def main(argv):
if len(argv) != 4:
print("Usage: python3 get_genre.py audiopath low medium high")
exit()
#le = LabelEncoder().fit(GENRES)
# ------------------------------- #
## LOAD TRAINED GENRENET MODEL
net = genreNet()
net.load_state_dict(torch.load(MODELPATH, map_location='cpu'))
# ------------------------------- #
## LOAD AUDIO
audio_path = argv[0]
y_val, sr = load(audio_path, mono=True, sr=22050)
y = y_val[sr * 30:sr * 60]
# ------------------------------- #
## GET CHUNKS OF AUDIO SPECTROGRAMS
S = melspectrogram(y, sr).T
S = S[:-1 * (S.shape[0] % 128)]
num_chunk = S.shape[0] / 128
data_chunks = np.split(S, num_chunk)
# ------------------------------- #
## CLASSIFY SPECTROGRAMS
genres = list()
for i, data in enumerate(data_chunks):
data = torch.FloatTensor(data).view(1, 1, 128, 128)
preds = net(data)
pred_val, pred_index = preds.max(1)
pred_index = pred_index.data.numpy()[0]
pred_val = np.exp(pred_val.data.numpy()[0])
#pred_genre = le.inverse_transform(pred_index)
if pred_val >= 0.5:
genres.append(pred_index)
# ------------------------------- #
#s = float(sum([v for k,v in dict(Counter(genres)).items()]))
#pos_genre = sorted([(k, v/s*100 ) for k,v in dict(Counter(genres)).items()], key=lambda x:x[1], reverse=True)
#for genre, pos in pos_genre:
#print("%10s: \t%.2f\t%%" % (genre, pos))
genre_index = str(np.bincount(genres).argmax())
low = argv[1]
medium = argv[2]
high = argv[3]
if genre_index in low:
print("Applying Low Effect!")
pitch_shift(y_val, sr)
elif genre_index in medium:
print("Applying Medium Effect!")
bass_operation(y_val, sr)
elif genre_index in high:
print("Applying High Effect")
print("high")
overlay(y_val, sr)
return
def main(argv):
if len(argv) != 1:
print("Usage: python3 get_genre.py audiopath")
exit()
le = LabelEncoder().fit(GENRES)
# ------------------------------- #
## LOAD TRAINED GENRENET MODEL
net = genreNet()
net.load_state_dict(torch.load(MODELPATH, map_location='cuda:0'))
# ------------------------------- #
## LOAD AUDIO
audio_path = argv[0]
y, sr = load(audio_path, mono=True, sr=22050)
# ------------------------------- #
## GET CHUNKS OF AUDIO SPECTROGRAMS
S = melspectrogram(y, sr).T
S = S[:-1 * (S.shape[0] % 128)]
num_chunk = S.shape[0] / 128
data_chunks = np.split(S, num_chunk)
# ------------------------------- #
## CLASSIFY SPECTROGRAMS
genres = []
for i, data in enumerate(data_chunks):
data = torch.FloatTensor(data).view(1, 1, 128, 128)
preds = net(data)
pred_val, pred_index = preds.max(1)
pred_index = pred_index.data.numpy()[0]
pred_val = np.exp(pred_val.data.numpy()[0])
pred_genre = le.inverse_transform([pred_index])[0]
#if pred_val >= 0.5:
genres.append(pred_genre)
# ------------------------------- #
s = float(sum([v for k, v in dict(Counter(genres)).items()]))
#pos_genre = sorted([(k, v/s ) for k,v in dict(Counter(genres)).items()], key=lambda x:x[1], reverse=True)
pos_genre = [(k, v / s) for k, v in dict(Counter(genres)).items()]
result = dict.fromkeys(GENRES, 0)
for genre, pos in pos_genre:
result[genre] = pos
str_out = re.sub('.mp3', '', os.path.basename(audio_path))
for genre in GENRES:
str_out = str_out + ' ' + str(result[genre])
print(str_out)
return
def main(argv):
if len(argv) != 1:
print("Usage: python3 get_genre.py audiopath")
exit()
le = LabelEncoder().fit(GENRES)
# ------------------------------- #
## LOAD TRAINED GENRENET MODEL
net = genreNet()
net.load_state_dict(torch.load(MODELPATH, map_location='cpu'))
# ------------------------------- #
## LOAD AUDIO
audio_path = argv[0]
y, sr = load(audio_path, mono=True, sr=22050)
# ------------------------------- #
## GET CHUNKS OF AUDIO SPECTROGRAMS
S = melspectrogram(y, sr).T
S = S[:-1 * (S.shape[0] % 128)]
num_chunk = S.shape[0] / 128
data_chunks = np.split(S, num_chunk)
# ------------------------------- #
## CLASSIFY SPECTROGRAMS
genres = list()
for i, data in enumerate(data_chunks):
data = torch.FloatTensor(data).view(1, 1, 128, 128)
preds = net(data)
pred_val, pred_index = preds.max(1)
pred_index = pred_index.data.numpy()
pred_val = np.exp(pred_val.data.numpy()[0])
pred_genre = le.inverse_transform(pred_index).item()
if pred_val >= 0.5:
genres.append(pred_genre)
# ------------------------------- #
s = float(sum([v for k, v in dict(Counter(genres)).items()]))
pos_genre = sorted([(k, v / s * 100)
for k, v in dict(Counter(genres)).items()],
key=lambda x: x[1],
reverse=True)
for genre, pos in pos_genre:
print("%10s: \t%.2f\t%%" % (genre, pos))
return
def main(argv):
if len(argv) != 1:
print("Usage: python3 get_genre.py audiopath")
exit()
le = LabelEncoder().fit(GENRES)
# ------------------------------- #
## LOAD TRAINED GENRENET MODEL
net = genreNet()
net.load_state_dict(torch.load(MODELPATH, map_location='cpu'))
# ------------------------------- #
## LOAD AUDIO
audio_path = argv[0]
y, sr = load(audio_path, mono=True, sr=22050)
# ------------------------------- #
## GET CHUNKS OF AUDIO SPECTROGRAMS
S = melspectrogram(y, sr).T
S = S[:-1 * (S.shape[0] % 128)]
num_chunk = S.shape[0] / 128
data_chunks = np.split(S, num_chunk)
# ------------------------------- #
## CLASSIFY SPECTROGRAMS
genres = list()
added = 0
total = 0
for i, data in enumerate(data_chunks):
data = torch.FloatTensor(data).view(1, 1, 128, 128)
preds = net(data)
pred_val = preds.item()
print(pred_val)
added += pred_val
total += 1
# ------------------------------- #
print('Predicted: {}'.format((added/total) + 1960))
s = float(sum([v for k,v in dict(Counter(genres)).items()]))
pos_genre = sorted([(k, v/s*100 ) for k,v in dict(Counter(genres)).items()], key=lambda x:x[1], reverse=True)
print('Should output something.')
for genre, pos in pos_genre:
print("%10s: \t%.2f\t%%" % (genre, pos))
return
def main():
# ------------------------------------------------------------------------------------------- #
## DATA
data = Data(GENRES, DATAPATH)
data.make_raw_data()
data.save()
data = Data(GENRES, DATAPATH)
data.load()
# ------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------- #
## SET
set_ = Set(data)
set_.make_dataset()
set_.save()
set_ = Set(data)
set_.load()
x_train, y_train = set_.get_train_set()
x_valid, y_valid = set_.get_valid_set()
x_test, y_test = set_.get_test_set()
# ------------------------------------------------------------------------------------------- #
TRAIN_SIZE = len(x_train)
VALID_SIZE = len(x_valid)
TEST_SIZE = len(x_test)
net = genreNet()
net.cuda()
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(net.parameters(), lr=1e-3, weight_decay=1e-4)
EPOCH_NUM = 200
BATCH_SIZE = 16
tra_loss = []
tra_acc = []
val_acc = []
val_loss = []
for epoch in range(EPOCH_NUM):
inp_train, out_train = Variable(
torch.from_numpy(x_train)).float().cuda(), Variable(
torch.from_numpy(y_train)).long().cuda()
inp_valid, out_valid = Variable(
torch.from_numpy(x_valid)).float().cuda(), Variable(
torch.from_numpy(y_valid)).long().cuda()
# ------------------------------------------------------------------------------------------------- #
## TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE #
# ------------------------------------------------------------------------------------------------- #
train_loss = 0
optimizer.zero_grad() # <-- OPTIMIZER
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
x_train_batch, y_train_batch = inp_train[
i:i + BATCH_SIZE], out_train[i:i + BATCH_SIZE]
pred_train_batch = net(x_train_batch)
loss_train_batch = criterion(pred_train_batch, y_train_batch)
train_loss += loss_train_batch.data.cpu().numpy()
loss_train_batch.backward()
optimizer.step() # <-- OPTIMIZER
epoch_train_loss = (train_loss * BATCH_SIZE) / TRAIN_SIZE
tra_loss.append(train_loss)
train_sum = 0
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
pred_train = net(inp_train[i:i + BATCH_SIZE])
indices_train = pred_train.max(1)[1]
train_sum += (indices_train == out_train[i:i + BATCH_SIZE]
).sum().data.cpu().numpy()
train_accuracy = train_sum / float(TRAIN_SIZE)
tra_acc.append(train_accuracy)
# ------------------------------------------------------------------------------------------------- #
## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE #
# ------------------------------------------------------------------------------------------------- #
valid_loss = 0
for i in range(0, VALID_SIZE, BATCH_SIZE):
x_valid_batch, y_valid_batch = inp_valid[
i:i + BATCH_SIZE], out_valid[i:i + BATCH_SIZE]
pred_valid_batch = net(x_valid_batch)
loss_valid_batch = criterion(pred_valid_batch, y_valid_batch)
valid_loss += loss_valid_batch.data.cpu().numpy()
epoch_valid_loss = (valid_loss * BATCH_SIZE) / VALID_SIZE
val_loss.append(epoch_valid_loss)
valid_sum = 0
for i in range(0, VALID_SIZE, BATCH_SIZE):
pred_valid = net(inp_valid[i:i + BATCH_SIZE])
indices_valid = pred_valid.max(1)[1]
valid_sum += (indices_valid == out_valid[i:i + BATCH_SIZE]
).sum().data.cpu().numpy()
valid_accuracy = valid_sum / float(VALID_SIZE)
val_acc.append(valid_accuracy)
print("Epoch: %d\t\tTrain loss : %.2f\t\tValid loss : %.2f\t\tTrain acc : %.2f\t\tValid acc : %.2f" % \
(epoch + 1, epoch_train_loss, epoch_valid_loss, train_accuracy, valid_accuracy))
# ------------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------------- #
## SAVE GENRENET MODEL
# ------------------------------------------------------------------------------------------------- #
torch.save(net.state_dict(), MODELPATH)
print("-> ptorch model is saved")
# ------------------------------------------------------------------------------------------------- #
save_dir = "../utils/"
plt.plot(tra_loss)
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.title("Training Loss")
plt.tight_layout()
plt.savefig("./evaluate.png", format="png", bbox_inches="tight")
plt.close()
plt.plot(tra_acc)
plt.xlabel("Epochs")
plt.ylabel("Accuracy")
plt.title("Training Accuracy")
plt.tight_layout()
plt.savefig("./tra_acc.png", format="png", bbox_inches="tight")
plt.close()
plt.plot(val_loss)
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.title("Validation Loss")
plt.tight_layout()
plt.savefig("./val_loss.png", format="png", bbox_inches="tight")
plt.close()
plt.plot(val_acc)
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.title("Validation Accuracy")
plt.tight_layout()
plt.savefig("./val_acc.png", format="png", bbox_inches="tight")
plt.close()
# ------------------------------------------------------------------------------------------------- #
## EVALUATE TEST ACCURACY
# ------------------------------------------------------------------------------------------------- #
test_acc = []
inp_test, out_test = Variable(
torch.from_numpy(x_test)).float().cuda(), Variable(
torch.from_numpy(y_test)).long().cuda()
test_sum = 0
for i in range(0, TEST_SIZE, BATCH_SIZE):
pred_test = net(inp_test[i:i + BATCH_SIZE])
indices_test = pred_test.max(1)[1]
test_sum += (
indices_test == out_test[i:i +
BATCH_SIZE]).sum().data.cpu().numpy()
test_acc.append(test_sum / float(TEST_SIZE))
test_accuracy = test_sum / float(TEST_SIZE)
print("Test acc: %.2f" % test_accuracy)
# ------------------------------------------------------------------------------------------------- #
plt.plot(test_acc)
plt.xlabel("Epochs")
plt.ylabel("Accuracy")
plt.title("Testing Accuracy")
plt.tight_layout()
plt.savefig("./test_acc.png", format="png", bbox_inches="tight")
plt.close()
return
def main():
# ------------------------------------------------------------------------------------------- #
## DATA
#data = Data_rank(RANKS, DATAPATH)
#data.make_raw_data() # computing features
#data.make_data_sets()
# ------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------- #
## SET
data = Data_rank(RANKS, DATAPATH)
set_ = Set(data)
set_.chunk_size =160 # files
set_.make_chunks()
#x_train, y_train = set_.get_train_set()
#TRAIN_SIZE = len(x_train)
# # ------------------------------------------------------------------------------------------- #
#rates = [5e-1,5e-2,5e-3,5e-4,5e-5]
EPOCH_NUM = 250
BATCH_SIZE = 64
# load check point if existing
point = 0
##
net = genreNet()
#optimizer = torch.optim.RMSprop(net.parameters(),lr=1e-2,momentum=0.9,weight_decay=1e-5)
optimizer = torch.optim.Adam(net.parameters(),lr=1e-2, weight_decay=1e-5)
#scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer,lr_lambda=[EPOCH_NUM,0.95])
#scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.5)
if not os.path.isfile(MODELCHECKPOINT):
print("Starting from zero point")
point = -1
else:
checkpoint = torch.load(MODELCHECKPOINT)
net = checkpoint['model']
optimizer = torch.optim.Adam(net.parameters())
point = checkpoint['epoch']
net.load_state_dict(checkpoint['state_dict'])
#scheduler.load_state_dict(checkpoint['scheduler'])
print("Starting from the checkpoint " + MODELCHECKPOINT + ' from epoch: ' + str(point))
criterion = torch.nn.CrossEntropyLoss()
net.cuda()
#
for epoch in range(EPOCH_NUM):
if epoch > point:
c_iter = 0
num_chunks = len(set_.train_chunks)
for chunk in set_.train_chunks:
x_train, y_train = set_.get_set(chunk)
if x_train is not None:
TRAIN_SIZE = len(x_train)
inp_, out_ = Variable(torch.from_numpy(x_train)).float().cuda(), Variable(torch.from_numpy(y_train)).long().cuda()
# ------------------------------------------------------------------------------------------------- #
train_loss = 0
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
optimizer.zero_grad() # <-- OPTIMIZER
x_train_batch, y_train_batch = inp_[i:i + BATCH_SIZE], out_[i:i + BATCH_SIZE]
pred_train_batch = net(x_train_batch)
loss_train_batch = criterion(pred_train_batch, y_train_batch)
train_loss += loss_train_batch.data.cpu().item()
loss_train_batch.backward()
#clip_grad_norm_(net.parameters(), 5)
optimizer.step() # <-- OPTIMIZER
epoch_train_loss = train_loss / (TRAIN_SIZE/BATCH_SIZE)
train_sum = 0
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
pred_train = net(inp_[i:i + BATCH_SIZE])
indices_train = pred_train.max(1)[1]
train_sum += (indices_train == out_[i:i + BATCH_SIZE]).sum().data.cpu().item()
train_accuracy = train_sum / float(TRAIN_SIZE)
print("Epoch: %d\t\tIter: %d/%d\t\tTrain loss : %.2f\t\tTrain accuracy: %.2f" % \
(epoch + 1, c_iter+1, num_chunks, epoch_train_loss, train_accuracy))
c_iter+=1
#
del inp_, out_
#scheduler.step()
# ------------------------------------------------------------------------------------------------- #
## SAVE checkpoint
# ------------------------------------------------------------------------------------------------- #
checkpoint = {'model': genreNet(),
'state_dict': net.state_dict(),
'optimizer' : optimizer.state_dict(),
#'scheduler' : scheduler.state_dict(),
'epoch' : epoch}
torch.save(checkpoint, MODELCHECKPOINT)
print("Saving checkpoint for epoch " + str(epoch))
# ------------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------------- #
## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE #
# ------------------------------------------------------------------------------------------------- #
torch.cuda.empty_cache()
valid_loss = 0
valid_sum = 0
for v_chunk in set_.valid_chunks:
x_valid, y_valid = set_.get_set(v_chunk)
if x_valid is not None:
VALID_SIZE = len(x_valid)
inp_, out_ = Variable(torch.from_numpy(x_valid)).float().cuda(), Variable(torch.from_numpy(y_valid)).long().cuda()
for i in range(0, VALID_SIZE, BATCH_SIZE):
x_valid_batch, y_valid_batch = inp_[i:i + BATCH_SIZE], out_[i:i + BATCH_SIZE]
pred_valid_batch = net(x_valid_batch)
loss_valid_batch = criterion(pred_valid_batch, y_valid_batch)
valid_loss += loss_valid_batch.data.cpu().item()
for i in range(0, VALID_SIZE, BATCH_SIZE):
pred_valid = net(inp_[i:i + BATCH_SIZE])
indices_valid = pred_valid.max(1)[1]
valid_sum += (indices_valid == out_[i:i + BATCH_SIZE]).sum().data.cpu().item()
del inp_, out_
valid_accuracy = valid_sum / float(VALID_SIZE)
epoch_valid_loss = (valid_loss * BATCH_SIZE) / VALID_SIZE
print("Epoch: %d\t\tTrain loss : %.2f\t\tValid loss : %.2f\t\tTrain acc : %.2f\t\tValid acc : %.2f" % \
(epoch + 1, epoch_train_loss, epoch_valid_loss, train_accuracy, valid_accuracy))
# ------------------------------------------------------------------------------------------------- #
torch.save(net.state_dict(), MODELPATH)
print('-> ptorch model is saved.')
# ------------------------------------------------------------------------------------------------- #
## EVALUATE TEST ACCURACY
# ------------------------------------------------------------------------------------------------- #
torch.cuda.empty_cache()
test_sum = 0
for t_chunk in set_.test_chunks:
x_test, y_test = set_.get_set(t_chunk)
if x_test is not None:
TEST_SIZE = len(x_test)
inp_, out_ = Variable(torch.from_numpy(x_test)).float().cuda(), Variable(torch.from_numpy(y_test)).long().cuda()
for i in range(0, TEST_SIZE, BATCH_SIZE):
pred_test = net(inp_[i:i + BATCH_SIZE])
indices_test = pred_test.max(1)[1]
test_sum += (indices_test == out_[i:i + BATCH_SIZE]).sum().data.cpu().item()
del inp_, out_
test_accuracy = test_sum / float(TEST_SIZE)
print("Test acc: %.2f" % test_accuracy)
# ------------------------------------------------------------------------------------------------- #
return
def main():
# ------------------------------------------------------------------------------------------- #
## DATA
data = Data(GENRES, DATAPATH)
#data.make_raw_data()
#data.save()
data = Data(GENRES, DATAPATH)
data.load()
# ------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------- #
## SET
set_ = Set(data)
set_.make_dataset()
set_.save()
set_ = Set(data)
set_.load()
x_train, y_train = set_.get_train_set()
x_valid, y_valid = set_.get_valid_set()
x_test, y_test = set_.get_test_set()
# ------------------------------------------------------------------------------------------- #
TRAIN_SIZE = len(x_train)
VALID_SIZE = len(x_valid)
TEST_SIZE = len(x_test)
net = genreNet()
net.cuda()
#criterion = torch.nn.CrossEntropyLoss()
criterion = torch.nn.L1Loss()
optimizer = torch.optim.RMSprop(net.parameters(), lr=1e-4)
EPOCH_NUM = 100 #250
BATCH_SIZE = 16 #16
for epoch in range(EPOCH_NUM):
inp_train, out_train = Variable(torch.from_numpy(x_train)).float().cuda(), Variable(torch.from_numpy(y_train)).long().cuda()
inp_valid, out_valid = Variable(torch.from_numpy(x_valid)).float().cuda(), Variable(torch.from_numpy(y_valid)).long().cuda()
# ------------------------------------------------------------------------------------------------- #
## TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE # TRAIN PHASE #
# ------------------------------------------------------------------------------------------------- #
train_loss = 0
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
optimizer.zero_grad() # <-- OPTIMIZER
x_train_batch, y_train_batch = inp_train[i:i + BATCH_SIZE], out_train[i:i + BATCH_SIZE]
pred_train_batch = net(x_train_batch)
loss_train_batch = criterion(pred_train_batch, y_train_batch)
train_loss += loss_train_batch.data.cpu().numpy() #[0]
loss_train_batch.backward()
optimizer.step() # <-- OPTIMIZER
epoch_train_loss = (train_loss * BATCH_SIZE) / TRAIN_SIZE
train_sum = 0
for i in range(0, TRAIN_SIZE, BATCH_SIZE):
pred_train = net(inp_train[i:i + BATCH_SIZE])
indices_train = pred_train.max(0)[0] #(1)[1]
train_sum += (indices_train == out_train[i:i + BATCH_SIZE]).sum().data.cpu().numpy() #[0]
train_accuracy = train_sum / float(TRAIN_SIZE)
# ------------------------------------------------------------------------------------------------- #
## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE ## VALIDATION PHASE #
# ------------------------------------------------------------------------------------------------- #
valid_loss = 0
for i in range(0, VALID_SIZE, BATCH_SIZE):
x_valid_batch, y_valid_batch = inp_valid[i:i + BATCH_SIZE], out_valid[i:i + BATCH_SIZE]
pred_valid_batch = net(x_valid_batch)
loss_valid_batch = criterion(pred_valid_batch, y_valid_batch)
valid_loss += loss_valid_batch.data.cpu().numpy() #[0]
epoch_valid_loss = (valid_loss * BATCH_SIZE) / VALID_SIZE
valid_sum = 0
for i in range(0, VALID_SIZE, BATCH_SIZE):
pred_valid = net(inp_valid[i:i + BATCH_SIZE])
indices_valid = pred_valid.max(0)[0] # 1 1
valid_sum += (indices_valid == out_valid[i:i + BATCH_SIZE]).sum().data.cpu().numpy() #[0]
valid_accuracy = valid_sum / float(VALID_SIZE)
print("Epoch: %d\t\tTrain loss : %.2f\t\tValid loss : %.2f\t\tTrain acc : %.2f\t\tValid acc : %.2f" % \
(epoch + 1, epoch_train_loss, epoch_valid_loss, train_accuracy, valid_accuracy))
# ------------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------------- #
## SAVE GENRENET MODEL
# ------------------------------------------------------------------------------------------------- #
torch.save(net.state_dict(), MODELPATH)
print('-> ptorch model is saved.')
# ------------------------------------------------------------------------------------------------- #
# ------------------------------------------------------------------------------------------------- #
## EVALUATE TEST ACCURACY
# ------------------------------------------------------------------------------------------------- #
inp_test, out_test = Variable(torch.from_numpy(x_test)).float().cuda(), Variable(torch.from_numpy(y_test)).long().cuda()
test_sum = 0
for i in range(0, TEST_SIZE, BATCH_SIZE):
pred_test = net(inp_test[i:i + BATCH_SIZE])
indices_test = pred_test.max(0)[0] # [0][0]
test_sum += (indices_test == out_test[i:i + BATCH_SIZE]).sum().data.cpu().numpy() #[0]
test_accuracy = test_sum / float(TEST_SIZE)
print("Test acc: %.2f" % test_accuracy)
# ------------------------------------------------------------------------------------------------- #
return
