In [1]:

    

import numpy as np
import pickle
import matplotlib.pyplot as plt
import tensorflow as tf
import DeepSparseCoding.utils.image_processing as ip
import DeepSparseCoding.utils.plot_functions as pf


    

    




---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-1-4fcb1a8f3090> in <module>()
      3 import matplotlib.pyplot as plt
      4 import tensorflow as tf
----> 5 import DeepSparseCoding.utils.image_processing as ip
      6 import DeepSparseCoding.utils.plot_functions as pf

ImportError: No module named 'DeepSparseCoding'

In [ ]:

    

weights = np.load("/home/dpaiton/CAE_Project/32768_min_compress_pcm_weights.npz")["data"]


    

In [ ]:

    

l1_weights = weights[5].reshape((81, 128))
bf_stats = ip.get_dictionary_stats(l1_weights, padding=256, num_gauss_fits=10, gauss_thresh=0.2)


    

In [ ]:

    

fig = pf.plot_data_tiled(l1_weights.T.reshape(128, 9, 9), normalize=True)


    

In [ ]:

    

fig = pf.plot_loc_freq_summary(bf_stats)


    

In [ ]:

    

fig = pf.plot_bf_stats(bf_stats, num_bf=10)


    

In [ ]:

    

#TODO: Why are these weird?
fig = pf.plot_gaussian_contours(bf_stats, num_plots=128)


    

In [ ]:

    

fig = pf.plot_ellipse_summaries(bf_stats, num_bf=128, lines=True)


    

In [ ]:

    

activities = pickle.load(open("/home/dpaiton/CAE_Project/32768_min_compress_pcm_ulist.npz", "rb"))


    

In [ ]:

    

num_imgs = activities[0].shape[0]
l1_act_corr = 0
for x_idx in range(64):
  for y_idx in range(64):
    l1_act = activities[1][:, x_idx, y_idx, :]
    cent_l1_act = l1_act - np.mean(l1_act, axis=0)
    norm_l1_act = np.divide(cent_l1_act, np.std(cent_l1_act, axis=0))
    conv_act_corr = np.dot(norm_l1_act.T, norm_l1_act) / num_imgs
    l1_act_corr += conv_act_corr
l1_act_corr /= 64*64

l2_act_corr = 0
for x_idx in range(32):
  for y_idx in range(32):
    l2_act = activities[2][:, x_idx, y_idx, :]
    cent_l2_act = l2_act - np.mean(l2_act, axis=0)
    norm_l2_act = np.divide(cent_l2_act, np.std(cent_l2_act, axis=0))
    conv_act_corr = np.dot(norm_l2_act.T, norm_l2_act) / num_imgs
    l2_act_corr += conv_act_corr
l2_act_corr /= 32*32


    

In [ ]:

    

fig = pf.plot_cov_matrix(l1_act_corr, str(num_imgs))
fig = pf.plot_cov_matrix(l2_act_corr, str(num_imgs))


    

In [2]:

    

l1_act_evals, l1_act_evecs = np.linalg.eig(l1_act_corr)
l2_act_evals, l2_act_evecs = np.linalg.eig(l2_act_corr)


    

    




---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-2-a3afcad7f1e4> in <module>()
----> 1 l1_act_evals, l1_act_evecs = np.linalg.eig(l1_act_corr)
      2 l2_act_evals, l2_act_evecs = np.linalg.eig(l2_act_corr)

NameError: name 'l1_act_corr' is not defined

In [3]:

    

l1_norm_evals = l1_act_evals - (np.identity(l1_act_evals.shape[0])*l1_act_evals)
l2_norm_evals = l2_act_evals - (np.identity(l2_act_evals.shape[0])*l2_act_evals)
fig, ax = plt.subplots(1)
ax.scatter(l1_norm_evals, l2_norm_evals)
ax.set_xlabel("layer 1 eigen values")
ax.set_ylabel("layer 2 eigen values")
ax.set_xlim([0,6])
ax.set_ylim([0,6])
ax.plot([0,6], [0,6], 'r')
plt.show()


    

    




---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-3-582abc169ffb> in <module>()
----> 1 l1_norm_evals = l1_act_evals - (np.identity(l1_act_evals.shape[0])*l1_act_evals)
      2 l2_norm_evals = l2_act_evals - (np.identity(l2_act_evals.shape[0])*l2_act_evals)
      3 fig, ax = plt.subplots(1)
      4 ax.scatter(l1_norm_evals, l2_norm_evals)
      5 ax.set_xlabel("layer 1 eigen values")

NameError: name 'l1_act_evals' is not defined

In [99]:

    

fig1 = pf.plot_eigenvalues(l1_act_evals, ylim=[0,100])
fig2 = pf.plot_eigenvalues(l2_act_evals, ylim=[0,100])


    

In [100]:

    

fig = pf.plot_pooling_summaries(bf_stats, act_evecs, num_pooling_filters=64, num_connected_weights=128, lines=True)


    

In [102]:

    

fig = pf.plot_pooling_centers(bf_stats, act_evecs, num_pooling_filters=64, fig_size=(10,10), spot_size=60)


    

In [1]:

    

gdn_weights = [weights[19+i] for i in range(3)]
fig, ax = plt.subplots(1, 4, figsize=(20,20))
for i in range(3):
  im = ax[i].imshow(gdn_weights[i], vmin=0, vmax=1, cmap="Greys_r")
  ax[i].set_title("GDN weights for layer "+str(i))
pf.clear_axis(ax[-1])
pf.add_colorbar_to_im(ax[-1].imshow(np.ones(gdn_weights[0].shape), vmin=0, vmax=1, cmap="Greys_r"))
plt.show()


    

    




---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-1-d64e37df41c6> in <module>()
----> 1 gdn_weights = [weights[19+i] for i in range(3)]
      2 fig, ax = plt.subplots(1, 4, figsize=(20,20))
      3 for i in range(3):
      4   im = ax[i].imshow(gdn_weights[i], vmin=0, vmax=1, cmap="Greys_r")
      5   ax[i].set_title("GDN weights for layer "+str(i))

<ipython-input-1-d64e37df41c6> in <listcomp>(.0)
----> 1 gdn_weights = [weights[19+i] for i in range(3)]
      2 fig, ax = plt.subplots(1, 4, figsize=(20,20))
      3 for i in range(3):
      4   im = ax[i].imshow(gdn_weights[i], vmin=0, vmax=1, cmap="Greys_r")
      5   ax[i].set_title("GDN weights for layer "+str(i))

NameError: name 'weights' is not defined