CraftDL
What is this?
CraftDL is a library for quickly solving simple & common deep learning tasks.
Requirements
You will need Python >= 3.9 and some recent version of pip (e.g. pip >= 20.0)
to use this library.
Installation
You can install CraftDL using the pip package manager:
pip install craftdlBasic example
Here is an example for how you might want to use CraftDL:
from craftdl import *
# Obtain the dataset
inputs, targets = circles_dataset(rs=[1, 3], ns=[50, 50])
# Create a train-test split
(inputs_train, targets_train), (inputs_test, targets_test) = train_test_split(inputs, targets)
# Plot the train and test datasets
plot_labels(inputs_train, targets_train)
plot_labels(inputs_test, targets_test)
# Create and fit a LinearClassificationNet
model = LinearClassificationNet(2, [6], 1)
losses = model.fit(inputs_train, targets_train, 100, lr=1.0)
# Plot the losses
plot_losses(losses)
# Get the accuracy on the test dataset
accuracy = model.accuracy(inputs_test, targets_test)
print(f"accuracy on test set={accuracy}")
# Get the predictions on the test dataset
predictions_test = model.predict(inputs_test)
# Show the decision boundary
x1_grid, x2_grid, predictions_grid = model.predict_grid(-4, 4, -4, 4)
plot_labels_with_decision_surface(
inputs_test, targets_test, x1_grid, x2_grid, predictions_grid
)You should get a bunch of images along with a low loss and a high accuracy on the test set.
The training dataset:
The testing dataset:
The loss plot:
The decision surface:
Matplotlib backend for plotting
The plotting facilities of CraftDL are built on top of matplotlib. Therefore you need
an appropriate matplotlib backend. If you encounter weird matplotlib errors when
trying to call a plotting function, you most probably don't have the appropriate backend
installed on selected.
Usually you want to use the TkAgg backend. You can use it like this:
import matplotlib
matplotlib.use("TkAgg")
