def image_grid(images=None,
data=None,
sort_by=None,
image_dir='',
image_size=None,
padding=None,
n_jobs=1,
**kwargs):
"""Create a grid of images.
Parameters
----------
images : str or array-like of shape [n_samples, width, height, channels], optional
Image array or name of the variable containing the image file
paths within `data`.
data : pandas.DataFrame, optional
Tidy ("long-form") dataframe where each column is a variable
and each row is an observation. If `images` is a variable name,
then it should be contained in `data`.
sort_by : str or array-like of shape [n_samples,], optional
Data or name of the variable to sort images by.
image_dir : str, optional
The location of the image files on disk. Images will
be loaded from files matching the pattern
'image_dir + os.path.sep + image_path'.
image_size : int, optional
The size of each image displayed in the scatter plot. Images
will be sampled to `image_size` if the size of the images
do not match `image_size`.
padding : int, optional
The padding between images in the grid.
n_jobs : int, optional
The number of parallel workers to use for loading
the image files when reading from disk. The default
uses a single core.
kwargs : key, value pairings
Additional keyword arguments are passed to the function used to draw
the plot on the Axes.
Returns
-------
ax : matplotlib Axes
Returns the Axes object with the plot for further tweaking.
See Also
--------
scatter_grid : Combines an image grid with a :func:`scatter_plot`.
Examples
--------
Create a image grid.
.. plot:: ../examples/image_grid.py
Create a image grid with custom ordering on MNIST.
.. plot:: ../examples/image_grid_mnist.py
"""
images = data_utils.get_images(data,
images,
as_image=True,
image_size=image_size,
n_jobs=n_jobs)
if sort_by is not None:
if sort_by in features.HSVFeatures.all_features():
hsv = features.extract_hsv_stats(images, n_jobs=n_jobs)
sort_by_values = hsv[:,
features.HSVFeatures.feature_index(sort_by)]
sorted_indices = np.argsort(sort_by_values)
images = [images[i] for i in sorted_indices]
else:
sort_by = data_utils.get_variable(data, sort_by)
images = [images[i] for i in np.argsort(sort_by)]
grid = images_to_grid(images, padding=padding)
return plots.pillow_to_matplotlib(grid, **kwargs)
def image_barplot(y,
images=None,
data=None,
sort_by=None,
bar_height=50,
image_dir='',
image_size=(40, 40),
n_jobs=1,
**kwargs):
"""Create a barplot where the bars are created from images in the dataset.
This visualization is useful for analyzing the relationship between images
and categorical variables.
Parameters
----------
y : str or array-like
Data or the name of a variables in `data`. This variable is assumed
to be categoricel, e.g. discrete, and is used to split the bars
on the vertical axis.
images : str or array-like of shape [n_samples, width, height, channels], optional
Image array or name of the variable containing the image file
paths within `data`.
data : pandas.DataFrame, optional
Tidy ("long-form") dataframe where each column is a variable
and each row is an observation. If `images` is a variable name,
then it should be contained in `data`.
sort_by : str or array-like of shape [n_samples,], optional
Data or name of the variable to sort images by in the horizontal
direction.
bar_height : int
The number of images placed in a single horizontal bar before
creating a new bar.
image_dir : str, optional
The location of the image files on disk. Images will
be loaded from files matching the pattern
'image_dir + os.path.sep + image_path'.
image_size : int, optional
The size of each image displayed in the scatter plot. Images
will be sampled to `image_size` if the size of the images
do not match `image_size`.
n_jobs : int, optional
The number of parallel workers to use for loading
the image files when reading from disk. The default
uses a single core.
kwargs : key, value pairings
Additional keyword arguments are passed to the function used to draw
the plot on the Axes.
Returns
-------
ax : matplotlib Axes
Returns the Axes object with the plot for further tweaking.
Examples
--------
Create a image barplot.
.. plot:: ../examples/image_barplot.py
"""
y = data_utils.get_variable(data, y)
images = data_utils.get_images(data,
images,
as_image=False,
image_size=image_size,
n_jobs=n_jobs)
if sort_by is not None:
if sort_by in features.HSVFeatures.all_features():
hsv = features.extract_hsv_stats(images, n_jobs=n_jobs)
sort_by_values = hsv[:,
features.HSVFeatures.feature_index(sort_by)]
images = images[np.argsort(sort_by_values)]
else:
sort_by = data_utils.get_variable(data, sort_by)
images = images[np.argsort(sort_by)]
return images_to_barplot(images, y, bar_height=bar_height, **kwargs)
def scatter_plot(x,
y,
images=None,
data=None,
hue=None,
image_dir='',
image_size=None,
threshold=None,
alpha=0.9,
color=None,
n_jobs=1,
**kwargs):
"""Create an image scatter plot based on columns `x` vs. `y`.
Parameters
----------
x, y : str or array-like
Data or names of variables in `data`. These variables are
used for the x and y axes respectively.
images : str or array-like, optional
Image arrays or names of the column pointing to the
image paths within `data`.
data : pandas.DataFrame, optional
Tidy ("long-form") dataframe where each column is a variable
and each row is an observation. If `images` is a variable name,
then it should be contained in `data`.
image_dir : str, optional
The location of the image files on disk. Images will
be loaded from files matching the pattern
'image_dir + os.path.sep + image_path'.
image_size : int, optional
The size of each image displayed in the scatter plot. Images
will be sampled to `image_size` if the size of the images
do not match `image_size`.
threshold : float, optional
In order to avoid clutter only one point in a ball of
radius `threshold` is displayed. Note that features
are re-scaled to lie on the unit square [0, 1] x [0, 1].
The default of None means all points are displayed.
alpha : float, optional
Alpha level used when displaying images.
n_jobs : int
The number of parallel jobs used to load the
images from disk.
Examples
--------
Create a scatter plot with hue labels.
.. plot:: ../examples/scatter_plot.py
"""
# get co-variates
x = data_utils.get_variable(data, x)
y = data_utils.get_variable(data, y)
# load images
images = data_utils.get_images(data,
images,
image_dir=image_dir,
as_image=False,
image_size=image_size,
n_jobs=n_jobs)
# TODO (seaborn is only required for a color palette. Remove this)
if hue is not None:
hue = data_utils.get_variable(data, hue)
values, value_map = np.unique(hue, return_inverse=True)
palette = sns.husl_palette(len(values))
images = [
features.color_image(img, hue=palette[val])
for img, val in zip(images, value_map)
]
elif color is not None:
images = [features.color_image(img, hue=color) for img in images]
return images_to_scatter(images,
x,
y,
threshold=threshold,
alpha=alpha,
**kwargs)
def image_histogram(x,
images=None,
data=None,
n_bins=None,
sort_by=features.HSVFeatures.SATURATION,
image_dir='',
image_size=None,
n_jobs=1,
**kwargs):
"""Create an univariate image histogram binned by the `x`
variable.
Parameters
----------
x : str or array-like of shape [n_samples,]
Data or names of variables in `data`.
images : str or array-like of shape [n_samples, width, height, channels], optional
Image array or name of the variable containing the image file
paths within `data`.
data : pandas.DataFrame, optional
Tidy ("long-form") dataframe where each column is a variable
and each row is an observation. If `images`, `x`, or `sort_by`
is a variable name, then it should be contained in `data`.
n_bins : int or None
Specification of the number of bins. If None, then the
Freedman-Diaconis estimator is used to determine the number of bins.
sort_by : str, HSVFeatures enum or array-like of shape [n_samples,], optional
Data or name of the variable to sort images by on the y-axis.
image_dir : str (default='')
The location of the image files on disk.
image_size : int
The size of each image in the scatter plot.
n_jobs : int (default=1)
The number of parallel workers to use for loading
the image files.
Returns
-------
ax : matplotlib Axes
Returns the Axes object with the plot for further tweaking.
Examples
--------
Create an image histogram.
.. plot:: ../examples/image_histogram.py
"""
images = data_utils.get_images(
data,
images,
image_dir=image_dir,
image_size=image_size,
index=None, #x.index,
as_image=False,
n_jobs=n_jobs)
x = data_utils.get_variable(data, x)
if sort_by is not None:
if sort_by in features.HSVFeatures.all_features():
hsv = features.extract_hsv_stats(images, n_jobs=n_jobs)
sort_by = hsv[:, features.HSVFeatures.feature_index(sort_by)]
else:
sort_by = data_utils.get_variable(data, sort_by)
#histo = images_to_histogram(images, x, n_bins=n_bins, sort_by=sort_by)
#return plots.pillow_to_matplotlib(histo, **kwargs)
return histogram_matplotlib(images,
x,
n_bins=n_bins,
sort_by=sort_by,
**kwargs)
def scatter_grid(x, y,
images=None,
data=None,
hue=None,
image_dir='',
image_size=None,
padding=None,
n_jobs=1,
**kwargs):
"""Draw a plot ordering images in a regularly spaced 2-d grid
based on their distance in the x-y plane. The distance between
points is assumed to be euclidean.
Parameters
----------
x, y : str or array-like
Data or names of variables in `data`.
These variables correspond to the x-y coordinates
in the euclidean space.
images : str or array-like
Image arrays or names of the column pointing to the
image paths within `data`.
data : pd.DataFrame
Pandas dataframe holding the dataset.
hue : str or array-like
Data or the name of the variable to use to color
the individual images on the grid.
image_dir : str (default='')
The location of the image files on disk.
image_size : int
The size of each image in the scatter plot.
padding : int, optional
The padding between images in the grid.
n_jobs : int (default=1)
The number of parallel workers to use for loading
the image files.
Returns
-------
A properly shaped NxWx3 image with any necessary padding.
Examples
--------
Create a grid plot with hue labels.
.. plot:: ../examples/scatter_grid.py
"""
x_var = data_utils.get_variable(data, x)
y_var = data_utils.get_variable(data, y)
# TODO (seaborn is only required for a color palette. Remove this)
if hue is not None:
images = data_utils.get_images(
data, images,
image_dir=image_dir,
as_image=False,
image_size=image_size,
n_jobs=n_jobs)
hue = data_utils.get_variable(data, hue)
values, value_map = np.unique(hue, return_inverse=True)
palette = sns.husl_palette(len(values))
images = [features.color_image(img, hue=palette[val]) for
img, val in zip(images, value_map)]
images = [image_io.to_pillow_image(img) for img in images]
else:
# load images
images = data_utils.get_images(
data, images,
image_dir=image_dir,
as_image=True,
image_size=image_size,
n_jobs=n_jobs)
return images_to_scatter_grid(images, x_var, y_var, padding=padding, **kwargs)