Can we use fastai for feature extraction?
Looks like it’s here: https://github.com/fastai/fastai/blob/master/docs_src/data_block.ipynb
worked for me - im using the dev version
is it possible to do multi label classification with a model trained on single labels?
Can we use DataBlock API with text DataSet?
If you mean getting the activations of a specific layer in a network yes, you can do this with hooks. You can see this thread or refer to the docs
Here the validation set is created by putting aside some part of the training set (20% randomly chosen). The test set doesn’t have labels (it’s a kaggle competition).
Is there a way to make the buttons in jupyter notebook when you hit shift+Tab not giant?
how to add test set after model was trained using data without test images?
Yes, very much so. You should get familiar with the process, but all steps come in a certain order to answer one specific problem. The block name comes from the fact you can use all the different strategies for labelling/putting aside a validation set/creating datasets/transforming plugged together like this.
Does this screw up the Shift+Tab-ability of these functions?
are these data augmentation features documented, which does what?
I get name 'Config' is not defined
when I try to run path = Config.data_path()/'planet'
Any idea why?
can i use custom define loss in fastai?
Not up to date with the library?
I think of DataBlocks as kind of like how we layered our models. All the piece gets stacked up in order:
nn.Sequential(
Flatten(),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(25088,256),
nn.ReLU(),
nn.BatchNorm1d(256),
nn.Dropout(0.5),
nn.Linear(256,4+len(cats)),
)
Is it possible to change the cost function in fastai?
Of course, you can plug it in learn.loss_func
.
Up until now I’ve heard the terms DataSet, DataBunch, and DataBlocks. Is there a doc somewhere that explains the difference between them?