def train_lstm_anger():
vocab_size, trainds, valds, testds, traindl, valdl, testdl = load_uncased_data(
'anger')
final_stats_anger = []
for dropout in [0.1, 0.2, 0.3, 0.4, 0.5]:
for hidden_size in [25, 50, 75, 100, 125, 150, 175, 200]:
for lr in [1e-1, 1e-2, 1e-3, 1e-4, 1e-5, 1e-6]:
model = LSTMClassifier(
vocab_size,
200,
hidden_size,
4,
trainds.fields['Clean_Tweet'].vocab.vectors,
dropout=dropout,
bidirectional=False)
training_stats = train(model,
traindl,
valdl,
lr=lr,
hidden_size=hidden_size,
dropout=dropout)
final_stats_anger.extend(training_stats)
pd.DataFrame(final_stats_anger).to_csv(
'./data/models/lstm/results/anger.csv')
def train_bilstm_sadness():
vocab_size, trainds, valds, testds, traindl, valdl, testdl = load_uncased_data('sadness')
final_stats_sadness = []
for dropout in [0.1, 0.2, 0.3, 0.4, 0.5]:
for hidden_size in [32, 64, 96, 128]:
for lr in [1e-1, 1e-2, 1e-3, 1e-4, 1e-5, 1e-6]:
model = LSTMClassifier(
vocab_size,
200,
hidden_size,
4,
trainds.fields['Clean_Tweet'].vocab.vectors,
dropout=dropout,
bidirectional=False
)
training_stats = train(model, traindl, valdl, lr=lr, hidden_size=hidden_size, dropout=dropout)
final_stats_sadness.extend(training_stats)
pd.DataFrame(final_stats_sadness).to_csv('./data/models/bilstm/results/sadness.csv')