omarsar

## importing dataset
from google.colab import drive
drive.mount('gdrive', force_remount=True)

omarsar

## configure root folder on your gdrive
data_dir = 'gdrive/My Drive/DAIR RESOURCES/TF to PT/datasets/hymenoptera_data'

## custom transformer to flatten the image tensors
class ReshapeTransform:
    def __init__(self, new_size):
        self.new_size = new_size

    def __call__(self, img):
        result = torch.reshape(img, self.new_size)

omarsar

## load the entire dataset
x, y = next(iter(train_dataset))

## print one example
dim = x.shape[1]
print("Dimension of image:", x.shape, "\n", 
      "Dimension of labels", y.shape)

plt.imshow(x[160].reshape(1, 3, 224, 224).squeeze().T.numpy())

omarsar

class LR(nn.Module):
    def __init__(self, dim, lr=torch.scalar_tensor(0.01)):
        super(LR, self).__init__()
        # intialize parameters
        self.w = torch.zeros(dim, 1, dtype=torch.float).to(device)
        self.b = torch.scalar_tensor(0).to(device)
        self.grads = {"dw": torch.zeros(dim, 1, dtype=torch.float).to(device),
                      "db": torch.scalar_tensor(0).to(device)}
        self.lr = lr.to(device)

omarsar

## model pretesting
x, y = next(iter(train_dataset))

## flatten/transform the data
x_flatten = x.T
y = y.unsqueeze(0) 

## num_px is the dimension of the images
dim = x_flatten.shape[0]

omarsar

## hyperparams
costs = []
dim = x_flatten.shape[0]
learning_rate = torch.scalar_tensor(0.0001).to(device)
num_iterations = 100
lrmodel = LR(dim, learning_rate)
lrmodel.to(device)

## transform the data
def transform_data(x, y):

omarsar

## the trend in the context of loss
plt.plot(costs)
plt.show()

omarsar

Cost after iteration 0: 0.6931470036506653 | Train Acc: 50.40983581542969 | Test Acc: 45.75163269042969
Cost after iteration 10: 0.6691471934318542 | Train Acc: 64.3442611694336 | Test Acc: 54.24836730957031
Cost after iteration 20: 0.6513187885284424 | Train Acc: 68.44261932373047 | Test Acc: 54.24836730957031
Cost after iteration 30: 0.6367831230163574 | Train Acc: 68.03278350830078 | Test Acc: 54.24836730957031
Cost after iteration 40: 0.6245343685150146 | Train Acc: 69.67213439941406 | Test Acc: 54.90196228027344
Cost after iteration 50: 0.6139233112335205 | Train Acc: 70.90164184570312 | Test Acc: 56.20914840698242
Cost after iteration 60: 0.6045243740081787 | Train Acc: 72.54098510742188 | Test Acc: 56.86274337768555
Cost after iteration 70: 0.5960519909858704 | Train Acc: 74.18032836914062 | Test Acc: 57.51633834838867
Cost after iteration 80: 0.5883094668388367 | Train Acc: 73.77049255371094 | Test Acc: 57.51633834838867
Cost after iteration 90: 0.581156849861145 | Train Acc: 74.59016418457031 | Test 

omarsar

.es(index=apa*,q=geoip.country_code2:FR,metric=sum:bytes).label("France").divide(.es(index=apa*, q=geoip.country_code2:,metric=sum:bytes)).multiply(100).yaxis(units="custom::%"), .es(index=apa,q=geoip.country_code2:DE,metric=sum:bytes).label("Germany",).divide(.es(index=apa*, q=geoip.country_code2:*,metric=sum:bytes)).multiply(100).yaxis(units="custom::%")

omarsar

Title

Applied Deep Learning for NLP Applications

Abstract

Natural language processing (NLP) has become an important field with interest from many important sectors that leverage modern deep learning methods for approaching several NLP problems and tasks such as text summarization, question answering, and sentiment classification, to name a few. In this tutorial, we will introduce several of the fundamental NLP techniques and more modern approaches (BERT, GTP-2, etc.) and show how they can be applied via transfer learning to approach many real-world NLP problems. We will focus on how to build an NLP pipeline using several open-source tools such as Transformers, Tokenizers, spaCy, TensorFlow, and PyTorch, among others. Then we will learn how to use the NLP model to search over documents based on semantic relationships. We will use open-source technologies such as BERT and Elasticsearch for this segment to build a proof of concept. In essence, the learner will take away the important theoretical pieces ne