def plot_clustering(container,
in_dir,
isi_hist,
ac_hist,
dend,
cluster,
opt_end=""):
dend_idxs = dend["leaves"][::-1]
sorted_isi = isi_hist[dend_idxs]
sorted_ac = ac_hist[dend_idxs]
ax1, fig1 = nc_plot._make_ax_if_none(None)
cmap = sns.cubehelix_palette(8,
start=0.5,
rot=-.75,
dark=0,
light=.95,
reverse=True)
# cmap = sns.color_palette("Blues")
sns.heatmap(sorted_isi, ax=ax1, yticklabels=5, xticklabels=10, cmap=cmap)
ax1.set_ylim([sorted_isi.shape[0], 0])
ax1.axvline(x=16, c="r", ls="--")
plot_loc = os.path.join(in_dir, "nc_plots",
"logisi_hist" + opt_end + ".png")
fig1.savefig(plot_loc, dpi=400)
fig1.clear()
ax1, fig1 = nc_plot._make_ax_if_none(None)
sns.heatmap(sorted_ac, ax=ax1, yticklabels=5, xticklabels=5, cmap=cmap)
ax1.axvline(x=6, c="r", ls="--")
ax1.set_ylim([sorted_ac.shape[0], 0])
plot_loc = os.path.join(in_dir, "nc_plots", "ac_hist" + opt_end + ".png")
fig1.savefig(plot_loc, dpi=400)
fig1.clear()
def ward_clustering(data,
in_dir,
plot_dim1=0,
plot_dim2=1,
opt_end="",
s_color=False):
"""
Perform heirarchical clustering using ward's method
Params
------
data - input data array
in_dir - where to save the result to
plot_dim1 - the PCA dimension to plot
plot_dim2 - the other PCA dimesion to plot
"""
ax, fig = nc_plot._make_ax_if_none(None)
if s_color:
shc.set_link_color_palette(
['#4A777A', "#6996AD", "#82CFFD", "#3579DC"])
else:
shc.set_link_color_palette(["k"])
atc = '#bcbddc'
dend = shc.dendrogram(shc.linkage(data,
method="ward",
optimal_ordering=True),
ax=ax,
above_threshold_color=atc,
orientation="right",
no_labels=True)
plot_loc = os.path.join(in_dir, "nc_plots", "dendogram" + opt_end + ".png")
fig.savefig(plot_loc, dpi=400)
shc.set_link_color_palette(None)
cluster = AgglomerativeClustering(n_clusters=2,
affinity="euclidean",
linkage="ward")
cluster.fit_predict(data)
ax, fig = nc_plot._make_ax_if_none(None)
markers = list(map(lambda a: "Burst" if a else "Regular", cluster.labels_))
# ax.scatter(data[:, plot_dim1], data[:, plot_dim2],
# c=markers)
sns.scatterplot(data[:, plot_dim1],
data[:, plot_dim2],
ax=ax,
style=markers,
hue=markers)
plot_loc = os.path.join(in_dir, "nc_plots", "PCAclust" + opt_end + ".png")
fig.savefig(plot_loc, dpi=400)
return cluster, dend
def calculate_isi_hist(container, in_dir, opt_end="", s_color=False):
"""Calculate a matrix of isi_hists for each unit in a container"""
ax1, fig1 = nc_plot._make_ax_if_none(None)
isi_hist_matrix = np.empty((len(container), 60), dtype=float)
color = iter(cm.gray(np.linspace(0, 0.8, len(container))))
for i, ndata in enumerate(container):
res_isi, bins = log_isi(ndata)
isi_hist_matrix[i] = res_isi
bin_centres = bins[:-1] + np.mean(np.diff(bins)) / 2
c = next(color) if s_color else "k"
ax1.plot(bin_centres, res_isi, c=c)
ax1.set_xlim([-3, 1])
ax1.set_xticks([-3, -2, -1, 0])
ax1.axvline(x=np.log10(0.006), c="r", ls="--")
plot_loc = os.path.join(in_dir, "nc_plots", "logisi" + opt_end + ".png")
fig1.savefig(plot_loc, dpi=400)
return isi_hist_matrix
def calculate_auto_corr(container, in_dir, opt_end="", s_color=False):
"""Calculate a matrix of autocorrs for each unit in a container"""
ax1, fig1 = nc_plot._make_ax_if_none(None)
auto_corr_matrix = np.empty((len(container), 20), dtype=float)
color = iter(cm.gray(np.linspace(0, 0.8, len(container))))
for i, ndata in enumerate(container):
auto_corr_data = ndata.isi_auto_corr(bins=1, bound=[0, 20])
auto_corr_matrix[i] = (auto_corr_data["isiCorr"] /
ndata.spike.get_unit_stamp().size)
bins = auto_corr_data['isiAllCorrBins']
bin_centres = bins[:-1] + np.mean(np.diff(bins)) / 2
c = next(color) if s_color else "k"
ax1.plot(bin_centres / 1000, auto_corr_matrix[i], c=c)
ax1.set_xlim([0.000, 0.02])
ax1.set_xticks([0.000, 0.005, 0.01, 0.015, 0.02])
ax1.axvline(x=0.006, c="r", ls="--")
plot_loc = os.path.join(in_dir, "nc_plots", "autocorr" + opt_end + ".png")
fig1.savefig(plot_loc, dpi=400)
return auto_corr_matrix