Welcome to our frequently asked questions page. We're glad that you're using Yellowbrick! If your question is not captured here, please submit it to our Google Groups Listserv. This is an email list/forum that you, as a Yellowbrick user, can join and interact with other users to address and troubleshoot Yellowbrick issues. The Google Groups Listserv is where you should be able to receive the quickest response. We would welcome and encourage you to join the group so that you can respond to others' questions! You can also ask questions on Stack Overflow and tag them with "yellowbrick". Finally, you can add issues on GitHub and you can tweet or direct message us on Twitter @scikit_yb.
You can change the size of a plot by passing in the desired dimensions in pixels on instantiation of the visualizer:
# Import the visualizer
from yellowbrick.features import RadViz
# Instantiate the visualizer using the ``size`` param
visualizer = RadViz(
classes=classes, features=features, size=(1080, 720)
)
...Note: we are considering adding support for passing in size in inches in a future Yellowbrick release. For a convenient inch-to-pixel converter, check out www.unitconversion.org.
You can change the title of a plot by passing in the desired title as a string on instantiation:
from yellowbrick.classifier import ROCAUC
from sklearn.linear_model import RidgeClassifier
# Create your custom title
my_title = "ROCAUC Curves for Multiclass RidgeClassifier"
# Instantiate the visualizer passing the custom title
visualizer = ROCAUC(
RidgeClassifier(), classes=classes, title=my_title
)
...To customize coloring in your plot, use the colors or cmap (or colormap) arguments. Note that different visualizers may require slightly different arguments depending on how they construct the plots.
For instance, the :doc:`api/features/manifold` accepts a colors argument, for which discrete targets should be the name of one of the :doc:`api/palettes` or a list of matplotlib colors represented as strings:
For instance, the :doc:`api/features/manifold` accepts a colors argument, for which discrete targets should be the name of a palette from the Yellowbrick :doc:`api/palettes` or a list of matplotlib colors represented as strings:
from yellowbrick.features.manifold import Manifold
visualizer = Manifold(
manifold="tsne", target="discrete", colors=["teal", "orchid"]
)
...... whereas for continuous targets, colors should be a colormap:
from yellowbrick.features.manifold import Manifold
visualizer = Manifold(
manifold="isomap", target="continuous", colors="YlOrRd"
)
...Other visualizers accept a cmap argument:
from sklearn.linear_model import LogisticRegression
from yellowbrick.classifier import ConfusionMatrix
visualizer = ConfusionMatrix(
LogisticRegression(), cmap="YlGnBu"
)
...Or a colormap argument:
from yellowbrick.features import ParallelCoordinates
# Instantiate the visualizer
visualizer = ParallelCoordinates(
classes=classes, features=features, colormap="PRGn"
)
...The :doc:`api/regressor/residuals` accepts color argument for the training and test points, train_color and test_color, respectively:
from yellowbrick.regressor import ResidualsPlot
from sklearn.linear_model import ElasticNet
visualizer = ResidualsPlot(
model=ElasticNet()
train_color=train_color, # color of points model was trained on
test_color=train_color, # color of points model was tested on
line_color=line_color # color of zero-error line
)Save your Yellowbrick plot by passing an outpath into poof():
from sklearn.cluster import MiniBatchKMeans
from yellowbrick.cluster import KElbowVisualizer
visualizer = KElbowVisualizer(MiniBatchKMeans(), k=(4,12))
visualizer.fit(X)
visualizer.poof(outpath="kelbow_minibatchkmeans.png")
...Most backends support png, pdf, ps, eps and svg to save your work!
You can use the alpha param to change the opacity of plotted points (where alpha=1 is complete opacity, and alpha=0 is complete transparency):
from yellowbrick.contrib.scatter import ScatterVisualizer
visualizer = ScatterVisualizer(
x="light", y="C02", classes=classes, alpha=0.5
)Visit the :doc:`api/datasets` page.