoder_output_size(self.encoder, dims)
self.discriminator = Discriminator(size=size)
self.train_acc = Accuracy()
self.val_acc = Accuracy()
print(self.hparams)
def __init__(self,
input_size: int = 784,
lin1_size: int = 256,
lin2_size: int = 256,
lin3_size: int = 256,
output_size: int = 10,
lr: float = 0.001,
weight_decay: float = 0.0005,
**kwargs):
super().__init__()
# this line ensures params passed to LightningModule will be saved to ckpt
# it also allows to access params with 'self.hparams' attribute
self.save_hyperparameters()
self.model = SimpleDenseNet(hparams=self.hparams)
# loss function
self.criterion = torch.nn.CrossEntropyLoss()
# use separate metric instance for train, val and test step
# to ensure a proper reduction over the epoch
self.train_accuracy = Accuracy()
self.val_accuracy = Accuracy()
self.test_accuracy = Accuracy()
self.metric_hist = {
"train/acc": [],
"val/acc": [],
"train/loss": [],
"val/loss": [],
}
def __init__(
self,
arch,
num_classes,
additional_layers=False,
resnet_variant='alpha',
noisy=False,
gamma=1.0,
optimizer='sgd',
lr=0.1,
beta1=0.9,
beta2=0.999,
weight_decay=0.0001,
momentum=0.9,
schedule='none',
steps=[100, 150],
step_factor=0.1,
):
super(BaselineModel, self).__init__()
self.save_hyperparameters()
self.encoder, self.processor, self.decoder = load_baseline_network(
arch, num_classes, resnet_variant, additional_layers)
self.train_acc = Accuracy()
self.val_acc = Accuracy()
print(self.hparams)
def __init__(self, hparams):
super(Model, self).__init__()
self.hparams = hparams
self.model = load_network(hparams)
self.train_acc = Accuracy()
self.val_acc = Accuracy()
self.criterion = nn.BCEWithLogitsLoss()
def prepare(self):
(self.train_transforms,
self.infer_transforms) = self._build_transforms(
**self.config.transforms)
(
self._train_dataloader,
self._val_dataloader,
self._test_dataloader,
) = self._build_dataloaders(**self.config.data)
num_slots = len(
self.train_transforms["utterance&slots"][0].vocab.idx.keys())
num_intents = len(self.train_transforms["intent"][0].vocab.idx.keys())
add_feat_len = 0
if self.config.model.use_intent:
add_feat_len += num_intents
self.joint_model = build_intent_joint_model(
**self.config.model,
num_slots=num_slots,
num_intents=num_intents,
vocab=self.train_transforms["utterance"][1].vocab,
add_feat_len=add_feat_len,
)
self.slot_accuracy_metric = Accuracy(num_classes=num_slots)
self.doc_accuracy_metric = Accuracy(num_classes=num_intents)
self.precision_metric = AveragePrecision()
self.slot_f1_metric = F1(num_classes=num_slots)
self.doc_f1_metric = F1(num_classes=num_intents)
self.optimizer = FairSeqAdam(self.joint_model.parameters(),
**self.config.optimizer)
def __init__(
self,
encoder_weights,
inversion_net_weights,
dims,
num_classes,
arch='resnet56',
resnet_variant='alpha',
additional_layers=False,
optimizer='sgd',
lr=0.1,
beta1=0.9,
beta2=0.999,
weight_decay=0.0001,
momentum=0.9,
schedule='none',
steps=[100, 150],
step_factor=0.1,
complex=False,
angle_dis_weights=None,
):
super(InferenceAttack3Model, self).__init__()
self.save_hyperparameters()
# Load pretrained models
if complex:
# Load complex encoder
self.protype_encoder = ComplexEncoder(encoder_weights)
else:
self.protype_encoder = BaselineEncoder(encoder_weights)
self.protype_encoder.freeze()
# Feature inversion network
in_size = get_encoder_output_size(self.protype_encoder, dims)
self.inversion_network = InversionNetwork(inversion_net_weights,
in_size[-3])
self.inversion_network.freeze()
if complex:
# Angle discriminator
self.angle_discriminator = AngleDiscriminator(
angle_dis_weights, in_size)
self.angle_discriminator.freeze()
self.complex_to_real = ComplexToReal()
# Load attack classification model
self.encoder, self.processor, self.decoder = load_baseline_network(
arch, num_classes, resnet_variant, additional_layers=False)
self.train_acc = Accuracy()
self.val_acc = Accuracy()
print(self.hparams)
def __init__(self, hidden_dim=128, learning_rate=1e-3):
super().__init__()
self.save_hyperparameters()
# mnist images are (1, 28, 28) (channels, width, height)
self.layer_1 = nn.Linear(28 * 28, self.hparams.hidden_dim)
self.layer_2 = nn.Linear(self.hparams.hidden_dim,
self.hparams.hidden_dim * 2)
self.layer_3 = nn.Linear(self.hparams.hidden_dim * 2, 10)
self.train_accuracy = Accuracy()
self.val_accuracy = Accuracy()
self.test_accuracy = Accuracy()
def __init__(
self,
architecture: str = "GraphSAGE",
aggregation_method: str = "concat",
num_node_features: int = 1,
add_pos_to_features: bool = False,
num_conv_layers: int = 3,
conv_size: int = 128,
lin_size: int = 128,
output_size: int = 10,
lr: float = 0.001,
weight_decay: float = 0,
**kwargs
):
super().__init__()
self.save_hyperparameters()
self.add_pos_to_features = add_pos_to_features
# init network architecture
if self.hparams.architecture == "GraphSAGE":
self.model = graph_sage.GraphSage(hparams=self.hparams)
elif self.hparams.architecture == "GAT":
self.model = gat.GAT(hparams=self.hparams)
elif self.hparams.architecture == "JumpingKnowledge":
self.model = jumping_knowledge.JK(hparams=self.hparams)
elif self.hparams.architecture == "GCN":
self.model = gcn.GCN(hparams=self.hparams)
elif self.hparams.architecture == "VectorSAGE":
self.model = vector_sage_module.VectorSAGEModule(hparams=self.hparams)
else:
raise Exception("Incorrect architecture name!")
# loss function
self.criterion = torch.nn.CrossEntropyLoss()
# use separate metric instance for train, val and test step
# to ensure a proper reduction over the epoch
self.train_accuracy = Accuracy()
self.val_accuracy = Accuracy()
self.test_accuracy = Accuracy()
self.metric_hist = {
"train/acc": [],
"val/acc": [],
"train/loss": [],
"val/loss": [],
}
def __init__(self):
super().__init__()
self.metrics_singleclass = nn.ModuleDict({
'loss':
BCEWithLogitsLoss(),
'acc':
SoftMaxWrapper(Accuracy(), multiclass=False),
'auc':
AUROC(),
})
self.metrics_multiclass = nn.ModuleDict({
'loss':
CrossEntropyLoss(),
'acc':
SoftMaxWrapper(Accuracy(), multiclass=True),
})
def __init__(self, HPARAMS):
super().__init__()
# HPARAMS
self.save_hyperparameters()
self.models = {
'wav2vecLSTMAttn': Wav2VecLSTM,
'spectralCNNLSTM': SpectralCNNLSTM,
'MultiScale': SpectralMultiScale,
}
self.model = self.models[HPARAMS['model_type']](HPARAMS['hidden_size'])
self.classification_criterion = MSE()
self.regression_criterion = MSE()
self.mae_criterion = MAE()
self.rmse_criterion = RMSELoss()
self.accuracy = Accuracy()
self.alpha = HPARAMS['alpha']
self.beta = HPARAMS['beta']
self.gamma = HPARAMS['gamma']
self.lr = HPARAMS['lr']
self.csv_path = HPARAMS['speaker_csv_path']
self.df = pd.read_csv(self.csv_path)
self.h_mean = self.df['height'].mean()
self.h_std = self.df['height'].std()
self.a_mean = self.df['age'].mean()
self.a_std = self.df['age'].std()
print(
f"Model Details: #Params = {self.count_total_parameters()}\t#Trainable Params = {self.count_trainable_parameters()}"
)
def __init__(self, model, criterion, config: dict, loaders: dict):
super().__init__()
self.model = model
self.criterion = criterion
self.config = config
self.loaders = loaders
self.metrics = {'acc': Accuracy(), 'f1': F1()}
def __init__(self, HPARAMS):
super().__init__()
# HPARAMS
self.save_hyperparameters()
self.model = Wav2VecLSTM(HPARAMS['model_hidden_size'])
self.classification_criterion = MSE()
self.regression_criterion = MSE()
self.mae_criterion = MAE()
self.rmse_criterion = RMSELoss()
self.accuracy = Accuracy()
self.alpha = HPARAMS['model_alpha']
self.beta = HPARAMS['model_beta']
self.gamma = HPARAMS['model_gamma']
self.lr = HPARAMS['training_lr']
self.csv_path = HPARAMS['speaker_csv_path']
self.df = pd.read_csv(self.csv_path, sep=' ')
self.h_mean = self.df['Height'].mean()
self.h_std = self.df['Height'].std()
self.a_mean = self.df['Age'].mean()
self.a_std = self.df['Age'].std()
print(f"Model Details: #Params = {self.count_total_parameters()}\t#Trainable Params = {self.count_trainable_parameters()}")
def test_accuracy(num_classes):
acc = Accuracy(num_classes=num_classes)
assert acc.name == 'accuracy'
result = acc(pred=torch.tensor([[0, 1, 1], [1, 0, 1]]),
target=torch.tensor([[0, 0, 1], [1, 0, 1]]))
assert isinstance(result, torch.Tensor)
def __init__(self,
model: torch.nn.Module,
train_data: str,
val_data: str,
test_data: str,
batch_size: int,
num_classes: int = 2,
num_workers: int = 1):
super().__init__()
self.num_workers = num_workers
self.train_data = train_data
self.val_data = val_data
self.test_data = test_data
self.batch_size = batch_size
self.model = model
self.criterion = torch.nn.BCELoss()
self.collate_fn = PaddingCollateFn(150)
self.train_metrics = []
self.val_metrics = [
Accuracy(num_classes=num_classes),
F1(num_classes=num_classes),
Precision(num_classes=num_classes),
Recall(num_classes=num_classes)
]
def __init__(self, out_classes: int = 3, lr_bert: float = 1e-5, lr_class: float = 1e-4,
weight_decay: float = 1e-2, freeze_base: bool = False, train_steps: int = 100):
super(SentBert, self).__init__()
self.lr_bert = lr_bert
self.lr_class = lr_class
self.weight_decay = weight_decay
self.train_steps = train_steps
self.save_hyperparameters()
self.bert = BertForSequenceClassification.from_pretrained(
'bert-base-uncased', num_labels=out_classes, return_dict=True)
if freeze_base:
for param in self.bert.base_model.parameters():
param.requires_grad = False
self.f1 = F1(num_classes=out_classes, average='macro')
self.train_acc = Accuracy()
self.val_acc = Accuracy()
def __init__(
self,
num_classes: int,
backbone: str = "prajjwal1/bert-tiny",
optimizer: Type[torch.optim.Optimizer] = torch.optim.Adam,
metrics: Union[Callable, Mapping, Sequence, None] = [Accuracy()],
learning_rate: float = 1e-3,
):
self.save_hyperparameters()
os.environ["TOKENIZERS_PARALLELISM"] = "TRUE"
# disable HF thousand warnings
warnings.simplefilter("ignore")
# set os environ variable for multiprocesses
os.environ["PYTHONWARNINGS"] = "ignore"
super().__init__(
model=None,
loss_fn=None,
optimizer=optimizer,
metrics=metrics,
learning_rate=learning_rate,
)
self.model = BertForSequenceClassification.from_pretrained(
backbone, num_labels=num_classes)
def __init__(self, *args, **kwargs):
super().__init__()
self.save_hyperparameters()
self.loss = BCELoss(reduction='mean')
self.accuracy = Accuracy()
# Initialize model architecture
self.architecture = XLMREncoder(hparams=self.hparams)
def __init__(self, config):
super().__init__()
self.train_config = config
self.roberta = RobertaForMaskedLM.from_pretrained('roberta-base')
_ = self.roberta.eval()
for param in self.roberta.parameters():
param.requires_grad = False
self.pred_model = self.roberta.roberta
self.enc_model = self.pred_model.embeddings.word_embeddings
self.proj_head = DVProjectionHead_EmbActi()
self.lossfunc = nn.BCEWithLogitsLoss()
self.acc = Accuracy(threshold=0.0)
self.f1 = F1(threshold=0.0)
def __init__(self, config, trial=None):
super(LitPSD, self).__init__()
if trial:
self.trial = trial
else:
self.trial = None
self.pylog = logging.getLogger(__name__)
logging.getLogger("lightning").setLevel(self.pylog.level)
self.config = config
if hasattr(config.system_config, "half_precision"):
self.needs_float = not config.system_config.half_precision
else:
self.needs_float = True
self.hparams = DictionaryUtility.to_dict(config)
self.n_type = config.system_config.n_type
self.lr = config.optimize_config.lr
self.modules = ModuleUtility(config.net_config.imports +
config.dataset_config.imports +
config.optimize_config.imports)
self.model_class = self.modules.retrieve_class(
config.net_config.net_class)
# self.data_module = PSDDataModule(config,self.device)
self.model = self.model_class(config)
self.criterion_class = self.modules.retrieve_class(
config.net_config.criterion_class)
self.criterion = self.criterion_class(
*config.net_config.criterion_params)
self.softmax = LogSoftmax(dim=1)
self.accuracy = Accuracy()
self.confusion = ConfusionMatrix(num_classes=self.n_type)
if hasattr(self.config.dataset_config, "calgroup"):
calgroup = self.config.dataset_config.calgroup
else:
calgroup = None
if self.config.dataset_config.dataset_class == "PulseDatasetDet":
self.evaluator = PhysEvaluator(
self.config.system_config.type_names,
self.logger,
device=self.device)
elif self.config.dataset_config.dataset_class == "PulseDatasetWaveformNorm":
self.evaluator = TensorEvaluator(self.logger,
calgroup=calgroup,
target_has_phys=False,
target_index=None,
metric_name="accuracy",
metric_unit="")
else:
if hasattr(self.config.dataset_config, "calgroup"):
self.evaluator = PSDEvaluator(
self.config.system_config.type_names,
self.logger,
device=self.device,
calgroup=self.config.dataset_config.calgroup)
else:
self.evaluator = PSDEvaluator(
self.config.system_config.type_names,
self.logger,
device=self.device)
def __init__(self, hparams):
super(YourModel, self).__init__()
"""
Initialise custom objects here
"""
self.hparams = hparams
self.model = ImageNet_Pretrained_Model()
self.accuracy = pl.metrics.Accuracy()
self.acc = Accuracy()
def __init__(self, hparams):
super().__init__()
self.hparams = hparams
self.train_batch_size = hparams.train_batch_size
self.test_batch_size = hparams.test_batch_size
self.input_size = ((3, ) if hparams.pseudo3d else
(1, )) + tuple(hparams.resize)
self.classifier = Encoder(num_convolutions=hparams.num_convolutions,
filters=hparams.filters,
input_size=self.input_size,
latent_dim=1,
use_weight_norm=hparams.use_weight_norm,
dropout_rate=hparams.dropout_rate)
self.classifier.fc = nn.Linear(np.prod(
self.classifier.intermediate_shape),
1,
bias=True)
self.train_acc = Accuracy()
self.val_acc = Accuracy()
self.test_acc = Accuracy()
if hparams.validate:
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.autograd.set_detect_anomaly(self.hparams.validate)
resize = None if self.hparams.resize == (0, 0) else self.hparams.resize
train_crop_type = self.hparams.train_crop_type if hasattr(
self.hparams, 'train_crop_type') else 'random'
crop_size = self.hparams.crop_size if hasattr(
self.hparams, 'crop_size') else (224, 224)
self.calabresi_train = CalabresiDataset(self.hparams.train_csv,
crop_size=crop_size,
crop_type=train_crop_type,
resize=resize) # noqa: E501
self.calabresi_val = CalabresiDataset(self.hparams.valid_csv,
crop_size=crop_size,
crop_type='center',
resize=resize)
self.calabresi_test = CalabresiDataset(self.hparams.test_csv,
crop_size=crop_size,
crop_type='center',
resize=resize)
def __init__(self, HPARAMS):
super().__init__()
# HPARAMS
self.save_hyperparameters()
lstm_inp=512
lstm_h = HPARAMS['model_hidden_size']
alpha = HPARAMS['model_alpha']
beta = HPARAMS['model_beta']
gamma = HPARAMS['model_gamma']
csv_path = HPARAMS['speaker_csv_path']
self.encoder = Wav2Vec(pretrained=True)
for param in self.encoder.parameters():
param.requires_grad = False
for param in self.encoder.feature_extractor.conv_layers[5:].parameters():
param.requires_grad = True
self.lstm = nn.LSTM(lstm_inp, lstm_h, batch_first=True)
self.attention = Attention(lstm_h)
self.height_regressor = nn.Linear(lstm_h, 1)
self.age_regressor = nn.Linear(lstm_h, 1)
self.gender_classifier = nn.Linear(lstm_h, 1)
self.classification_criterion = MSE()
self.regression_criterion = MSE()
self.mae_criterion = MAE()
self.rmse_criterion = RMSELoss()
self.accuracy = Accuracy()
self.alpha = alpha
self.beta = beta
self.gamma = gamma
self.lr = HPARAMS['training_lr']
self.csv_path = csv_path
self.df = pd.read_csv(self.csv_path, sep=' ')
self.h_mean = self.df['Height'].mean()
self.h_std = self.df['Height'].std()
self.a_mean = self.df['Age'].mean()
self.a_std = self.df['Age'].std()
print(f"Model Details: #Params = {self.count_total_parameters()}\t#Trainable Params = {self.count_trainable_parameters()}")
def __init__(self,
classifier_weights,
encoder_weights,
inversion_net_weights,
dims,
complex=False,
angle_dis_weights=None,
verbose=False):
super(InferenceAttack1Model, self).__init__()
self.save_hyperparameters()
# Load classification network
self.classifier = BaselineNetwork(classifier_weights)
self.classifier.freeze()
if complex:
# Load complex encoder
self.encoder = ComplexEncoder(encoder_weights)
else:
self.encoder = BaselineEncoder(encoder_weights)
self.encoder.freeze()
# Load feature inversion network
in_size = get_encoder_output_size(self.encoder, dims)
self.inversion_network = InversionNetwork(inversion_net_weights,
in_size[-3])
self.inversion_network.freeze()
if complex:
# Load angle discriminator
self.angle_discriminator = AngleDiscriminator(
angle_dis_weights, in_size)
self.angle_discriminator.freeze()
self.complex_to_real = ComplexToReal()
self.acc = Accuracy()
if verbose:
print(self.hparams)
download_data("https://pl-flash-data.s3.amazonaws.com/titanic.zip", 'data/')
# 2. Load the data
datamodule = TabularData.from_csv(
"./data/titanic/titanic.csv",
test_csv="./data/titanic/test.csv",
categorical_input=[
"Sex", "Age", "SibSp", "Parch", "Ticket", "Cabin", "Embarked"
],
numerical_input=["Fare"],
target="Survived",
val_size=0.25,
)
# 3. Build the model
model = TabularClassifier.from_data(
datamodule, metrics=[Accuracy(), Precision(),
Recall()])
# 4. Create the trainer. Run 10 times on data
trainer = flash.Trainer(max_epochs=10)
# 5. Train the model
trainer.fit(model, datamodule=datamodule)
# 6. Test model
trainer.test()
# 7. Save it!
trainer.save_checkpoint("tabular_classification_model.pt")
def accuracy_score(labels, predictions):
predictions = torch.Tensor(predictions)
labels = torch.Tensor(labels)
accuracy = Accuracy(threshold=cfg['training']['threshold'])
return accuracy(predictions, labels).item()
class LightningLongformerCLS(pl.LightningModule):
def __init__(self, config):
super().__init__()
self.train_config = config
self.roberta = RobertaForMaskedLM.from_pretrained('roberta-base')
_ = self.roberta.eval()
for param in self.roberta.parameters():
param.requires_grad = False
self.pred_model = self.roberta.roberta
self.enc_model = self.pred_model.embeddings.word_embeddings
# self.proj_head = DVProjectionHead()
# self.proj_head = DVProjectionHead_ActiFirst()
self.proj_head = DVProjectionHead_EmbActi()
self.tkz = RobertaTokenizer.from_pretrained("roberta-base")
self.collator = TokenizerCollate(self.tkz)
self.lossfunc = nn.BCEWithLogitsLoss()
self.acc = Accuracy(threshold=0.0)
self.f1 = F1(threshold=0.0)
def configure_optimizers(self):
optimizer = torch.optim.AdamW(self.parameters(),
lr=self.train_config["learning_rate"])
scheduler = transformers.get_cosine_with_hard_restarts_schedule_with_warmup(
optimizer,
num_warmup_steps=10,
num_training_steps=5000,
num_cycles=10)
schedulers = [{
'scheduler': scheduler,
'interval': 'step',
'frequency': 1
}]
return [optimizer], schedulers
def train_dataloader(self):
# self.dataset_train = PANDataset('./data_pickle_cutcombo/pan_all_cls/train_kucombo_only.pickle')
# self.dataset_train = PANDataset('./data_pickle_cutcombo/pan_14e_cls/train_essays.pickle')
self.dataset_train = PANDataset(
'./data_pickle_cutcombo/pan_14n_cls/train_novels_kucombo_only.pickle'
)
self.loader_train = DataLoader(
self.dataset_train,
batch_size=self.train_config["batch_size"],
collate_fn=self.collator,
num_workers=4,
pin_memory=True,
drop_last=False,
shuffle=False)
return self.loader_train
def val_dataloader(self):
# self.dataset_val = PANDataset('./data_pickle_cutcombo/pan_14e_cls/test02_essays_onecut.pickle')
self.dataset_val = PANDataset(
'./data_pickle_cutcombo/pan_14n_cls/test02_novels_onecut.pickle')
self.loader_val = DataLoader(
self.dataset_val,
batch_size=self.train_config["batch_size"],
collate_fn=self.collator,
num_workers=4,
pin_memory=True,
drop_last=False,
shuffle=False)
return self.loader_val
def test_dataloader(self):
# self.dataset_test = PANDataset('./data_pickle_cutcombo/pan_14e_cls/test02_essays_onecut.pickle')
self.dataset_test = PANDataset(
'./data_pickle_cutcombo/pan_14n_cls/test02_novels_onecut.pickle')
self.loader_test = DataLoader(
self.dataset_test,
batch_size=self.train_config["batch_size"],
collate_fn=self.collator,
num_workers=4,
pin_memory=True,
drop_last=False,
shuffle=False)
return self.loader_test
@autocast()
def forward(self, inputs, onedoc_enc=False):
def one_doc_embed(input_ids, input_mask, mask_n=1):
uniq_mask = []
uniq_input, inverse_indices = torch.unique(input_ids,
return_inverse=True,
dim=0)
invi = inverse_indices.detach().cpu().numpy()
for i in range(uniq_input.shape[0]):
first_index = np.where(invi == i)[0][0]
uniq_mask.append(input_mask[first_index, :])
input_ids = uniq_input
input_mask = torch.stack(uniq_mask, dim=0)
embed = self.enc_model(input_ids)
result_embed = []
result_pred = []
# skip start and end symbol
masked_ids = input_ids.clone()
for i in range(1, input_ids.shape[1] - mask_n):
masked_ids[:, i:(i + mask_n)] = self.tkz.mask_token_id
output = self.pred_model(input_ids=masked_ids,
attention_mask=input_mask,
return_dict=False)[0]
result_embed.append(embed[:, i:(i + mask_n), :])
result_pred.append(output[:, i:(i + mask_n), :])
masked_ids[:, i:(i + mask_n)] = input_ids[:, i:(i + mask_n)]
# stack along doc_len
result_embed = torch.cat(result_embed, dim=1)
result_pred = torch.cat(result_pred, dim=1)
rec_embed = []
rec_pred = []
for i in invi:
rec_embed.append(result_embed[i, :, :])
rec_pred.append(result_pred[i, :, :])
rec_embed = torch.stack(rec_embed, dim=0)
rec_pred = torch.stack(rec_pred, dim=0)
return rec_embed, rec_pred
if onedoc_enc:
doc_ids, doc_mask = inputs
doc_embed, doc_pred = one_doc_embed(input_ids=doc_ids,
input_mask=doc_mask)
doc_dv = doc_pred - doc_embed
return doc_pred, doc_embed, doc_dv
else:
labels, kno_ids, kno_mask, unk_ids, unk_mask = inputs
kno_embed, kno_pred = one_doc_embed(input_ids=kno_ids,
input_mask=kno_mask)
unk_embed, unk_pred = one_doc_embed(input_ids=unk_ids,
input_mask=unk_mask)
kno_dv = kno_pred - kno_embed
unk_dv = unk_pred - unk_embed
# logits = self.proj_head(kno_dv, kno_mask[:,1:-1], unk_dv, unk_mask[:,1:-1])
logits = self.proj_head(kno_embed, kno_dv, kno_mask[:, 1:-1],
unk_embed, unk_dv, unk_mask[:, 1:-1])
logits = torch.squeeze(logits)
labels = labels.float()
loss = self.lossfunc(logits, labels)
return (loss, logits, (kno_embed, kno_pred, unk_embed, unk_pred))
def training_step(self, batch, batch_idx):
labels, kno_ids, kno_mask, unk_ids, unk_mask = batch
loss, logits, outputs = self(
(labels, kno_ids, kno_mask, unk_ids, unk_mask))
self.log("train_loss", loss)
self.log("logits mean", logits.mean())
self.log("LR", self.trainer.optimizers[0].param_groups[0]['lr'])
return loss
def validation_step(self, batch, batch_idx):
labels, kno_ids, kno_mask, unk_ids, unk_mask = batch
loss, logits, outputs = self(
(labels, kno_ids, kno_mask, unk_ids, unk_mask))
self.acc(logits, labels.float())
self.f1(logits, labels.float())
return {"val_loss": loss}
def validation_epoch_end(self, validation_step_outputs):
avg_loss = torch.stack(
[x['val_loss'] for x in validation_step_outputs]).mean()
self.log("val_loss", avg_loss)
self.log('eval accuracy', self.acc.compute())
self.log('eval F1', self.f1.compute())
str(commit)[2:-3],
shell=True)
return tag
if __name__ == '__main__':
with open("params.yaml", 'r') as fd:
params = yaml.safe_load(fd)
MODEL_PATH = params['test']['model_path']
module = TrainingModule.load_from_checkpoint(MODEL_PATH).to('cuda')
module.eval()
acc = Accuracy()
f1 = F1()
precision = Precision()
recall = Recall()
mlflow.set_tracking_uri('file-plugin:/content/NLP_Emotions/mlruns')
if get_closest_gittag() == 'v2.0':
mlflow.set_experiment('SGD')
mlflow.set_tag('Version', 'SGD')
mlflow.set_tag('Stage', 'test')
mlflow.set_tag('Commit', get_commit())
mlflow.set_tag('Time', get_commit_time())
mlflow.set_tag('Model', module.model_name)
mlflow.log_params({
'batch_size': module.hparams.batch_size,
'epochs': module.hparams.epochs,
def __init__(self, *args, **kwargs):
super(LanguageModel, self).__init__(*args, **kwargs)
self.lossf = nn.CrossEntropyLoss()
self.metric = Accuracy(self.outclasses)
def get_acc(self, batch, infer):
pred = self.get_pred(infer)
targ = infer[1]
flat_pred = pred.reshape(-1)
flat_targ = targ.reshape(-1)
return Accuracy()(flat_pred, flat_targ)
if __name__ == "__main__":
# 1. Download the data
download_data("https://pl-flash-data.s3.amazonaws.com/titanic.zip", 'data/')
# 2. Load the data
datamodule = TabularData.from_csv(
"./data/titanic/titanic.csv",
test_csv="./data/titanic/test.csv",
categorical_input=["Sex", "Age", "SibSp", "Parch", "Ticket", "Cabin", "Embarked"],
numerical_input=["Fare"],
target="Survived",
val_size=0.25,
)
# 3. Build the model
model = TabularClassifier.from_data(datamodule, metrics=[Accuracy(), Precision(), Recall()])
# 4. Create the trainer. Run 10 times on data
trainer = flash.Trainer(max_epochs=10)
# 5. Train the model
trainer.fit(model, datamodule=datamodule)
# 6. Test model
trainer.test()
# 7. Save it!
trainer.save_checkpoint("tabular_classification_model.pt")