Switch scripts to use pyNN.spiNNaker and minor tidying up

eprop_testing/create_pops_for_incremental_learning.py

@@ -67,6 +67,9 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
 def first_create_pops():
     pynn.setup(timestep=1)
     pynn.set_number_of_neurons_per_core(pynn.extra_models.EPropAdaptive, 6)
+
+    print("Check rates: ", rates)
+
     input_pop = pynn.Population(input_size,
                                 pynn.SpikeSourcePoisson(rate=rates),
                                 label='input_pop')
@@ -101,6 +104,9 @@ def first_create_pops():
                                    weight=in_weight)  # connect all 2 prompt
     from_list_in += range_connector(ps * 2, ps * 4, 0, neuron_pop_size, delay_offset=0,
                                     weight=prompt_weight)  # connect all 2 prompt
+
+    print("length ", len(from_list_in))
+
     in_proj = pynn.Projection(input_pop,
                               neuron,
                               pynn.FromListConnector(from_list_in),
@@ -303,7 +309,7 @@ def run_until(experiment_label, runtime, pynn, in_proj, recurrent_proj, out_proj
         pynn.run(window_size)
         # in_spikes = input_pop.get_data('spikes', clear=True)
         # neuron_res = neuron.get_data('all', clear=True)
-        readout_res = readout_pop.get_data('all')  #, clear=True)
+        readout_res = readout_pop.get_data('all', clear=True)
 
         for cycle in range(window_cycles):
             cycle_error.append(0.0)
@@ -312,7 +318,8 @@ def run_until(experiment_label, runtime, pynn, in_proj, recurrent_proj, out_proj
                 instantaneous_error = np.abs(float(
                     readout_res.segments[0].filter(
                         name='gsyn_inh')[0][time_index + (
-                            (cycle+current_iter) * cycle_time)][0]))
+                            cycle * cycle_time)][0]))
+                            # (cycle+current_iter) * cycle_time)][0]))
                 cycle_error[-1] += instantaneous_error
             if cycle_error[-1] < 75:
                 correct_or_not[-1] = 1
@@ -375,9 +382,12 @@ def run_until(experiment_label, runtime, pynn, in_proj, recurrent_proj, out_proj
         print(experiment_label)
 
         graph_directory = './big_with_labels/'
-        draw_graph_from_list(new_connections_in, new_connections_rec, new_connections_out, graph_directory,
-                             experiment_label + ' {}'.format(current_window), rec_flag=recurrent_connections,
-                             save_flag=True)
+        draw_graph_from_list(
+            new_connections_in, new_connections_rec, new_connections_out,
+            graph_directory,
+            experiment_label + ' {}'.format(current_window),
+            rec_flag=recurrent_connections,
+            save_flag=True)
         plot_learning_curve(correct_or_not, cycle_error, graph_directory,
                             experiment_label + ' {}'.format(current_window),
                             save_flag=True,

eprop_testing/delta_synapse_test.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/hand_crafted_left_right.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz

eprop_testing/incremental_graph_drawing.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz
@@ -139,7 +139,9 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
     # "w_fb": [(np.random.random() * 2) - 1. for i in range(neuron_pop_size)],
     # "w_fb": [4*np.random.random() - 4*np.random.random() for i in range(neuron_pop_size)],  ## for both feedback weights
     # "w_fb": [-3]*(neuron_pop_size/2) + [3]*(neuron_pop_size/2),
-    "w_fb": [3]*int(neuron_pop_size/4) + [-3]*int(neuron_pop_size/4) + [3]*int(neuron_pop_size/4) + [-3]*int(neuron_pop_size/4),
+    "w_fb": [3.0]*int(neuron_pop_size/4) + [-3.0]*int(
+        neuron_pop_size/4) + [3.0]*int(neuron_pop_size/4) + [-3.0]*int(
+            neuron_pop_size/4),
     # "B": 0.0,
     "beta": beta,
     "target_rate": 10,
@@ -254,15 +256,16 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
     pynn.run(window_size)
     in_spikes = input_pop.get_data('spikes', clear=True)
     neuron_res = neuron.get_data('all', clear=True)
-    readout_res = readout_pop.get_data('all')  #, clear=True)
+    readout_res = readout_pop.get_data('all', clear=True)
     plot_start = (window_size*current_window)
     current_window += 1
     plot_end = (window_size*current_window)
 
     total_error = 0.0
     cycle_error = [0.0 for i in range(current_window*window_cycles)]
     correct_or_not = [0 for i in range(current_window*window_cycles)]
-    for cycle in range(current_window*window_cycles):
+    # for cycle in range(current_window*window_cycles):
+    for cycle in range(window_cycles):
         for time_index in range(cycle_time):
             instantaneous_error = np.abs(float(
                 readout_res.segments[0].filter(name='gsyn_inh')[0][time_index+(cycle*cycle_time)][0]))

eprop_testing/input_buffer_overflow_test.py

@@ -1,10 +1,10 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz
 from pyNN.random import NumpyRNG, RandomDistribution
 from pyNN.utility.plotting import Figure, Panel
-# from spynnaker.pyNN.spynnaker_external_device_plugin_manager import SpynnakerExternalDevicePluginManager
+from spynnaker.pyNN.spynnaker_external_device_plugin_manager import SpynnakerExternalDevicePluginManager
 
 def weight_distribution(pop_size):
     base_weight = np.random.randn() / np.sqrt(pop_size) #+ 0.5
@@ -75,7 +75,7 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
 readout_neuron_params = {
     "v": 0,
     "v_thresh": 30, # controls firing rate of error neurons
-    "poisson_pop_size": input_size / 4,
+    "poisson_pop_size": input_size // 4,
     "rate_on": 40,
     "rate_off": 0,
     # "tau_m": tau_a,
@@ -85,7 +85,7 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
     }
 rates = []
 for i in range(input_size):
-    if i >= (3*input_size) / 4:
+    if i >= (3*input_size) // 4:
         rates.append(10)
     else:
         rates.append(0)
@@ -99,7 +99,7 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
 ratio_of_LIF = 0.5
 beta = []
 for i in range(neuron_pop_size):
-    if i < neuron_pop_size/2:
+    if i < neuron_pop_size//2:
     # if i % 2 == 0:
         beta.append(25.8)
     else:
@@ -145,8 +145,8 @@ def range_connector(pre_min, pre_max, post_min, post_max, weight=1.5, delay_offs
 
 # from_list_in, max_syn_per_input = probability_connector(input_size, neuron_pop_size, p_connect_in)
 ps = readout_neuron_params["poisson_pop_size"]
-from_list_in = range_connector(0, ps, 0, neuron_pop_size/2, weight=in_weight)
-from_list_in += range_connector(ps, ps*2, neuron_pop_size/2, neuron_pop_size, weight=in_weight)
+from_list_in = range_connector(0, ps, 0, neuron_pop_size//2, weight=in_weight)
+from_list_in += range_connector(ps, ps*2, neuron_pop_size//2, neuron_pop_size, weight=in_weight)
 from_list_in += range_connector(ps*2, ps*3, 0, neuron_pop_size, delay_offset=0, weight=prompt_weight)
 in_proj = pynn.Projection(input_pop,
                           neuron,

eprop_testing/learn_sinusoid_from_frozen_poisson.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz

eprop_testing/learn_sinusoid_only_readout.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz

eprop_testing/learn_sinusoid_only_readout_single_cycle_input.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz
@@ -95,7 +95,7 @@ def probability_connector(pre_pop_size, post_pop_size, prob, offset=0):
 for i in range(num_repeats):
     print('\nrepeat: {}\n'.format(i))
     pynn.run(cycle_time)
-    
+
 in_spikes = input_pop.get_data('spikes')
 readout_res = readout_pop.get_data('all')
 
@@ -117,7 +117,7 @@ def probability_connector(pre_pop_size, post_pop_size, prob, offset=0):
 connection_diff_in = []
 for i in range(len(from_list_in)):
     connection_diff_in.append(new_connections_in[i][2] - from_list_in[i][2])
-    
+
 print("Input connections\noriginal\n", np.array(from_list_in))
 print("new\n", np.array(new_connections_in))
 print("diff\n", np.array(connection_diff_in))

eprop_testing/one_neuron.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/signed_weights_test.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/test_8_neuron_regularisation.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/test_adaptive_threshold.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/test_learning_signal_by_payload.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/test_left_right_readout.py

@@ -1,7 +1,7 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
-from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz
+# from frozen_poisson import build_input_spike_train, frozen_poisson_variable_hz
 from pyNN.random import NumpyRNG, RandomDistribution
 from pyNN.utility.plotting import Figure, Panel
 from spynnaker.pyNN.spynnaker_external_device_plugin_manager import SpynnakerExternalDevicePluginManager

eprop_testing/test_single_plastic_update.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution
@@ -29,7 +29,7 @@
 eprop_learning = pynn.STDPMechanism(
     timing_dependence=pynn.extra_models.TimingDependenceEprop(),
     weight_dependence=pynn.extra_models.WeightDependenceEpropReg(
-        w_min=-2.0, w_max=2.0, reg_rate=1.5), 
+        w_min=-2.0, w_max=2.0, reg_rate=1.5),
     weight=start_w, delay=[0, 0])
 
 proj = pynn.Projection(spike_source, neuron,

eprop_testing/test_sinusoid_learning.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution

eprop_testing/test_synapse_state_evol.py

@@ -1,4 +1,4 @@
-import spynnaker8 as pynn
+import pyNN.spiNNaker as pynn
 import numpy as np
 import matplotlib.pyplot as plt
 from pyNN.random import NumpyRNG, RandomDistribution