def main(argv=None):
if(argv==None):
argv = sys.argv
options = options_desc.parse_args(argv)[0]
assert(not(options.refine_all and options.extend_all))
pool = Pool()
needy_nodes = pool.where("isa_partition and is_sampled").multilock()
# 1. Trying to detect fake convergence
for n in pool.where("state == 'converged'"):
means = kmeans(n.trajectory, k=2)
d = (means[0] - means[1]).norm2()
if(d > 2.0 and (options.refine_all or userinput("%s has converged but appears to have a bimodal distribution.\nDo you want to refine?"%n.name, "bool"))): #TODO decide upon threshold (per coordinate?)
refine(n, options)
# 2. Dealing with not-converged nodes
for n in pool.where("state == 'not-converged'"):
if(not(options.refine_all or options.extend_all)):
choice = userchoice("%s has not converged. What do you want to do?"%n.name, ['_refine', '_extend', '_ignore'])
if(options.refine_all or choice=="r"):
refine(n, options)
elif(options.extend_all or choice=="e"):
extend(n)
elif(choice=="i"):
continue
for n in needy_nodes:
n.save()
n.unlock()
zgf_setup_nodes.main()
zgf_grompp.main()
zgf_cleanup.main()
def main(argv=None):
if (argv == None):
argv = sys.argv
options = options_desc.parse_args(argv)[0]
assert (not (options.refine_all and options.extend_all))
pool = Pool()
needy_nodes = pool.where("isa_partition and is_sampled").multilock()
# 1. Trying to detect fake convergence
for n in pool.where("state == 'converged'"):
means = kmeans(n.trajectory, k=2)
d = (means[0] - means[1]).norm2()
if (d > 2.0 and (options.refine_all or userinput(
"%s has converged but appears to have a bimodal distribution.\nDo you want to refine?"
% n.name,
"bool"))): #TODO decide upon threshold (per coordinate?)
refine(n, options)
# 2. Dealing with not-converged nodes
for n in pool.where("state == 'not-converged'"):
if (not (options.refine_all or options.extend_all)):
choice = userchoice(
"%s has not converged. What do you want to do?" % n.name,
['_refine', '_extend', '_ignore'])
if (options.refine_all or choice == "r"):
refine(n, options)
elif (options.extend_all or choice == "e"):
extend(n)
elif (choice == "i"):
continue
for n in needy_nodes:
n.save()
n.unlock()
zgf_setup_nodes.main()
zgf_grompp.main()
zgf_cleanup.main()
def main():
options = options_desc.parse_args(sys.argv)[0]
pool = Pool()
active_nodes = pool.where("isa_partition")
if options.transition_level == "clusters":
npz_file = np.load(pool.chi_mat_fn)
chi_matrix = npz_file['matrix']
n_clusters = npz_file['n_clusters']
default_cluster_threshold = options.coreset_power
# determine cluster
#TODO this part is too cryptic
# amount_phi[j] = amount of basis functions per cluster j
amount_phi=np.ones(n_clusters,dtype=np.uint64)
amount_phi=amount_phi*len(chi_matrix)
amount_phi_total=len(chi_matrix)
# sort columns of chi and return new sorted args
arg_sort_cluster=np.argsort(chi_matrix,axis=0)
# sort columns of chi and return new sorted chi
# notice that the last row has to be [1 ... 1]
sort_cluster=np.sort(chi_matrix,axis=0)
# show_cluster contains arrays of the type [a b] where a is the row
# and b the column of the entry from chi matrix where
# chi_sorted(a,b) > default_cluster_threshold
show_cluster=np.argwhere(sort_cluster > 0.5 )
# from the above it could be clear
# that the amount of phi function
# of cluster i is given by x where x the number so that
# [x i] is in show_cluster and for all
# [y i] in show_cluster we have x>y
# we define amount_phi[i]=x
for element in show_cluster:
index=element[0]
cluster=element[1]
if amount_phi[cluster]>index:
amount_phi[cluster]=index
# create cluster list which contains arrays
# each array consinst of a set of numbers corresponding to
# the phi function of node_number
cluster=[]
for i in range(0,n_clusters):
cluster_set=[]
for j in range(amount_phi[i],amount_phi_total):
#if (j < amount_phi[i] + 3):
cluster_set.append(arg_sort_cluster[j][i])
cluster.append(cluster_set)
for i in range(len(cluster)):
counter = 0
for node_index in cluster[i]:
counter += 1
# and ignore nodes which have a higher chi value then default_cluster_threshold
if( chi_matrix[node_index][i] > default_cluster_threshold and counter>options.min_nodes):
continue
node = active_nodes[node_index]
trajectory= node.trajectory
print "-----"
print "Generating transition nodes for node %s..."%node.name
neighbour_frames = get_indices_equidist(node, options.num_tnodes)
# create transition node for node_index
for frame_number in neighbour_frames:
print "Using frame %d as starting configuration."%frame_number
n = Node()
n.parent_frame_num = frame_number
n.parent = node
n.state = "created"
n.extensions_counter = 0
n.extensions_max = options.num_runs-1
n.extensions_length = options.sampling_length
n.sampling_length = options.sampling_length
n.internals = trajectory.getframe(frame_number)
n.save_mode = options.save_mode
pool.append(n)
n.save()
print "%d transition nodes generated."%options.num_tnodes
print "-----"
zgf_setup_nodes.main()
zgf_grompp.main()
cluster_dict = {}
for (ic,c) in enumerate(cluster):
cluster_dict['cluster_%d'%ic] = c
# save cluster
np.savez(pool.analysis_dir+"core_set_cluster.npz", **cluster_dict)
elif options.transition_level == "nodes":
for node in active_nodes:
trajectory= node.trajectory
# TODO duplicate code... use the one above
print "-----"
print "Generating transition nodes for node %s..."%node.name
neighbour_frames = get_indices_equidist(node, options.num_tnodes)
# create transition point for node_index
for frame_number in neighbour_frames:
print "Using frame %d as starting configuration."%frame_number
n = Node()
n.parent_frame_num = frame_number
n.parent = node
n.state = "created"
n.extensions_counter = 0
n.extensions_max = options.num_runs-1
n.extensions_length = options.sampling_length
n.sampling_length = options.sampling_length
n.internals = trajectory.getframe(frame_number)
n.save_mode = options.save_mode
pool.append(n)
n.save()
print "%d transition nodes generated."%options.num_tnodes
print "-----"
zgf_setup_nodes.main()
zgf_grompp.main()
instructionFile = pool.analysis_dir+"instruction.txt"
f = open(instructionFile, "w")
f.write("{'power': %f, 'tnodes': %d, 'level': '%s', 'min_nodes': %d}"%(options.coreset_power, options.num_tnodes, options.transition_level, options.min_nodes))
f.close()