def bayes_save_to_csv(version, ansatz_name, size=3, minimizer='bayes'):
base_dir = join(ROOT_DIR, f'data/final_bayes/v{version}')
data_file1 = f'{ansatz_name}_{minimizer}_size={size}_part_{1}.pkl'
data_file2 = f'{ansatz_name}_{minimizer}_size={size}_part_{2}.pkl'
data_1 = data.load(data_file1, base_dir)[0]
data_2 = data.load(data_file2, base_dir)[0]
data_ = data_1+data_2
fel = np.zeros([36, 60])
samples_lst = []
max_meas_lst = []
nr = np.zeros([36, 60])
for i, y in enumerate(data_):
identifier, result = y
samples = identifier[6]
max_meas = identifier[5]
arr = np.asarray(np.abs(np.linspace(2750, 256500, 36) - samples))
samples_idx = np.argmin(arr)
max_meas_idx = np.argmin(np.abs(np.linspace(50000, 3e6, 60) - max_meas))
eig = result['correct']
fun_none = result['fun']
error = (eig - fun_none) / eig * 100
max_meas_lst.append(max_meas)
samples_lst.append(samples)
fel[samples_idx][max_meas_idx] +=error
nr[samples_idx][max_meas_idx] +=1
for j in range(36):
for k in range(60):
if np.any(nr[j,k] != 0):
fel[j,k]/=nr[j,k]
else:
fel[j,k] = 0
import csv
vec1, vec2, vec3 = [],[],[]
for i,sample in enumerate(np.linspace(2750, 256500, 36)):
for j, evals in enumerate(np.linspace(50000, 3e6, 60)):
vec1.append(evals)
vec2.append(sample)
if fel[i,j] == 0.0: vec3.append('nan')
else: vec3.append(np.log10(fel[i,j]))
with open('text.csv', 'w') as f:
writer = csv.writer(f, delimiter=' ')
writer.writerows(zip(vec1, vec2, vec3))
def NM_save(version, size, ansatz_name, minimizer):
base_dir = join(ROOT_DIR, f'data/final_nm/v{version}')
data_file = f'{ansatz_name}_{minimizer}_size={size}.pkl'
data_ = data.load(data_file, base_dir)[0]
fel = []
samples_lst = []
max_meas = np.linspace(50000, 3e6, 60)
nr = []
identifier_set = set()
for i, y in enumerate(data_):
print(f'Klar med {i+1}/{len(data_)}')
identifier, result = y
if identifier in identifier_set:
continue
identifier_set.add(identifier)
samples = identifier[6]
x = result['iteration_params_all']
fun = result['expectation_vals_all']
eig = result['correct']
fun_evals = np.round(max_meas / samples)
fun_evals[fun_evals == 0] = 1
fel_none = NM_fel_measmax.fel_measmax(x, fun, identifier, fun_evals)
error = (eig - fel_none) / eig * 100
if samples not in samples_lst:
samples_lst.append(samples)
print(samples_lst)
fel.append(error)
nr.append(1)
elif samples in samples_lst:
idx = np.argmin(np.abs(np.array(samples_lst) - samples))
fel[idx] = (fel[idx] + error)
nr[idx] += 1
print(nr)
fel = np.array(fel)
print(nr)
for i, rows in enumerate(fel):
fel[i, :] = rows / nr[i]
samples_lst = np.array(samples_lst)
sort_idx = np.argsort(samples_lst)
samples_lst = samples_lst[sort_idx]
fel = fel[sort_idx, :]
file = f'NM_heatmap/v{version}/{ansatz_name}_{minimizer}_size={size}.pkl'
data2 = max_meas, samples_lst, fel
data.save(file, data2, extract=True)
def NM_save(version, size, ansatz_name, matidx=None):
base_dir = join(ROOT_DIR, f'data/final_nm/v{version}')
data_file = f'{ansatz_name}_nelder-mead_size={size}.pkl'
data_ = data.load(data_file, base_dir)[0]
max_meas_lst = np.linspace(50000, 3e6, 60)
samples_lst = np.linspace(2750, 256500, 36)
fel = np.zeros([36, 60])
nr = np.zeros([36, 60])
identifier_set = set()
for y in data_:
identifier, result = y
samples = identifier[-2]
if identifier in identifier_set or samples > 256500:
continue
identifier_set.add(identifier)
i = np.argmin(np.abs(samples_lst - samples))
for j, max_meas in enumerate(max_meas_lst):
fun_evals = int(round(max_meas / samples))
if fun_evals <= 4:
continue
funs = np.array(result['expectation_vals_all'][:fun_evals])
params = np.array(result['iteration_params_all'][:fun_evals])
idx = np.argmin(funs)
x = params[idx, :]
idx_lst = np.linalg.norm(params - x, axis=1) <= 1e-3
fun = np.mean(funs[idx_lst])
eig = result['correct']
fel[i, j] += np.abs((fun - eig) / eig * 100)
nr[i, j] += 1
for rows in nr:
print(rows)
for i in range(36):
for j in range(60):
if nr[i, j] != 0:
fel[i, j] /= nr[i, j]
if matidx == None:
file = f'heatmap_data/v2/nm_{ansatz_name}_size={size}.pkl'
else:
file = f'heatmap_data/v2/nm_{ansatz_name}_size={size}_matidx={matidx}.pkl'
data2 = max_meas_lst, samples_lst, fel
data.save(file, data2, extract=True)
def _get_metadata(directory, warn=True, base=base_dir):
"""
Look for metadata in directory. Returns [], {} if not found.
:param directory: Directory to search through.
:return: Metadata and meta-metadata.
"""
try:
return data.load(join(directory, 'total', 'metadata'), base_dir=base)
except FileNotFoundError:
if warn:
warnings.warn('Metadata has not been initialized correctly, '
'run parallel.run with init=True and then with '
'cleanup=True.')
return [], {}
def bayes_save(version, size, ansatz_name, minimizer):
base_dir = join(ROOT_DIR, f'data/final_bayes/v{version}')
data_file = f'{ansatz_name}_{minimizer}_size={size}.pkl'
data_ = data.load(data_file, base_dir)[0]
fel = np.zeros([36, 60])
samples_lst = []
max_meas_lst = []
nr = np.zeros([36, 60])
for i, y in enumerate(data_):
identifier, result = y
samples = identifier[6]
max_meas = identifier[5]
arr = np.asarray(np.abs(np.linspace(2750, 256500, 36) - samples))
samples_idx = np.argmin(arr)
max_meas_idx = np.argmin(
np.abs(np.linspace(50000, 3e6, 60) - max_meas))
eig = result['correct']
fun_none = result['fun_none']
error = (eig - fun_none) / eig * 100
max_meas_lst.append(max_meas)
samples_lst.append(samples)
fel[samples_idx][max_meas_idx] += error
nr[samples_idx][max_meas_idx] += 1
for j in range(36):
for k in range(60):
if np.any(nr[j, k] != 0):
fel[j, k] /= nr[j, k]
else:
fel[j, k] = 5
print(fel.shape[0])
file = f'NM_heatmap/v{version}/{ansatz_name}_{minimizer}_size={size}.pkl'
data2 = np.linspace(50000, 3e6, 60), np.linspace(2750, 256500, 36), fel
data.save(file, data2, extract=True)
def bayes_save(version, size, ansatz_name, matidx=None):
base_dir = join(ROOT_DIR, f'data/final_bayes/v{version}')
data_file1 = f'{ansatz_name}_bayes_size={size}_part_1.pkl'
data_file2 = f'{ansatz_name}_bayes_size={size}_part_2.pkl'
data1 = data.load(data_file1, base_dir)[0]
data2 = data.load(data_file2, base_dir)[0]
data_ = data1 + data2
fel = np.zeros([36, 60])
nr = np.zeros([36, 60])
samples_lst = np.zeros([36])
max_meas_lst = np.zeros([60])
fun_evals_lst = np.zeros([36, 60])
identifier_set = set()
for i, y in enumerate(data_):
identifier, result = y
samples = identifier[-1]
max_meas = identifier[-2]
fun_evals = int(round(max_meas / samples))
if identifier in identifier_set or samples > 256500 or not (
matidx != None and matidx == identifier[4]):
continue
identifier_set.add(identifier)
samples_idx = np.argmin(np.abs(np.linspace(2750, 256500, 36) - samples))
max_meas_idx = np.argmin(np.abs(np.linspace(50000, 3e6, 60) - max_meas))
eig = result['correct']
fun = result['fun']
error = abs((eig - fun) / eig * 100)
max_meas_lst[max_meas_idx] = max_meas
samples_lst[samples_idx] = samples
fun_evals_lst[samples_idx, max_meas_idx] = fun_evals
if 4 < int(round(max_meas / samples)) <= 300:
fel[samples_idx, max_meas_idx] += error
nr[samples_idx, max_meas_idx] += 1
for i in range(36):
for j in range(60):
if nr[i, j] != 0:
fel[i, j] /= nr[i, j]
for i in range(36):
idx_dict = {}
for idx, fun_evals in enumerate(fun_evals_lst[i, :]):
if fun_evals not in idx_dict:
idx_dict[fun_evals] = []
idx_dict[fun_evals].append(idx)
for fun_evals, idx_lst in zip(idx_dict.keys(), idx_dict.values()):
fel[i, idx_lst] = np.mean(fel[i, idx_lst])
for row in nr:
print(row)
if matidx == None:
file = f'heatmap_data/v2/bayes_{ansatz_name}_size={size}.pkl'
else:
file = f'heatmap_data/v2/bayes_{ansatz_name}_size={size}_matidx={matidx}.pkl'
data2 = np.linspace(50000, 3e6, 60), np.linspace(2750, 256500, 36), fel
data.save(file, data2, extract=True)
version = 1
# TODO: select directory and basename of file to save to.
directory = 'test_recipe' # directory to save to
file = 'test' # file to save to (basename)
# Append version number to file
file += f'_v{version}'
# Metadata about what has been done previously will be saved here:
path_metadata = join(directory, file + '_metadata.pkl')
# Load/initialize metadata
try:
# Try to load the file (will raise FileNotFoundError if not existing)
metadata, metametadata = data.load(path_metadata)
except FileNotFoundError:
metadata = []
metametadata = {
'description':
"File that keeps track of what's been done "
"previously in this script "
f"({basename(__file__)})."
}
data.save(file=path_metadata,
data=metadata,
metadata=metametadata,
extract=True)
# Extract identifiers to previously completed simulations
ids = set()
def _cleanup_big(identifier_generator, directory, script_file):
"""
Cleanup directory by going trough all subdirectories, collect results and
fix metadata.
:param directory: Directory of cleanup.
:return:
"""
# TODO: collect metadata regarding time, success rate, etc. from
# mac-subdirs to total/metadata, save as dict with mac as key(?)
start_time = perf_counter()
# Get total/metadata
metadata, metametadata = _get_metadata(directory, warn=False)
if (metadata, metametadata) == ([], {}):
_init_metadata(identifier_generator, directory, script_file)
# Try again
metadata, metametadata = _get_metadata(directory, warn=False)
if (metadata, metametadata) == ([], {}):
raise ValueError("_init_metadata doesn't work")
# Convert metadata to dict
meta_dict = {}
for x in metadata:
# Keep only the last exception for given identifier.
if x[0] not in meta_dict or meta_dict[x[0]] is not True:
meta_dict[x[0]] = x[1]
del metadata
# Find mac-subdirectories
subdirs = set()
with os.scandir(join(base_dir, directory)) as it:
for entry in it:
if entry.is_dir() and entry.name.isdigit():
subdirs.add(join(base_dir, directory, entry.name))
# Add data/directory to subdirs to not overwrite data from other
# file-systems
data_dir = join(data.BASE_DIR, directory)
if isdir(data_dir):
subdirs.add(data_dir)
# Find data-files and keep track of which exists in which subdir
files = {}
for subdir in subdirs:
with os.scandir(subdir) as it:
for entry in it:
if entry.is_file() and entry.name != 'metadata.pkl':
if entry.name not in files:
files[entry.name] = []
files[entry.name].append(subdir)
# Go through files, create file in total (if not existing), add data
# from files in subdirs and update meta_dict
count = 0
metadata = None # To not crash at del metadata if no file.
for file in files:
print(f"\nSaving results in {join('total', file)}."
"\nUsing data from the following directories:")
# Load file from total
try:
content, metadata = data.load(file=join(directory, 'total', file),
base_dir=base_dir)
except FileNotFoundError:
content, metadata = [], {}
# Convert content to dict
content_dict = _add_result_to_dict(content, {})
del content
# Add content from other files
for subdir in files[file]:
print(subdir)
content_new, metadata_new = data.load(file=join(subdir, file),
base_dir='')
content_dict = _add_result_to_dict(content_new, content_dict)
# Change metadata if no previous
if metadata == {}:
metadata = metadata_new
metadata['created_by'] = data.get_name()
metadata['created_datetime'] = datetime.now().\
strftime("%Y-%m-%d, %H:%M:%S")
# Convert content_dict back to list and update meta_dict
content = []
for id_ in content_dict:
if id_ not in meta_dict or meta_dict[id_] is not True:
meta_dict[id_] = True
for result in content_dict[id_]:
count += 1
content.append([id_, result])
# Save file
data.save(file=join(directory, 'total', file),
data=content,
metadata=metadata,
extract=True,
base_dir=base_dir,
disp=False)
del metadata
print('\nFinishing cleanup.')
# Convert meta_dict back to list and save
metadata = []
for id_ in meta_dict:
metadata.append([id_, meta_dict[id_]])
data.save(file=join(directory, 'total', 'metadata'),
base_dir=base_dir,
data=metadata,
metadata=metametadata,
extract=True,
disp=False)
# Update metadata in subdirs
for subdir in subdirs:
try:
metametadata_new = data.load(file=join(subdir, 'metadata'),
base_dir='')[1]
except FileNotFoundError:
metametadata_new = metametadata
data.save(file=join(subdir, 'metadata'),
base_dir='',
data=metadata,
metadata=metametadata_new,
extract=True,
disp=False)
# Copy content of base_dir/directory/total to data.BASE_DIR/directory
destination = join(data.BASE_DIR, directory)
os.makedirs(destination, exist_ok=True)
with os.scandir(join(base_dir, directory, 'total')) as it:
for entry in it:
if entry.is_file():
shutil.copy(entry.path, destination)
remaining = sum(x is not True for x in meta_dict.values())
# Print some stats
stop_time = perf_counter()
print(f'\nCleanup completed in {stop_time - start_time:.1f} s.')
print(f'A total of {len(metadata)} identifiers where handled, {count} '
f'results saved and {remaining} tasks remaining.\n')
def _run_internal(simulate, identifier_generator, input_functions, directory,
script_file, file_from_id, metadata_from_id, num_workers,
start_range, stop_range, max_task, chunksize, debug):
"""
Internal parallel run.
:return: number of remaining tasks.
"""
# Files and paths
directory_nomac = directory
directory = join(directory, str(mac))
path_metadata = join(directory, 'metadata')
try:
# Try to load the file (will raise FileNotFoundError if not existing)
metadata, metametadata = data.load(path_metadata, base_dir=base_dir)
except FileNotFoundError:
metadata = _get_metadata(directory_nomac)[0]
metametadata = {
'description':
"File that keeps track of what's been done "
"previously in this script "
f"({script_file}).",
'run_time': [],
'num_workers': [],
'num_tasks_completed': [],
'success_rate': [],
'created_from':
script_file
}
data.save(file=path_metadata,
data=metadata,
metadata=metametadata,
extract=True,
base_dir=base_dir)
# Extract identifiers to previously completed simulations
ids = set()
for id_, value in metadata:
if value is True:
if id_ in ids:
raise KeyError('Multiple entries for same id in metadata.')
ids.add(id_)
# Cleanup (metadata can contain thousands of Exception objects)
del metadata, metametadata
# Wrap simulate to get expected input/output and handle exceptions
wrap = Wrap(simulate, debug=debug)
# Generator for pool
generator = Bookkeeper(identifier_generator, ids, input_functions,
[start_range, stop_range])
# Counters and such
files = set()
success = 0
fail = 0
start_time = perf_counter()
# Actual run
try:
with Pool(num_workers, maxtasksperchild=max_task) as p:
result_generator = p.imap_unordered(wrap,
generator,
chunksize=chunksize)
for identifier, result in result_generator:
# Handle exceptions:
if isinstance(result, Exception):
fail += 1
# Save the error
data.append(path_metadata, [identifier, result],
base_dir=base_dir)
else:
success += 1
file = file_from_id(identifier)
if file not in files:
files.add(file)
if not isfile(join(base_dir, directory,
file + '.pkl')):
# Create file
metadata = metadata_from_id(identifier)
data.save(join(directory, file), [],
metadata,
extract=True,
base_dir=base_dir)
data.append(join(directory, file), [identifier, result],
base_dir=base_dir)
# Mark the task as completed (last in the else,
# after saving result)
data.append(path_metadata, [identifier, True],
base_dir=base_dir)
except Exception as e:
if debug:
raise e
finally:
stop_time = perf_counter()
total = success + fail
if total == 0:
total = 1 # To avoid division by zero
# Post simulation.
metadata, metametadata = data.load(path_metadata, base_dir=base_dir)
metadata = _cleanup_small(metadata)
metametadata['run_time'].append(stop_time - start_time)
metametadata['num_workers'].append((num_workers, max_num_workers))
metametadata['num_tasks_completed'].append(success)
metametadata['success_rate'].append(success / total)
data.save(file=path_metadata,
data=metadata,
metadata=metametadata,
extract=True,
base_dir=base_dir)
# Print some stats
print('\nSimulation completed.')
print(f'Total number of tasks this far: {len(metadata)}')
print(f'Completed tasks this run: {success}')
print(f'Success rate this run: {success / total}')
remaining = len(metadata) - sum(x[1] for x in metadata if x[1] is True)
print(f'Minimum number of tasks remaining for this run: {fail}')
print(f'Total number of tasks remaining: {remaining}')
# No success and no fail => no restart
if not debug:
return success + fail