Update the notebooks, make k.inner=False the default

h-mayorquin · Mar 30, 2017 · 3dba22d · 3dba22d
1 parent 3a59cb8
commit 3dba22d
Show file tree

Hide file tree

Showing 9 changed files with 842 additions and 207 deletions.
diff --git a/__init__.py b/__init__.py
diff --git a/benchmarking/__init__.py b/benchmarking/__init__.py
diff --git a/benchmarking/standard_program.py b/benchmarking/standard_program.py
diff --git a/benchmarking/standard_program_times.py b/benchmarking/standard_program_times.py
diff --git a/network.py b/network.py
@@ -163,7 +163,7 @@ def update_continuous(self, dt=1.0, sigma=None):
 class BCPNNFast:
     def __init__(self, hypercolumns, minicolumns, beta=None, w=None, G=1.0, tau_m=0.050, g_w=1, g_w_ampa=1.0, g_beta=1,
                  tau_z_pre=0.150, tau_z_post=0.005, tau_z_pre_ampa=0.005, tau_z_post_ampa=0.005, tau_p=5.0, tau_k=0.010,
-                 tau_a=2.70, g_a=97.0, g_I=10.0, p=1.0, k=0.0, sigma=1.0, epsilon=1e-20, k_inner=True, prng=np.random):
+                 tau_a=2.70, g_a=97.0, g_I=10.0, p=1.0, k=0.0, sigma=1.0, epsilon=1e-20, k_inner=False, prng=np.random):
         # Initial values are taken from the paper on memory by Marklund and Lansner also from Phil's paper
 
         # Random number generator
@@ -714,17 +714,30 @@ def create_overload_chain(self, number_of_sequences, half_width, units_to_overlo
 
             sequence = []
 
-            for i in range(half_width):
-                sequence.append(number)
-                number += 1
-            sequence += units_to_overload
+            # The first half
+            i = 0
+            while i < half_width:
+                if number in units_to_overload:
+                    number += 1
+
+                else:
+                    sequence.append(number)
+                    number += 1
+                    i += 1
 
-            for i in range(half_width):
-                sequence.append(number)
-                number += 1
+            # The overload units in the middle
+            sequence += units_to_overload
 
+            # The second half
+            i = 0
+            while i < half_width:
+                if number in units_to_overload:
+                    number += 1
+                else:
+                    sequence.append(number)
+                    number += 1
+                    i += 1
 
             chain.append(sequence)
 
         return chain
-
diff --git a/notebooks/2017-03-24(Speed of patterns study).ipynb b/notebooks/2017-03-24(Speed of patterns study).ipynb
@@ -122,7 +122,6 @@
     "\n",
     "for epochs in epochs_range:\n",
     "    # Protocol\n",
-    "    training_time = 0.1\n",
     "    inter_sequence_interval = 1.0\n",
     "    inter_pulse_interval = 0.0\n",
     "\n",

diff --git a/notebooks/2017-03-28(Unit load study).ipynb b/notebooks/2017-03-28(Unit load study).ipynb
diff --git a/notebooks/2017-03-30(Unit Load II).ipynb b/notebooks/2017-03-30(Unit Load II).ipynb
diff --git a/play.py b/play.py
@@ -1,77 +0,0 @@
-from __future__ import print_function
-
-import pprint
-
-import numpy as np
-import matplotlib.pyplot as plt
-import matplotlib
-import matplotlib.gridspec as gridspec
-from mpl_toolkits.axes_grid1 import make_axes_locatable
-
-from network import BCPNN, NetworkManager, BCPNNFast, Protocol
-from plotting_functions import plot_state_variables_vs_time, plot_network_activity, plot_network_activity_angle
-from plotting_functions import plot_adaptation_dynamics, plot_weight_matrix
-from analysis_functions import calculate_compression_factor, calculate_recall_success, calculate_timings
-from plotting_functions import plot_winning_pattern, plot_sequence, plot_network_activity_angle
-
-# np.set_printoptions(suppress=True, precision=2)
-
-# Patterns parameters
-hypercolumns = 4
-minicolumns = 30
-
-# Manager properties
-dt = 0.001
-T_recall = 3.0
-values_to_save = ['o']
-
-compression_list = []
-
-
-# Protocol
-training_time = 0.1
-inter_sequence_interval = 1.0
-inter_pulse_interval = 0.0
-epochs = 3
-
-# Build the network
-nn = BCPNNFast(hypercolumns, minicolumns)
-nn.k_inner = False
-
-# Build the manager
-manager = NetworkManager(nn=nn, dt=dt, values_to_save=values_to_save)
-
-# Build the protocol for
-protocol = Protocol()
-
-number_of_sequences = 3
-half_width = 2
-units_to_overload = [10]
-chain = []
-
-protocol = Protocol()
-chain = protocol.create_overload_chain(number_of_sequences, half_width, units_to_overload)
-protocol.cross_protocol(chain, training_time, inter_sequence_interval, epochs)
-# Train
-manager.run_network_protocol(protocol, verbose=False, values_to_save_epoch=None, reset=True, empty_history=True)
-
-# Recall
-if True:
-    manager.run_network_recall(T_recall=T_recall, T_cue=0.1, I_cue=0, reset=True, empty_history=True)
-    manager.run_network_recall(T_recall=T_recall, T_cue=0.1, I_cue=5, reset=True, empty_history=False)
-    manager.run_network_recall(T_recall=T_recall, T_cue=0.1, I_cue=11, reset=True, empty_history=False)
-    manager.run_network_recall(T_recall=T_recall, T_cue=0.1, I_cue=15, reset=True, empty_history=False)
-
-    # Timings
-    timings = calculate_timings(manager, remove=0.010)
-    print(timings)
-
-    plot_winning_pattern(manager, remove=0.010)
-    plt.show()
-
-if True:
-    n = 5
-    for sequence in chain:
-        success = calculate_recall_success(manager, T_recall=T_recall, I_cue=sequence[0], T_cue=0.1, n=n,
-                                        patterns_indexes=sequence)
-        print(success)