def test_compute_fixed_hash():
complex_obj = [1, 'd', {'a': 4, 'b': np.arange(10)}, (7, range(10))]
original_code = compute_fixed_hash(complex_obj)
assert compute_fixed_hash(complex_obj) == original_code == '6c98fabc301361863f321f6149a8a12a'
complex_obj[2]['b'][6]=0
assert compute_fixed_hash(complex_obj) == '8aee1f739fc9a612ed72e14682026627' != original_code
complex_obj[2]['b'][6]=6 # Revert to old value
assert compute_fixed_hash(complex_obj) == original_code
def test_compute_fixed_hash():
complex_obj = [1, 'd', {'a': 4, 'b': np.arange(10)}, (7, range(10))]
assert compute_fixed_hash(
complex_obj) == '285f0ff77c0cb6a5fc529ae55775f9dd'
complex_obj[2]['b'][6] = 0
assert compute_fixed_hash(
complex_obj) == '3124854bda236cd50f0cc04bd6f84935'
complex_obj[2]['b'][6] = 6 # Revert to old value
assert compute_fixed_hash(
complex_obj) == '285f0ff77c0cb6a5fc529ae55775f9dd'
def is_valid(self, last_run_args=None, current_args=None):
"""
:return: True if the experiment arguments have changed, otherwise false.
"""
if last_run_args is None:
last_run_args = dict(self.info.get_field(ExpInfoFields.ARGS))
if current_args is None:
current_args = dict(
load_experiment(
record_id_to_experiment_id(
self.get_identifier())).get_args())
return compute_fixed_hash(last_run_args) == compute_fixed_hash(
current_args)
def test_compute_fixed_hash_terminates():
a = []
b = [a]
a.append(b)
code = compute_fixed_hash(a)
assert code == 'cffaee424a62cd1893825a5811c34b8d'
c = []
d = [c]
c.append(d)
code = compute_fixed_hash(c)
assert code == 'cffaee424a62cd1893825a5811c34b8d'
def test_compute_fixed_hash_terminates():
a = []
b = [a]
a.append(b)
code = compute_fixed_hash(a)
assert code == 'cffaee424a62cd1893825a5811c34b8d'
c = []
d = [c]
c.append(d)
code = compute_fixed_hash(c)
assert code == 'cffaee424a62cd1893825a5811c34b8d'
def args_valid(self, last_run_args=None, current_args=None):
"""
:return: True if the experiment arguments have not changed
False if they have changed
None if it cannot be determined because arguments are not hashable objects.
"""
if last_run_args is None: # Cast to dict (from OrderedDict) because different arg order shouldn't matter
last_run_args = self.info.get_field(ExpInfoFields.ARGS) # A list of 2-tuples
if current_args is None:
current_args = dict(self.get_experiment().get_args())
try:
return compute_fixed_hash(dict(last_run_args), try_objects=True) == compute_fixed_hash(dict(current_args), try_objects=True)
except NotImplementedError: # Happens when we have unhashable arguments
return None
def test_compute_fixed_hash():
# Not really sure why the fixed hash differes between python 2 and 4 here (maybe something do do with changes to strings)
complex_obj = [1, 'd', {'a': 4, 'b': np.arange(10)}, (7, list(range(10)))]
original_code = compute_fixed_hash(complex_obj)
expected_code = 'c9b83dd2e1099c3bcbb05e3c69327c72' if _IS_PYTHON_3 else '6c98fabc301361863f321f6149a8a12a'
assert compute_fixed_hash(complex_obj) == original_code == expected_code
complex_obj[2]['b'][6] = 0
expected_code = 'a783b1f098fca8ccaf977a3123be5ac4' if _IS_PYTHON_3 else '8aee1f739fc9a612ed72e14682026627'
assert compute_fixed_hash(complex_obj) == expected_code != original_code
complex_obj[2]['b'][6] = 6 # Revert to old value
assert compute_fixed_hash(complex_obj) == original_code
def test_compute_fixed_hash():
# Not really sure why the fixed hash differes between python 2 and 4 here (maybe something do do with changes to strings)
complex_obj = [1, 'd', {'a': 4, 'b': np.arange(10)}, (7, list(range(10)))]
original_code = compute_fixed_hash(complex_obj)
expected_code = 'c9b83dd2e1099c3bcbb05e3c69327c72' if _IS_PYTHON_3 else '6c98fabc301361863f321f6149a8a12a'
assert compute_fixed_hash(complex_obj) == original_code == expected_code
complex_obj[2]['b'][6]=0
expected_code = 'a783b1f098fca8ccaf977a3123be5ac4' if _IS_PYTHON_3 else '8aee1f739fc9a612ed72e14682026627'
assert compute_fixed_hash(complex_obj) == expected_code != original_code
complex_obj[2]['b'][6]=6 # Revert to old value
assert compute_fixed_hash(complex_obj) == original_code
def get_function_hash_filename(fcn, argname_argvalue_list, create_dir_if_not = False):
args_code = compute_fixed_hash(argname_argvalue_list)
# TODO: Include function path in hash? Or module path, which would allow memos to be shareable.
full_path = os.path.join(get_memo_dir(fcn), '{}.pkl'.format(args_code, ))
if create_dir_if_not:
make_file_dir(full_path)
return full_path
def _get_record_rows_cached(record_id, headers, raise_display_errors, truncate_to, ignore_valid_keys = ()):
"""
We want to load the saved row only if:
- The record is complete
-
:param record_id:
:param headers:
:return:
"""
cache_key = compute_fixed_hash((record_id, [h.value for h in headers], truncate_to, ignore_valid_keys))
path = get_artemis_data_path(os.path.join('_ui_cache', cache_key), make_local_dir=True)
if os.path.exists(path):
try:
with open(path, 'rb') as f:
record_rows = pickle.load(f)
if len(record_rows)!=len(headers):
os.remove(path) # This should never happen. But in case it somehow does, we just go ahead and compute again.
else:
return record_rows
except:
logging.warn('Failed to load cached record info: {}'.format(record_id))
info_plus_status = _get_record_rows(record_id=record_id, headers=headers+[ExpRecordDisplayFields.STATUS],
raise_display_errors=raise_display_errors, truncate_to=truncate_to, ignore_valid_keys=ignore_valid_keys)
record_rows, status = info_plus_status[:-1], info_plus_status[-1]
if status == ExpStatusOptions.STARTED: # In this case it's still running (maybe) and we don't want to cache because it'll change
return record_rows
else:
with open(path, 'wb') as f:
pickle.dump(record_rows, f)
return record_rows
def _filter_experiments(user_range, exp_record_dict):
if user_range in exp_record_dict:
is_in = [k==user_range for k in exp_record_dict]
else:
number_range = interpret_numbers(user_range)
if number_range is not None:
# experiment_ids = [experiment_list[i] for i in number_range]
is_in = [i in number_range for i in xrange(len(exp_record_dict))]
elif user_range == 'all':
# experiment_ids = experiment_list
is_in = [True]*len(exp_record_dict)
elif user_range.startswith('has:'):
phrase = user_range[len('has:'):]
# experiment_ids = [exp_id for exp_id in experiment_list if phrase in exp_id]
is_in = [phrase in exp_id for exp_id in exp_record_dict]
elif user_range.startswith('1diff:'):
# select experiments whose arguments differ by one element from the selected experiments
base_range = user_range[len('1diff:'):]
base_range_exps = select_experiments(base_range, exp_record_dict) # list<experiment_id>
all_exp_args_hashes = {eid: set(compute_fixed_hash(a) for a in load_experiment(eid).get_args().items()) for eid in exp_record_dict} # dict<experiment_id : set<arg_hashes>>
# assert all_equal_length(all_exp_args_hashes.values()), 'All variants must have the same number of arguments' # Note: we diable this because we may have lists of experiments with different root functions.
is_in = [any(len(all_exp_args_hashes[eid].difference(all_exp_args_hashes[other_eid]))<=1 for other_eid in base_range_exps) for eid in exp_record_dict]
elif user_range.startswith('hasnot:'):
phrase = user_range[len('hasnot:'):]
# experiment_ids = [exp_id for exp_id in experiment_list if phrase not in exp_id]
is_in = [phrase not in exp_id for exp_id in exp_record_dict]
elif user_range in ('unfinished', 'invalid', 'corrupt'): # Return all experiments where all records are unfinished/invalid/corrupt
record_filters = _filter_records(user_range, exp_record_dict)
# experiment_ids = [exp_id for exp_id in experiment_list if len(record_filters[exp_id])]
is_in = [all(record_filters[exp_id]) for exp_id in exp_record_dict]
else:
raise Exception("Don't know how to use input '{}' to select experiments".format(user_range))
return OrderedDict((exp_id, exp_is_in) for exp_id, exp_is_in in izip_equal(exp_record_dict, is_in))
def _get_record_rows_cached(record_id, headers, raise_display_errors, truncate_to, ignore_valid_keys = ()):
"""
We want to load the saved row only if:
- The record is complete
-
:param record_id:
:param headers:
:return:
"""
cache_key = compute_fixed_hash((record_id, [h.value for h in headers], truncate_to, ignore_valid_keys))
path = get_artemis_data_path(os.path.join('_ui_cache', cache_key), make_local_dir=True)
if os.path.exists(path):
try:
with open(path, 'rb') as f:
record_rows = pickle.load(f)
if len(record_rows)!=len(headers):
os.remove(path) # This should never happen. But in case it somehow does, we just go ahead and compute again.
else:
return record_rows
except:
logging.warn('Failed to load cached record info: {}'.format(record_id))
info_plus_status = _get_record_rows(record_id=record_id, headers=headers+[ExpRecordDisplayFields.STATUS],
raise_display_errors=raise_display_errors, truncate_to=truncate_to, ignore_valid_keys=ignore_valid_keys)
record_rows, status = info_plus_status[:-1], info_plus_status[-1]
if status == ExpStatusOptions.STARTED: # In this case it's still running (maybe) and we don't want to cache because it'll change
return record_rows
else:
with open(path, 'wb') as f:
pickle.dump(record_rows, f)
return record_rows
def get_function_hash_filename(fcn, argname_argvalue_list, create_dir_if_not = False):
args_code = compute_fixed_hash(argname_argvalue_list)
# TODO: Include function path in hash? Or module path, which would allow memos to be shareable.
full_path = os.path.join(get_memo_dir(fcn), '{}.pkl'.format(args_code, ))
if create_dir_if_not:
make_file_dir(full_path)
return full_path
def args_valid(self, last_run_args=None, current_args=None):
"""
:param Optional[OrderedDict] last_run_args: The arguments from the last run
:param Optional[OrderedDict] current_args: The arguments from the current experiment in code
:return: True if the experiment arguments have not changed
False if they have changed
None if it cannot be determined because arguments are not hashable objects.
"""
if last_run_args is None: # Cast to dict (from OrderedDict) because different arg order shouldn't matter
last_run_args = self.get_args() # A list of 2-tuples
if any(isinstance(v, UnPicklableArg) for k, v in last_run_args.items()):
return None
if current_args is None:
current_args = dict(self.get_experiment().get_args())
try:
return compute_fixed_hash(dict(last_run_args), try_objects=True) == compute_fixed_hash(dict(current_args), try_objects=True)
except NotImplementedError: # Happens when we have unhashable arguments
return None
def _filter_experiments(user_range, exp_record_dict, return_is_in = False):
if user_range.startswith('~'):
is_in = _filter_experiments(user_range=user_range[1:], exp_record_dict=exp_record_dict, return_is_in=True)
is_in = [not r for r in is_in]
else:
if user_range in exp_record_dict:
is_in = [k==user_range for k in exp_record_dict]
else:
number_range = interpret_numbers(user_range)
if number_range is not None:
is_in = [i in number_range for i in xrange(len(exp_record_dict))]
elif user_range == 'all':
is_in = [True]*len(exp_record_dict)
elif user_range.startswith('has:'):
phrase = user_range[len('has:'):]
is_in = [phrase in exp_id for exp_id in exp_record_dict]
elif user_range.startswith('1diff:'):
base_range = user_range[len('1diff:'):]
base_range_exps = select_experiments(base_range, exp_record_dict) # list<experiment_id>
all_exp_args_hashes = {eid: set(compute_fixed_hash(a) for a in load_experiment(eid).get_args().items()) for eid in exp_record_dict} # dict<experiment_id : set<arg_hashes>>
is_in = [any(len(all_exp_args_hashes[eid].difference(all_exp_args_hashes[other_eid]))<=1 for other_eid in base_range_exps) for eid in exp_record_dict]
elif user_range.startswith('hasnot:'):
phrase = user_range[len('hasnot:'):]
is_in = [phrase not in exp_id for exp_id in exp_record_dict]
elif user_range in ('unfinished', 'invalid', 'corrupt'): # Return all experiments where all records are unfinished/invalid/corrupt
record_filters = _filter_records(user_range, exp_record_dict)
is_in = [all(record_filters[exp_id]) for exp_id in exp_record_dict]
elif user_range == 'started':
is_in = [has_experiment_record(exp_id) for exp_id in exp_record_dict]
else:
raise RecordSelectionError("Don't know how to use input '{}' to select experiments".format(user_range))
if return_is_in:
return is_in
else:
return OrderedDict((exp_id, exp_is_in) for exp_id, exp_is_in in izip_equal(exp_record_dict, is_in))
def test_dictarraylist():
# We build the same thing in four different ways
a = Duck()
a['a', 'aa1'] = 1
a['a', 'aa2'] = 2
a['b', 0, 'subfield1'] = 4
a['b', 0, 'subfield2'] = 5
a['b', 1, 'subfield1'] = 6
a['b', 1, 'subfield2'] = 7
a['b', 2, 'subfield2'] = 9 # Note: Out of order assignment here!
a['b', 2, 'subfield1'] = 8
b = Duck.from_struct(a.to_struct())
assert a.to_struct() == b.to_struct()
c = Duck()
c['a', :] = OrderedDict([('aa1', 1), ('aa2', 2)])
c['b', next, :] = OrderedDict([('subfield1', 4), ('subfield2', 5)])
c['b', next, :] = OrderedDict([('subfield1', 6), ('subfield2', 7)])
c['b', next, :] = OrderedDict([('subfield2', 9), ('subfield1', 8)])
d = Duck()
d['a', :] = OrderedDict([('aa1', 1), ('aa2', 2)])
d['b', :, :] = [
OrderedDict([('subfield1', 4), ('subfield2', 5)]),
OrderedDict([('subfield1', 6), ('subfield2', 7)]),
OrderedDict([('subfield2', 9), ('subfield1', 8)])
]
e = Duck()
e['a', ...] = OrderedDict([('aa1', 1), ('aa2', 2)])
e['b', ...] = [
OrderedDict([('subfield1', 4), ('subfield2', 5)]),
OrderedDict([('subfield1', 6), ('subfield2', 7)]),
OrderedDict([('subfield2', 9), ('subfield1', 8)])
]
f = a[:]
g = a[:, :]
for i, same_struct in enumerate([a, b, c, d, e, f, g]):
print('Test {}'.format(i))
assert a.to_struct() == same_struct.to_struct()
assert a == same_struct
assert compute_fixed_hash(a.to_struct()) == compute_fixed_hash(
b.to_struct())
assert same_struct['a', 'aa2'] == 2
assert list(same_struct['b', 1, :].values()) == [6, 7]
assert list(same_struct['b', :, 'subfield1'].values()) == [4, 6, 8]
assert same_struct['b', :, :].deepvalues() == [[4, 5], [6, 7], [
9, 8
]] # Note that the order-switching gets through.
assert same_struct['b', :,
['subfield1', 'subfield2']].deepvalues() == [[4, 5],
[6, 7],
[8, 9]]
assert same_struct['b', :, :].to_struct() == [
OrderedDict([('subfield1', 4), ('subfield2', 5)]),
OrderedDict([('subfield1', 6), ('subfield2', 7)]),
OrderedDict([('subfield2', 9), ('subfield1', 8)])
]
assert same_struct['b', 1:, :].to_struct() == [
OrderedDict([('subfield1', 6), ('subfield2', 7)]),
OrderedDict([('subfield2', 9), ('subfield1', 8)])
]
with pytest.raises(KeyError):
x = same_struct['a', 'aa3']
with pytest.raises(KeyError):
x = same_struct['a', 1, 'subfield1']
with pytest.raises(IndexError):
x = same_struct['b', 4, 'subfield1']
with pytest.raises(
KeyError
): # This should raise an error because key 'a' does not have subkeys 1, 'subfield1'
x = same_struct[:, 1, 'subfield1']
assert np.array_equal(
same_struct['b'].to_array(),
[[4, 5], [6, 7], [8, 9]]) # Note... the order is corrected.
with pytest.raises(InvalidKeyError):
assert np.array_equal(
same_struct.to_array()) # Note... the order is corrected.
new_struct = same_struct.arrayify_axis(axis=1, subkeys=('b', ))
assert np.array_equal(new_struct['b', 'subfield1'], [4, 6, 8])
assert np.array_equal(new_struct['b', 'subfield2'], [5, 7, 9])
def test_dictarraylist():
# We build the same thing in four different ways
a = Duck()
a['a', 'aa1'] = 1
a['a', 'aa2'] = 2
a['b', 0, 'subfield1'] = 4
a['b', 0, 'subfield2'] = 5
a['b', 1, 'subfield1'] = 6
a['b', 1, 'subfield2'] = 7
a['b', 2, 'subfield2'] = 9 # Note: Out of order assignment here!
a['b', 2, 'subfield1'] = 8
b = Duck.from_struct(a.to_struct())
assert a.to_struct() == b.to_struct()
c = Duck()
c['a', :] = OrderedDict([('aa1', 1), ('aa2', 2)])
c['b', next, :] = OrderedDict([('subfield1', 4), ('subfield2', 5)])
c['b', next, :] = OrderedDict([('subfield1', 6), ('subfield2', 7)])
c['b', next, :] = OrderedDict([('subfield2', 9), ('subfield1', 8)])
d = Duck()
d['a', :] = OrderedDict([('aa1', 1), ('aa2', 2)])
d['b', :, :] = [OrderedDict([('subfield1', 4), ('subfield2', 5)]), OrderedDict([('subfield1', 6), ('subfield2', 7)]), OrderedDict([('subfield2', 9), ('subfield1', 8)])]
e = Duck()
e['a', ...] = OrderedDict([('aa1', 1), ('aa2', 2)])
e['b', ...] = [OrderedDict([('subfield1', 4), ('subfield2', 5)]), OrderedDict([('subfield1', 6), ('subfield2', 7)]), OrderedDict([('subfield2', 9), ('subfield1', 8)])]
f = a[:]
g = a[:, :]
for i, same_struct in enumerate([a, b, c, d, e, f, g]):
print ('Test {}'.format(i))
assert a.to_struct()==same_struct.to_struct()
assert a==same_struct
assert compute_fixed_hash(a.to_struct()) == compute_fixed_hash(b.to_struct())
assert same_struct['a', 'aa2'] == 2
assert list(same_struct['b', 1, :].values()) == [6, 7]
assert list(same_struct['b', :, 'subfield1'].values()) == [4, 6, 8]
assert same_struct['b', :, :].deepvalues() == [[4, 5], [6, 7], [9, 8]] # Note that the order-switching gets through.
assert same_struct['b', :, ['subfield1', 'subfield2']].deepvalues() == [[4, 5], [6, 7], [8, 9]]
assert same_struct['b', :, :].to_struct() == [OrderedDict([('subfield1', 4), ('subfield2', 5)]), OrderedDict([('subfield1', 6), ('subfield2', 7)]), OrderedDict([('subfield2', 9), ('subfield1', 8)])]
assert same_struct['b', 1:, :].to_struct() == [OrderedDict([('subfield1', 6), ('subfield2', 7)]), OrderedDict([('subfield2', 9), ('subfield1', 8)])]
with pytest.raises(KeyError):
x = same_struct['a', 'aa3']
with pytest.raises(KeyError):
x = same_struct['a', 1, 'subfield1']
with pytest.raises(IndexError):
x = same_struct['b', 4, 'subfield1']
with pytest.raises(KeyError): # This should raise an error because key 'a' does not have subkeys 1, 'subfield1'
x = same_struct[:, 1, 'subfield1']
assert np.array_equal(same_struct['b'].to_array(), [[4, 5], [6, 7], [8, 9]]) # Note... the order is corrected.
with pytest.raises(InvalidKeyError):
assert np.array_equal(same_struct.to_array()) # Note... the order is corrected.
new_struct = same_struct.arrayify_axis(axis=1, subkeys=('b',))
assert np.array_equal(new_struct['b', 'subfield1'], [4, 6, 8])
assert np.array_equal(new_struct['b', 'subfield2'], [5, 7, 9])
def _filter_experiments(user_range, exp_record_dict, return_is_in=False):
if '|' in user_range:
is_in = [
any(xs) for xs in zip(*(_filter_experiments(
subrange, exp_record_dict, return_is_in=True)
for subrange in user_range.split('|')))
]
elif '&' in user_range:
is_in = [
all(xs) for xs in zip(*(_filter_experiments(
subrange, exp_record_dict, return_is_in=True)
for subrange in user_range.split('&')))
]
elif user_range.startswith('~'):
is_in = _filter_experiments(user_range=user_range[1:],
exp_record_dict=exp_record_dict,
return_is_in=True)
is_in = [not r for r in is_in]
else:
if user_range in exp_record_dict:
is_in = [k == user_range for k in exp_record_dict]
else:
number_range = interpret_numbers(user_range)
if number_range is not None:
is_in = [
i in number_range for i in xrange(len(exp_record_dict))
]
elif user_range == 'all':
is_in = [True] * len(exp_record_dict)
elif user_range.startswith('has:'):
phrase = user_range[len('has:'):]
is_in = [phrase in exp_id for exp_id in exp_record_dict]
elif user_range.startswith('tag:'):
tag = user_range[len('tag:'):]
is_in = [
tag in load_experiment(exp_id).get_tags()
for exp_id in exp_record_dict
]
elif user_range.startswith('1diff:'):
base_range = user_range[len('1diff:'):]
base_range_exps = select_experiments(
base_range, exp_record_dict) # list<experiment_id>
all_exp_args_hashes = {
eid: set(
compute_fixed_hash(a)
for a in load_experiment(eid).get_args().items())
for eid in exp_record_dict
} # dict<experiment_id : set<arg_hashes>>
is_in = [
any(
len(all_exp_args_hashes[eid].difference(
all_exp_args_hashes[other_eid])) <= 1
for other_eid in base_range_exps)
for eid in exp_record_dict
]
elif user_range.startswith('hasnot:'):
phrase = user_range[len('hasnot:'):]
is_in = [phrase not in exp_id for exp_id in exp_record_dict]
elif user_range in (
'unfinished', 'invalid', 'corrupt'
): # Return all experiments where all records are unfinished/invalid/corrupt
record_filters = _filter_records(user_range, exp_record_dict)
is_in = [
all(record_filters[exp_id]) for exp_id in exp_record_dict
]
elif user_range == 'started':
is_in = [
has_experiment_record(exp_id) for exp_id in exp_record_dict
]
else:
raise RecordSelectionError(
"Don't know how to use input '{}' to select experiments".
format(user_range))
if return_is_in:
return is_in
else:
return OrderedDict(
(exp_id, exp_is_in)
for exp_id, exp_is_in in izip_equal(exp_record_dict, is_in))
def get_function_hash_filename(fcn, argname_argvalue_list):
args_code = compute_fixed_hash(argname_argvalue_list)
# TODO: Include function path in hash? Or module path, which would allow memos to be shareable.
return os.path.join(MEMO_DIR, '%s-%s.pkl' % (fcn.__name__, args_code))
