def stashfig(name, **kws):
if SAVEFIGS:
savefig(name,
foldername=FNAME,
fmt=DEFAULT_FMT,
dpi=DEFUALT_DPI,
**kws)
color_graphs,
labels=graph_type_labels,
outer_hier_labels=side_labels,
inner_hier_labels=simple_class_labels,
transform="simple-all",
height=20,
hier_label_fontsize=22,
sort_nodes=False,
sizes=(4, 20),
palette=palette,
title=r"All paired neurons",
font_scale=2.2,
title_pad=260,
legend_name="Edge type",
)
savefig("color_graphs_pretty", pad_inches=1, bbox_inches="tight", fmt="png", dpi=200)
countplot(color_graphs, graph_type_labels)
savefig(
"color_graphs_pretty_prop", bbox_inches="tight", pad_inches=0.5, fmt="png", dpi=200
)
#%% try just one induced subgraph
# Plot left to left
color_graphs_small = [c[:n_per_side, :n_per_side] for c in color_graphs]
sns.set_palette(palette, desat=0.7)
sns.set_context("talk", font_scale=1.2)
gridplot(
color_graphs_small,
labels=graph_type_labels,
inner_hier_labels=simple_class_labels[:n_per_side],
transform="simple-all",
class_ind_map = {}
class_ids_map = {}
for i, class_name in enumerate(unique_classes):
inds = np.where(inverse_classes == i)[0]
ids = cell_ids[inds]
class_ind_map[class_name] = inds
class_ids_map[class_name] = ids
class_ind_map
#%%
plt.figure(figsize=(10, 5))
sns.distplot(adj.sum(axis=0), norm_hist=False)
plt.xlabel("Proportion of input w/in current graph")
plt.ylabel("Frequency")
plt.tight_layout()
savefig("proportion_win_graph", fmt="png")
# degreeplot(adj)
#%%
# node = str(df_ids[class_ids_map["ORN mPNs"][0]])
# neighbor_graph = nx.ego_graph(graph, node)
# neighbor_names = meta_to_array(neighbor_graph, "Class")
# neighbor_nodes = list(neighbor_graph.nodes())
# labels = dict(zip(neighbor_nodes, neighbor_names))
# plt.figure(figsize=(20, 20))
# nx.draw_networkx(neighbor_graph, labels=labels)
# plt.show()
#%%
# Plot the adjacency matrix for the summed graph
plt.figure(figsize=(5, 5))
ax = heatmap(
sum_adj,
inner_hier_labels=simple_class_labels,
transform="simple-all",
hier_label_fontsize=18,
sort_nodes=False,
cbar=False,
title="Right Mushroom Body (summed 4 channels)",
title_pad=90,
font_scale=1.7,
)
annotate_arrow(ax, (0.135, 0.88))
savefig("flat_mb", fmt="png", dpi=150, bbox_inches="tight", pad_inches=0.5)
# Plot the adjacency matrix for the 4-color graphs
fig, ax = plt.subplots(2, 2, figsize=(20, 20))
ax = ax.ravel()
for i, g in enumerate(color_adjs):
heatmap(
g,
inner_hier_labels=simple_class_labels,
transform="simple-all",
hier_label_fontsize=18,
sort_nodes=False,
ax=ax[i],
cbar=False,
title=GRAPH_TYPE_LABELS[i],
title_pad=70,
pred_lhn_ids).mean()) # how many of the og lhn i got
print("Precision:")
print(np.isin(pred_lhn_ids,
true_lhn_ids).mean()) # this is how many of mine are in og
print(len(pred_lhn_ids))
# update the predicted labels and class map
pred_classes = update_classes(classes, pred_lhn_ids, "LHN")
class_ids_map, class_ind_map = update_class_map(cell_ids, pred_classes)
# plot output
if plot:
for t in pn_types:
incidence_plot(adj, pred_classes, t)
savefig("lhn_" + t)
#%% Estimate CN neurons
innate_input_types = ["ORN mPNs", "ORN uPNs", "tPNs", "vPNs", "LHN"]
innate_thresh = 5 * [0.05]
mb_input_types = ["MBON"]
mb_thresh = [0.05]
pred_innate_ids = proportional_search(adj,
class_ind_map,
innate_input_types,
cell_ids,
thresh=innate_thresh)
pred_learn_ids = proportional_search(adj,
class_ind_map,
mb_input_types,
result_df["Left ID"] = left_nodes
result_df["Right ID"] = right_nodes
result_df["Flat Distance"] = norm_diff_summed
result_df["4color Distance"] = norm_diff_colors
result_df["Flat Rank"] = rank_summed
result_df["4color Rank"] = rank_colors
result_df["Reciprocal Rank Diff"] = reciprocal_diff
result_df["Rank Diff"] = rank_diff
plt.figure(figsize=(10, 10))
sns.set_context("talk", font_scale=1)
sns.scatterplot(data=result_df,
x="Flat Distance",
y="4color Distance",
alpha=0.8)
savefig("color_distance_comparison", fmt="png", dpi=200)
plt.figure(figsize=(10, 10))
sns.scatterplot(
data=result_df,
x="Flat Distance",
y="4color Distance",
# hue="Reciprocal Rank Diff",
alpha=0.8,
palette="gist_heat_r",
size="Reciprocal Rank Diff",
sizes=(10, 1000),
)
savefig("color_distance_comparison_heat", fmt="png", dpi=200)
plt.figure(figsize=(10, 10))
[embed_graph],
inner_hier_labels=hier1_labels,
title="whole graph sorted by hierclust, n_clust=6",
**gridplot_kws,
)
###########
#%% focus on subgraph id 2
subvert_inds = np.where(hier1_labels == 2)[0]
subgraph = embed_graph[np.ix_(subvert_inds, subvert_inds)]
subgraph = get_lcc(subgraph)
subgraph_latent = normalized_ase(subgraph, n_components=4)
dists = pairwise_distances(subgraph_latent, metric="cosine")
hierplot(dists, figsize=(20, 20))
savefig("hier2_dissim_heat", fmt="png")
#%% cluster a subgraph
hierclust = AgglomerativeClustering(
affinity="precomputed", compute_full_tree=True, n_clusters=7, linkage="average"
)
hier2_labels = hierclust.fit_predict(dists)
gridplot([subgraph], inner_hier_labels=hier2_labels, **gridplot_kws)
#%% embed a subgraph
plt.style.use("seaborn-white")
l = normalized_ase(lcc_graph_t, n_components=6)
dists = pairwise_distances(l, metric="cosine")
c = AgglomerativeClustering(n_clusters=7, affinity="precomputed", linkage="average")
labels = c.fit_predict(dists)
np.unique(labels, return_counts=True)
with timer(prefix='plot oof-test-distributions '):
for c, ax in zip(oofs.columns, axes):
sns.histplot(
data=pd.DataFrame({
'OOF': oofs[c],
'Test': preds[c]
}),
ax=ax,
stat='density',
common_norm=False,
)
ax.set_facecolor('white')
ax.set_title(category_id2code[c])
ax.grid()
fig.tight_layout()
savefig(fig, to='out-of-fold_vs_test')
# %%
nfigs = len(oofs.columns)
ncols = 5
nrows = -(-nfigs // ncols)
fig, axes = plt.subplots(figsize=(5 * ncols, 3 * nrows),
ncols=ncols,
nrows=nrows,
sharex=True)
axes = np.ravel(axes)
with timer(prefix='plot oof-test-distributions '):
for c, ax in zip(oofs.columns, axes):
sns.histplot(
print(graph_type)
edgelist_path = data_path / data_date / (graph_type + ".csv")
adj = pd.read_csv(edgelist_path, index_col=0)
print((adj.index.values == meta_data_df.index.values).all())
graph = df_to_nx(adj, meta_data_dict)
nx_graphs_raw[graph_type] = graph
df_graphs_raw[graph_type] = adj
name = graph_type + "_raw_PTR"
heatmap(adj.values, title=name, **heatmap_kws)
if save_figures:
savefig(name, fmt="png", dpi=300)
print()
#%% Normalize weights for the raw graphs
df_graphs_norm = {}
nx_graphs_norm = {}
input_counts = input_counts_df["axon_inputs"].values
input_counts[input_counts == 0] = 1
for graph_type in ["axon-axon", "dendrite-axon"]:
print(graph_type)
df_adj_raw = df_graphs_raw[graph_type]
if (input_counts_df.index.values == adj.index.values).all():
print("Same indexing!")
adj_raw = df_adj_raw.values
# uni_labels = np.unique(pred_labels)
sankey(ax[1], class_labels, pred_labels)
ax[1].axis("off")
rand_perm = np.random.permutation(len(pred_labels))
heatmap(
binarize(adj),
inner_hier_labels=pred_labels[rand_perm],
# outer_hier_labels=side_labels,
hier_label_fontsize=18,
ax=ax[2],
cbar=False,
sort_nodes=True,
)
sankey(ax[3], class_labels, pred_labels[rand_perm])
ax[3].axis("off")
print(k)
if k == 6:
print(k)
savefig("sankey6", fmt="png", dpi=200)
# %%
fig, ax = plt.subplots(1, 1, figsize=(20, 20))
sns.set_context("talk", font_scale=2)
sankey(ax, class_labels, pred_labels, aspect=20, fontsize=24)
ax.axis("off")
# %%
