From 0cf8e39b2b5e9c4e6fb30b88f180f73ca48dec80 Mon Sep 17 00:00:00 2001
From: Nikolai Hartmann <Nikolai.Hartmann@physik.uni-muenchen.de>
Date: Tue, 31 Jul 2018 18:28:23 +0200
Subject: [PATCH] switching pixel pots to pcolormesh
---
plotting.py | 42 ++++++++++++++++++++++++++++--------------
1 file changed, 28 insertions(+), 14 deletions(-)
diff --git a/plotting.py b/plotting.py
index 50e5888..4340f7a 100644
--- a/plotting.py
+++ b/plotting.py
@@ -6,7 +6,7 @@ import math
import matplotlib.pyplot as plt
import matplotlib.colors
from matplotlib.ticker import LogFormatter
-from mpl_toolkits.axes_grid1 import ImageGrid
+from mpl_toolkits.axes_grid1 import ImageGrid, make_axes_locatable
import numpy as np
from .keras_visualize_activations.read_activations import get_activations
@@ -48,7 +48,9 @@ def plot_NN_vs_var_2D(plotname, means,
var2_label=None,
logscale=False,
ncontours=20,
- black_contourlines=False):
+ only_pixels=False,
+ black_contourlines=False,
+ cmap="inferno"):
print("Creating varied events (2d)")
# example: vary var1 vs var2
@@ -79,14 +81,23 @@ def plot_NN_vs_var_2D(plotname, means,
zmax = np.max(scores)
if logscale:
+ if zmin <= 0:
+ zmin = 1e-5
+ print("Setting zmin to {}".format(zmin))
lvls = np.logspace(math.log10(zmin), math.log10(zmax), ncontours)
- pcm = ax.contourf(sequence1, sequence2, scores, levels=lvls, norm=matplotlib.colors.LogNorm(vmin=zmin, vmax=zmax))
+ if only_pixels:
+ pcm = ax.pcolormesh(sequence1, sequence2, scores, norm=matplotlib.colors.LogNorm(vmin=zmin, vmax=zmax), cmap=cmap)
+ else:
+ pcm = ax.contourf(sequence1, sequence2, scores, levels=lvls, norm=matplotlib.colors.LogNorm(vmin=zmin, vmax=zmax), cmap=cmap)
if black_contourlines:
ax.contour(sequence1, sequence2, scores, levels=lvls, colors="k", linewidths=1)
l_f = LogFormatter(10, labelOnlyBase=False, minor_thresholds=(np.inf, np.inf))
cbar = fig.colorbar(pcm, ax=ax, extend='max', ticks=lvls, format=l_f)
else:
- pcm = ax.contourf(sequence1, sequence2, scores, ncontours, norm=matplotlib.colors.Normalize(vmin=0, vmax=1))
+ if only_pixels:
+ pcm = ax.pcolormesh(sequence1, sequence2, scores, norm=matplotlib.colors.Normalize(vmin=zmin, vmax=zmax), cmap=cmap)
+ else:
+ pcm = ax.contourf(sequence1, sequence2, scores, ncontours, norm=matplotlib.colors.Normalize(vmin=zmin, vmax=zmax), cmap=cmap)
if black_contourlines:
ax.contour(sequence1, sequence2, scores, ncontours, colors="k", linewidths=1)
cbar = fig.colorbar(pcm, ax=ax, extend='max')
@@ -107,7 +118,9 @@ def plot_NN_vs_var_2D_all(plotname, model, means,
var1_label=None,
var2_label=None,
zrange=None, logz=False,
- plot_last_layer=False):
+ plot_last_layer=False,
+ log_default_ymin=1e-5,
+ cmap="inferno"):
"Similar to plot_NN_vs_var_2D, but creates a grid of plots for all neurons."
@@ -130,8 +143,6 @@ def plot_NN_vs_var_2D_all(plotname, model, means,
acts = get_activations(model, events, print_shape_only=True)
- aspect = (var1_vals[-1]-var1_vals[0])/(var2_vals[-1]-var2_vals[0])
-
if plot_last_layer:
n_neurons = [len(i[0]) for i in acts]
else:
@@ -142,8 +153,11 @@ def plot_NN_vs_var_2D_all(plotname, model, means,
fig = plt.figure(1, figsize=nrows_ncols)
grid = ImageGrid(fig, 111, nrows_ncols=nrows_ncols[::-1], axes_pad=0,
label_mode="1",
+ aspect=False,
cbar_location="top",
- cbar_mode="single",)
+ cbar_mode="single",
+ cbar_pad=.2,
+ cbar_size="5%",)
grid_array = np.array(grid)
grid_array = grid_array.reshape(*nrows_ncols[::-1])
@@ -158,11 +172,10 @@ def plot_NN_vs_var_2D_all(plotname, model, means,
output_max_default = 1
if global_min <= 0 and logz:
- min_exponent = -5
- global_min = 10**min_exponent
- output_min_default = global_min
- print("Changing global_min to {}".format(global_min))
+ global_min = log_default_ymin
+ print("Changing global_min to {}".format(log_default_ymin))
+ ims = []
for layer in range(layers):
for neuron in range(len(acts[layer][0])):
acts_neuron = acts[layer][:,neuron]
@@ -179,12 +192,13 @@ def plot_NN_vs_var_2D_all(plotname, model, means,
extra_opts["norm"] = norm(vmin=zrange[0], vmax=zrange[1])
else:
extra_opts["norm"] = norm(vmin=global_min, vmax=global_max)
- im = ax.imshow(acts_neuron, origin="lower", extent=[var1_vals[0], var1_vals[-1], var2_vals[0], var2_vals[-1]], aspect=aspect, cmap="jet", **extra_opts)
+ im = ax.pcolormesh(var1_vals, var2_vals, acts_neuron, cmap=cmap, linewidth=0, rasterized=True, **extra_opts)
+ ax.set_facecolor("black")
if var1_label is not None:
ax.set_xlabel(var1_label)
if var2_label is not None:
ax.set_ylabel(var2_label)
- ax.text(0., 0.5, "{}, {}".format(layer, neuron), transform=ax.transAxes)
+ ax.text(0., 0.5, "{}, {}".format(layer, neuron), transform=ax.transAxes, color="white")
cb = fig.colorbar(im, cax=grid[0].cax, orientation="horizontal")
cb.ax.xaxis.set_ticks_position('top')
--
GitLab