def train_individual_cased_joy():
train, dev, _ = get_bert_data_loader('joy', uncased=False)
final_training_stats = []
for d in [0.1, 0.2, 0.3]:
for w in [0.8, 0.85, 0.9, 0.95]:
for e in [1e-06, 1e-07, 1e-08]:
for lr in [2e-5, 3e-5, 5e-5]:
cased_model = BertForCasedClassification(dropout=d)
cased_trained, _, _, _ = train_model(
cased_model,
'joy_cased',
train,
dev,
filepath='./models/joy_cased/',
lr=lr,
eps=e,
weight_decay=w)
final_training_stats.extend(cased_trained)
pd.DataFrame(final_training_stats).to_csv(
'./data/models/cased-bert/results/joy/results.csv')
def train_individual_uncased_sadness():
train, dev, _ = get_bert_data_loader('sadness')
final_training_stats = []
for d in [0.1, 0.2, 0.3]:
for w in [0.8, 0.85, 0.9, 0.95]:
for e in [1e-06, 1e-07, 1e-08]:
for lr in [2e-5, 3e-5, 5e-5]:
uncased_model = BertForUncasedClassification(dropout=d)
uncased_trained, _, _, _ = train_model(
uncased_model,
'sadness_uncased',
train,
dev,
filepath='./models/sadness_uncased/',
lr=lr,
eps=e,
weight_decay=w)
final_training_stats.extend(uncased_trained)
pd.DataFrame(final_training_stats).to_csv(
'./data/models/uncased-bert/results/sadness/results.csv')
def train_ordinal_fear():
fear_train_dataloader_uncased, fear_val_dataloader_uncased, fear_test_dataloader_uncased = get_bert_data_loader(
'fear')
fear_uncased_model_1 = BertForUncasedClassification_1()
#fear_cased_model_1 = BertForCasedClassification_1()
fear_uncased_model_2 = BertForUncasedClassification_2()
#fear_cased_model_2 = BertForCasedClassification_2()
fear_uncased_model_3 = BertForUncasedClassification_3()
#fear_cased_model_3 = BertForCasedClassification_3()
### has to be based on the data distribution for each emotion
class_weights1 = torch.tensor([1, 2], dtype=torch.float)
class_weights2 = torch.tensor([1, 4], dtype=torch.float)
class_weights3 = torch.tensor([1, 10], dtype=torch.float)
final_training_stats_au_1 = []
for lr in [2e-5, 3e-5, 5e-5]:
fear_uncased_model_1 = BertForUncasedClassification_1()
uncased_fear_train_1, model1_fearUncased = train(
class_weights1,
1,
fear_uncased_model_1,
fear_train_dataloader_uncased,
fear_test_dataloader_uncased,
filepath='fear_uncased_ordinal',
lr=lr)
final_training_stats_au_1.append(uncased_fear_train_1)
pd.DataFrame(final_training_stats_au_1[0]).to_csv(
'./data/models/ordinal/fear/lr_2e-5_model1.csv')
pd.DataFrame(final_training_stats_au_1[1]).to_csv(
'./data/models/ordinal/fear/lr_3e-5_model1.csv')
pd.DataFrame(final_training_stats_au_1[2]).to_csv(
'./data/models/ordinal/fear/lr_5e-5_model1.csv')
final_training_stats_au_2 = []
for lr in [2e-5, 3e-5, 5e-5]:
fear_uncased_model_2 = BertForUncasedClassification_2()
uncased_fear_train_2, fear_uncased_model2 = train(
class_weights2,
2,
fear_uncased_model_2,
fear_train_dataloader_uncased,
fear_test_dataloader_uncased,
filepath='fear_uncased_ordinal',
lr=lr)
final_training_stats_au_2.append(uncased_fear_train_2)
pd.DataFrame(final_training_stats_au_2[0]).to_csv(
'./data/models/ordinal/fear/lr_2e-5_model2.csv')
pd.DataFrame(final_training_stats_au_2[1]).to_csv(
'./data/models/ordinal/fear/lr_3e-5_model2.csv')
pd.DataFrame(final_training_stats_au_2[2]).to_csv(
'./data/models/ordinal/fear/lr_5e-5_model2.csv')
final_training_stats_au_3 = []
for lr in [2e-5, 3e-5, 5e-5]:
fear_uncased_model_3 = BertForUncasedClassification_3()
uncased_fear_train_3, fear_uncased_model3 = train(
class_weights3,
3,
fear_uncased_model_3,
fear_train_dataloader_uncased,
fear_test_dataloader_uncased,
filepath='fear_uncased_ordinal',
lr=lr)
final_training_stats_au_3.append(uncased_fear_train_3)
pd.DataFrame(final_training_stats_au_3[0]).to_csv(
'./data/models/ordinal/fear/lr_2e-5_model3.csv')
pd.DataFrame(final_training_stats_au_3[1]).to_csv(
'./data/models/ordinal/fear/lr_3e-5_model3.csv')
pd.DataFrame(final_training_stats_au_3[2]).to_csv(
'./data/models/ordinal/fear/lr_5e-5_model3.csv')
def train_ordinal_sadness():
sadness_train_dataloader_uncased, sadness_val_dataloader_uncased, sadness_test_dataloader_uncased = get_bert_data_loader('sadness')
sadness_uncased_model_1 = BertForUncasedClassification_1()
#sadness_cased_model_1 = BertForCasedClassification_1()
sadness_uncased_model_2 = BertForUncasedClassification_2()
#sadness_cased_model_2 = BertForCasedClassification_2()
sadness_uncased_model_3 = BertForUncasedClassification_3()
#sadness_cased_model_3 = BertForCasedClassification_3()
### has to be based on the data distribution for each emotion
class_weights1 = torch.tensor([1.56,1], dtype = torch.float)
class_weights2 = torch.tensor([1,1.29], dtype = torch.float)
class_weights3 = torch.tensor([1,3.96], dtype = torch.float)
final_training_stats_au_1 = []
for lr in [2e-5, 3e-5, 5e-5]:
sadness_uncased_model_1 = BertForUncasedClassification_1()
uncased_sadness_train_1, model1_sadnessUncased = train(
class_weights1,
1,
sadness_uncased_model_1,
sadness_train_dataloader_uncased,
sadness_test_dataloader_uncased,
filepath='sadness_uncased_ordinal',
lr=lr
)
final_training_stats_au_1.append(uncased_sadness_train_1)
pd.DataFrame(final_training_stats_au_1[0]).to_csv('./data/models/ordinal/sadness/lr_2e-5_model1.csv')
pd.DataFrame(final_training_stats_au_1[1]).to_csv('./data/models/ordinal/sadness/lr_3e-5_model1.csv')
pd.DataFrame(final_training_stats_au_1[2]).to_csv('./data/models/ordinal/sadness/lr_5e-5_model1.csv')
final_training_stats_au_2 = []
for lr in [2e-5, 3e-5, 5e-5]:
sadness_uncased_model_2 = BertForUncasedClassification_2()
uncased_sadness_train_2, sadness_uncased_model2 = train(
class_weights2,
2,
sadness_uncased_model_2,
sadness_train_dataloader_uncased,
sadness_test_dataloader_uncased,
filepath='sadness_uncased_ordinal',
lr=lr
)
final_training_stats_au_2.append(uncased_sadness_train_2)
pd.DataFrame(final_training_stats_au_2[0]).to_csv('./data/models/ordinal/sadness/lr_2e-5_model2.csv')
pd.DataFrame(final_training_stats_au_2[1]).to_csv('./data/models/ordinal/sadness/lr_3e-5_model2.csv')
pd.DataFrame(final_training_stats_au_2[2]).to_csv('./data/models/ordinal/sadness/lr_5e-5_model2.csv')
final_training_stats_au_3 = []
for lr in [2e-5, 3e-5, 5e-5]:
sadness_uncased_model_3 = BertForUncasedClassification_3()
uncased_sadness_train_3, sadness_uncased_model3 = train(
class_weights3,
3,
sadness_uncased_model_3,
sadness_train_dataloader_uncased,
sadness_test_dataloader_uncased,
filepath='sadness_uncased_ordinal',
lr=lr
)
final_training_stats_au_3.append(uncased_sadness_train_3)
pd.DataFrame(final_training_stats_au_3[0]).to_csv('./data/models/ordinal/sadness/lr_2e-5_model3.csv')
pd.DataFrame(final_training_stats_au_3[1]).to_csv('./data/models/ordinal/sadness/lr_3e-5_model3.csv')
pd.DataFrame(final_training_stats_au_3[2]).to_csv('./data/models/ordinal/sadness/lr_5e-5_model3.csv')
