Econometric is the study of time series data
Feature engineering with Rossman dataset
Converting a timeseries problem into Tabular data form to do regression on it
Preprocessor
they are functions which are applied to training/test set
for example - Categorising the string values of a column.
FillMissing is another preprocessor
Normalize the data (substract the mean and divide by standard deviation)
Dropout is one of a kind of Regularization
We do regularization to avoid overfitting
What dropout does is - it skips some activations randomly from the hidden layers
Batch normalization
With this - we can have higher learning rate with a surety that we will still be able to reach global minimum.
Data Augmentation
another kind of regularization. There are many data augmentation techniques in fastai library - have a look at the documentation. Using a single image - and then creating different versions of it is also another interpretation that i have right now
increasing following of images
brightness
contrast
dihedral (flipping images)
warping (perspective of camera)
you can do data augmentation with NLP also
We will now dive deep inside CNN. We will now understand.
Below is a heatmap that shows that the colored part are the deciding factors that let CNN to decide if its a picture of cat or dog
