def run_flat_gradient(model_path, drop_list=None):
mdata = c.ModelData(model_path)
gpn = MoTypes(False).network_model()
gpn.load(mdata.ModelDefinition, mdata.LastCheckpoint)
flt_params = GlobalDefs.circle_sim_params.copy()
flt_params["t_max"] = flt_params["t_min"]
sim = MoTypes(False).rad_sim(gpn, std, **flt_params)
sim.t_max = sim.t_min # reset gradient to be flat
sim.remove = drop_list
evo_path = model_path + '/evolve/generation_weights.npy'
evo_weights = np.load(evo_path)
w = np.mean(evo_weights[-1, :, :], 0)
sim.bf_weights = w
return sim.run_simulation(GlobalDefs.n_steps, False)
conn_mat = np.zeros((8, 8, len(paths_512_ce)))
for i, p in enumerate(paths_512_ce):
model_cids = clust_ids_ce[all_ids_ce[0, :] == i]
layers_ids = all_ids_ce[1, :][all_ids_ce[0, :] == i]
l_0_mask = np.full(8, False)
ix = model_cids[layers_ids == 0]
ix = ix[ix != -1]
l_0_mask[ix] = True
l_1_mask = np.full(8, False)
ix = model_cids[layers_ids == 1]
ix = ix[ix != -1]
l_1_mask[ix] = True
m_path = mpath(base_path_ce, p)
mdata = c.ModelData(m_path)
gpn = MoTypes(True).network_model()
gpn.load(mdata.ModelDefinition, mdata.LastCheckpoint)
input_result = gpn.parse_layer_input_by_cluster(
't', 1, model_cids[layers_ids == 0], model_cids[layers_ids == 1])
for k, l0 in enumerate(np.arange(8)[l_0_mask]):
for l, l1 in enumerate(np.arange(8)[l_1_mask]):
conn_mat[l0, l1, i] = input_result[k, l]
# reordered version of conn_mat based on known types
cm_order = [1, 7, 0, 2, 3, 4, 5, 6]
cm_reordered = conn_mat[:, cm_order, :]
cm_reordered = cm_reordered[cm_order, :, :]
m = np.mean(cm_reordered, 2)
s = np.std(cm_reordered, 2)
cross_0 = np.sign((m + s) * (m - s)) <= 0
m[cross_0] = 0
s[cross_0] = 0
fig, axes = pl.subplots(nrows=4, sharex=True, sharey=True)