def visualize(
self,
fig: plt.Figure,
X: np.ndarray,
energy: Callable[[np.ndarray], np.ndarray] = None,
):
"""
Visualize a set of samples, returning a figure.
Args:
X (tensor ~ (n_samples, n_features):
Returns:
Figure
"""
fig.clear()
ax = fig.subplots(1, 1)
if X.shape[1] == 1:
ax.hist(X.reshape([-1]), density=True, label="Data")
x_min = X.min()
x_max = X.max()
xs = np.linspace(x_min, x_max, 100)
if hasattr(self, "logpdf"):
ys_ = np.exp(self.logpdf(xs.reshape([-1, 1])))
ys = ys_.reshape(-1)
ax.plot(xs, ys, label="Actual")
if energy is not None:
ys_ = np.exp(-energy(xs.reshape([-1, 1])))
ys = ys_.reshape(-1)
Z = ys_.mean() * (x_max - x_min)
ax.plot(xs, ys / Z, label="Energy", color="red")
ax.legend()
elif X.shape[1] == 2:
ax.scatter(X[:, 0], X[:, 1], label="Data")
x_min, x_max = X[:, 0].min(), X[:, 0].max()
y_min, y_max = X[:, 1].min(), X[:, 1].max()
x_support = np.linspace(x_min, x_max, 100)
y_support = np.linspace(y_min, y_max, 100)
xx, yy = np.meshgrid(x_support, y_support)
XY = np.hstack([xx.reshape([-1, 1]), yy.reshape([-1, 1])])
if hasattr(self, "logpdf"):
z_ = np.exp(self.logpdf(XY))
z = z_.reshape(xx.shape)
ax.contour(xx, yy, z, 10)
if energy is not None:
z_ = np.exp(-energy(XY))
z = z_.reshape(xx.shape)
ax.contour(xx, yy, z, 10, cmap="Reds")
ax.legend()
else:
from sklearn.manifold import TSNE
tsne = TSNE(n_components=2)
emb = tsne.fit_transform(X)
ax.scatter(emb[:, 0], emb[:, 1])
def plot_image_samples(im_size: Tuple[int, ...],
binarize,
fig: plt.Figure,
X: np.ndarray,
energy=None):
"""Plots model samples on the given figure."""
fig.clear()
if binarize:
X = X > 0.5
n = min(int(sqrt(X.shape[0])), 7)
ax = fig.subplots(ncols=n, nrows=n)
for i in range(n):
for j in range(n):
ax[i, j].imshow(X[i + n * j, :].reshape(im_size),
cmap="gray",
vmin=-1,
vmax=1)
ax[i, j].axis("off")
def npsPlotByKey(pkg: NotePkg, fig:plt.Figure = None, shape: Tuple = None,
window=1000, stride=None, title=True, legend=True, barKwargs=None) -> plt.Figure:
""" This creates an NPS Plot with the axes.
:param pkg: Any Note Package
:param fig: The figure to plot on, if None, we use gcf()
:param shape: The shape of the axes to take. (rows, columns)
:param window: The window of the roll
:param stride: The stride length of the roll
:param title: Whether to show the key titles
:param legend: Whether to show legend. False to show none, True to show on first, 'all' to show on all
:param barKwargs: The kwargs to pass into plot()
"""
if fig is None: fig = plt.gcf()
if barKwargs is None: barKwargs = {}
keys = pkg.maxColumn() + 1 # This gives us the keys
if shape is None:
rows = keys
cols = 1
shape = (rows, cols)
else:
assert shape[0] * shape[1] >= keys, "Shape must be able to hold all keys."
ax: np.ndarray = fig.subplots(nrows=shape[0], ncols=shape[1],
sharex='all', sharey='all')
ax = ax.flatten()
for key in range(keys):
if legend == 'all':
npsPlot(pkg.inColumns([key]), ax=ax[key], window=window, stride=stride, legend=True, barKwargs=barKwargs)
elif legend is True and key == 0:
npsPlot(pkg.inColumns([key]), ax=ax[key], window=window, stride=stride, legend=True, barKwargs=barKwargs)
else:
npsPlot(pkg.inColumns([key]), ax=ax[key], window=window, stride=stride, legend=False, barKwargs=barKwargs)
ax: List[plt.Axes]
if title: ax[key].set_title(f"Key: {key}")
fig.tight_layout()
return fig
from matplotlib.backends.backend_tkagg import (
FigureCanvasTkAgg, # not FigureCanvasTk
NavigationToolbar2Tk # not NavigationToolbar2TkAgg
)
import tkinter as tk
# --- random data ---
import random
y = [random.randint(-100, 100) for x in range(200)]
# --- GUI ---
root = tk.Tk()
#fig = plt.Figure()
fig = Figure()
canvas = FigureCanvasTkAgg(fig, root)
canvas.get_tk_widget().pack()
navbar = NavigationToolbar2Tk(canvas, root)
#navbar.update() # not needed
#ax = fig.add_subplot(111)
ax = fig.subplots()
ax.plot(y)
root.mainloop()
