from attention_decoder import AttentionDecoderRNN
from encoder import EncoderRNN
from language import Language
from torch.autograd import Variable
import torch.nn as nn
import torch.optim as optim
import random
# Parse argument for input sentence
# parser = argparse.ArgumentParser()
# parser.add_argument('language')
# parser.add_argument('input')
# args = parser.parse_args()
language = 'spa-eng'
helpers.validate_language_params(language)
input_lang, output_lang, pairs = etl.prepare_data(language)
attn_model = 'general'
hidden_size = 500
n_layers = 2
dropout_p = 0.05
teacher_forcing_ratio = .5
clip = 5.
criterion = nn.NLLLoss()
# Initialize models
encoder = EncoderRNN(input_lang.n_words, hidden_size, n_layers)
decoder = AttentionDecoderRNN(attn_model,
hidden_size,
output_lang.n_words,
n_layers,
import numpy as np
from etl import prepare_data, prepare_submission
from sklearn.feature_selection import SelectFromModel
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import train_test_split
from xgboost import XGBRegressor
# load and split train/dev/test
(X_train, y_train), (X_test, test_id) = prepare_data()
X_train, X_dev, y_train, y_dev = train_test_split(X_train, y_train, test_size=0.2, random_state=69)
# run without hyperparams for fscore calculations
model = XGBRegressor()
model.fit(X_train, y_train)
y_hat = model.predict(X_dev)
mae = mean_absolute_error(np.expm1(y_dev), np.expm1(y_hat))
print("Mae: {}".format(mae))
thresholds = np.sort(model.feature_importances_)
thresholds = np.unique(thresholds)
threshold = 0
best_mae = mae
for thresh in thresholds[:50]:
selection = SelectFromModel(model, threshold=thresh, prefit=True)
select_X_train = selection.transform(X_train)
# train model
selection_model = XGBRegressor()
selection_model.fit(select_X_train, y_train)