def plotSortedProfile(original_series,
reconstructed_series,
x_label,
y_label,
scale="log",
name=None):
figure_name = "profile"
if name:
figure_name = name + "_" + figure_name
sort_idx = argsort(original_series)[::-1]
D = original_series.shape[0]
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
x = linspace(0, D, D)
axis.plot(x, original_series[sort_idx], zorder=1, label='Original')
axis.plot(x,
reconstructed_series[sort_idx],
zorder=0,
label='Reconstruction')
axis.legend()
axis.set_yscale(scale)
axis.set_xscale(scale)
axis.set_xlabel(x_label)
axis.set_ylabel(y_label)
data.saveFigure(figure, figure_name)
def plotKLdivergenceHeatmap(KL_all, name=None):
print("Plotting KL-divergence heatmap (activity of latent units).")
figure_name = "KL_divergence_heatmap"
if name:
figure_name = name + "/" + figure_name
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
KL_array = array(KL_all)
print("Dimensions of KL-activations:")
print(KL_array.shape)
seaborn.heatmap(log(KL_array.T),
xticklabels=True,
yticklabels=False,
cbar=True,
center=None,
square=True,
ax=axis)
axis.set_xlabel("Epoch")
axis.set_ylabel("$log KL(p_i||q_i)$")
data.saveFigure(figure, figure_name, no_spine=False)
def plotProfile(series,
x_label,
y_label,
scale="linear",
bar=False,
name=None):
figure_name = "profile"
if name:
figure_name = name + "_" + figure_name
D = series.shape[0]
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
x = linspace(0, D, D)
if bar:
axis.bar(x, series)
else:
axis.plot(x, series)
axis.set_yscale(scale)
axis.set_xlabel(x_label)
axis.set_ylabel(y_label)
data.saveFigure(figure, figure_name)
def plotHeatMap(data_set,
data_set_headers=None,
clusters=None,
center=None,
simple=False,
name=None):
figure_name = "heat_map"
if name:
figure_name = name + "_" + figure_name
if data_set_headers and clusters:
sorted_data_set = empty(data_set.shape)
N_seen = 0
for cluster_id, cluster in sorted(clusters.items()):
subset = []
for cell in cluster:
index = where(data_set_headers["cells"] == cell)[0]
if len(index) == 0:
continue
subset.append(int(index))
N_subset = len(subset)
if N_subset == 0:
continue
sorted_data_set[N_seen:(N_seen + N_subset)] = data_set[subset]
N_seen += N_subset
data_set = sorted_data_set[:N_seen]
figure_name += "_sorted"
N, M = data_set.shape
# figure = pyplot.figure(figsize = (N/500, M/500))
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.heatmap(data_set.T,
xticklabels=False,
yticklabels=False,
cbar=True,
square=True,
center=center,
ax=axis)
axis.set_xlabel("Cell")
axis.set_ylabel("Gene")
data.saveFigure(figure, figure_name, no_spine=False)
def plotLatentSpace(latent_set,
latent_set_headers=None,
clusters=None,
name=None):
figure_name = "latent_space"
if name:
figure_name = name + "/" + figure_name
N, M = latent_set.shape
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
if M > 2:
# TODO Try t-SNE.
pca = PCA(n_components=2)
pca.fit(latent_set)
latent_set = pca.transform(latent_set)
axis.set_xlabel("PC 1")
axis.set_ylabel("PC 2")
else:
axis.set_xlabel("z_1")
axis.set_ylabel("z_2")
for cluster_id in clusters:
cluster = clusters[cluster_id]
subset = []
for cell in cluster:
index = where(latent_set_headers["cells"] == cell)[0]
if len(index) == 0:
continue
subset.append(int(index))
if len(subset) == 0:
continue
axis.scatter(latent_set[subset, 0],
latent_set[subset, 1],
c=data.cluster_colours[cluster_id],
edgecolors=None,
label=cluster_id)
# axis.legend(loc="best")
data.saveFigure(figure, figure_name)
def plotLearningCurves(curves, name=None):
print("Plotting learning curves.")
figure_name = "learning_curves"
if name:
figure_name = name + "/" + figure_name
figure, (axis_1, axis_2) = pyplot.subplots(2,
sharex=True,
figsize=(6.4, 9.6))
for i, (curve_set_name, curve_set) in enumerate(sorted(curves.items())):
colour = palette[i]
for curve_name, curve in sorted(curve_set.items()):
if curve_name == "LB":
curve_name = "$\\mathcal{L}$"
line_style = "solid"
axis = axis_1
elif curve_name == "ENRE":
curve_name = "$\\log p(x|z)$"
line_style = "dashed"
axis = axis_1
elif curve_name == "KL":
line_style = "dashed"
curve_name = "$KL(p||q)$"
axis = axis_2
elif curve_name == "KL_all":
continue
epochs = arange(len(curve)) + 1
label = curve_name + " ({} set)".format(curve_set_name)
axis.plot(curve, color=colour, linestyle=line_style, label=label)
handles, labels = axis_1.get_legend_handles_labels()
labels, handles = zip(*sorted(zip(labels, handles), key=lambda t: t[0]))
axis_1.legend(handles, labels, loc="best")
axis_2.legend(loc="best")
# axis_1.set_xlabel("Epoch")
axis_2.set_xlabel("Epoch")
data.saveFigure(figure, figure_name)
def plotCountHistogram(data_set, k_min, k_max, name=None):
figure_name = "count_histogram"
if name:
figure_name = name + "_" + figure_name
k = linspace(0, k_max, k_max + 1)
C = empty(k_max + 1)
for i in range(k_max + 1):
if k[i] < k_max:
c = (data_set == k[i]).sum()
if k[i] == k_max:
c = (data_set >= k_max).sum()
C[i] = c
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
axis.bar(k, C)
axis.set_yscale("log")
axis.set_xlabel("Counts")
axis.set_ylabel("Number of counts")
data.saveFigure(figure, figure_name)
for i in reversed(range(k_min, k_max + 1)):
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
if i < k_max:
C[i] += C[i + 1]
C[i + 1] = 0
axis.bar(k[:i + 1], C[:i + 1])
axis.set_yscale("log")
axis.set_xlabel("Counts")
axis.set_ylabel("Number of counts")
data.saveFigure(figure, figure_name + "_" + str(i))
def plotHistogram(series, x_label, scale="linear", normed=False, name=None):
figure_name = "histogram"
if name:
figure_name = name + "_" + figure_name
figure = pyplot.figure()
axis = figure.add_subplot(1, 1, 1)
seaborn.distplot(series, kde=False, norm_hist=normed, ax=axis)
axis.set_yscale(scale)
axis.set_xlabel(x_label)
# axis.set_ylabel(y_label)
data.saveFigure(figure, figure_name)
