print(ex)
print("Could't load Doc2Vec, BUILDING...")
doc2vec_model = generate_doc2vec(doc2vec_name, datasets)
# Step 3 Gather Features
print("FEATURIZING")
# Labels
# yTrain = torch.Tensor([y_val for y_val in dataset_A["sentiment"][0:200]])
yTrain = getSentimentArray(dataset_A["sentiment"], useSmall=None)
yTrain_list = torch.Tensor([y_val for y_val in dataset_A["sentiment"]])
yDev = getSentimentArray(dataset_B["sentiment"], useSmall=None)
yDev_list = torch.Tensor([y_val for y_val in dataset_B["sentiment"]])
# Get word2vec
xTrain = getFeatureList(getCleanReviews(dataset_A, useSmall=None), word2vec_model, NUM_FEATURES, num_splits=4)
xDev = getFeatureList(getCleanReviews(dataset_B, useSmall=None), word2vec_model, NUM_FEATURES, num_splits=4)
# Step 4 Training NN model
MODEL_PATH = "recurrent_model.pt"
doTraining = False
if doTraining:
print("TRAINING recurrent nn model")
r_model = nn_recurrent_model.RNN(input_size=NUM_FEATURES)
r_model.train(xTrain, yTrain, learning_rate = 0.01, epochs=4)
torch.save(r_model, MODEL_PATH)
else:
r_model = torch.load(MODEL_PATH)
# Step 5 Evaluate performance
print()
print("Could't load Doc2Vec, BUILDING...")
doc2vec_model = generate_doc2vec(doc2vec_name, datasets)
# Step 3 Gather Features
print("FEATURIZING")
# Labels
yTrain = torch.Tensor([y_val for y_val in dataset_A["sentiment"]])
yDev = torch.Tensor([y_val for y_val in dataset_B["sentiment"]])
# Get word2vec
# xTrain = getAvgFeatureVecs(getCleanReviews(dataset_A), word2vec_model, NUM_FEATURES)
# xTrain = torch.tensor(xTrain)
# Get doc2vec
xTrainDoc = getDocFeatureVec(getCleanReviews(dataset_A), doc2vec_model,
NUM_FEATURES)
xTrainDoc = torch.tensor(xTrainDoc)
# Step 4 Training NN model
MODEL_PATH = "feedforward_model.pt"
doTraining = False
if doTraining:
print("TRAINING nn model")
l_model = nn_model.NeuralNetwork(input_nodes=NUM_FEATURES)
l_model.train_model_persample(xTrainDoc, yTrain)
torch.save(l_model, MODEL_PATH)
else:
l_model = torch.load(MODEL_PATH)
# Step 5 Evaluate performance
header=0,
delimiter="\t",
quoting=3)
dataset_D = pd.read_csv("dataset/processed/D.tsv",
header=0,
delimiter="\t",
quoting=3)
dataset_E = pd.read_csv("dataset/processed/E.tsv",
header=0,
delimiter="\t",
quoting=3)
datasets = [dataset_A, dataset_D, dataset_E]
yTrain = torch.Tensor([y_val for y_val in dataset_A["sentiment"]])
yDev = torch.Tensor([y_val for y_val in dataset_B["sentiment"]])
xTrainRaw = getCleanReviews(dataset_A)
xDevRaw = getCleanReviews(dataset_B)
# Step 2 Define sweep range
evaluationRunSpecifications = []
# for num_features in [100, 150, 200, 250, 300]:
# for num_features in [350, 400, 450]:
# runSpecification = {}
# runSpecification['model'] = 'Standard Neural Network'
# runSpecification['num_features'] = num_features
# runSpecification['context'] = 5
# runSpecification['optimizing'] = 'num_features'
# evaluationRunSpecifications.append(runSpecification)
print(ex)
print("Could't load Doc2Vec, BUILDING...")
doc2vec_model = generate_doc2vec(doc2vec_name, datasets)
# Step 3 Gather Features
print("FEATURIZING")
# Labels
# yTrain = torch.Tensor([y_val for y_val in dataset_A["sentiment"][0:200]])
yTrain = getSentimentArray(dataset_A["sentiment"], useSmall=None)
yTrain_list = torch.Tensor([y_val for y_val in dataset_A["sentiment"]])
yDev = getSentimentArray(dataset_B["sentiment"], useSmall=None)
yDev_list = torch.Tensor([y_val for y_val in dataset_B["sentiment"]])
# Get word2vec
xTrain = getFeatureList(getCleanReviews(dataset_A, useSmall=None),
word2vec_model,
NUM_FEATURES,
num_splits=4)
xDev = getFeatureList(getCleanReviews(dataset_B, useSmall=None),
word2vec_model,
NUM_FEATURES,
num_splits=4)
# Step 4 Training NN model
MODEL_PATH = "recurrent_model.pt"
doTraining = True
if doTraining:
print("TRAINING recurrent nn model")
r_model = nn_recurrent_model.RNN(input_size=NUM_FEATURES)
r_model.train(xTrain, yTrain, learning_rate=0.01, epochs=4)