Domain specific language for configuration spaces in Python/Cython. Useful for hyperparameter optimization and algorithm configuration.
Home Page: https://automl.github.io/ConfigSpace/
License: Other
Hi,
This is more of a question than an issue; sorry if this is the wrong place to post.
I have been working on extending auto-sklearn to include other types of components, and several of these have regularization penalty-esque hyperparameters. For similar types of hyperparameters (e.g., alpha and the thetas for Gaussian processes), the log option is always set to True when creating the respective ConfigSpace hyperparameters.
Does this result in the optimizer using a type of logarithmic search space for the respective hyperparameter (e.g., I just evaluated d=0.11 now consider something like d=0.011 or d=1.1 rather than d=0.12)? Also, are there better/other ways to specify these types of hyperparameters?
Have a good day,
Brandon
I noticed that there is a problem with ordinal hyperparameters.
I do not know if it is a bug or a feature but they do not behave like
integer hyperparameters.
Here is an example of what happens:
Define ConfigSpace in the .pcs file:
batch_size integer [8, 128] [64] log
sequence_size ordinal {128, 256, 512, 1024, 2048, 4096} [512]
Load this ConfigSpace and then sample random values and convert to a vector:
config_space = ... # Some loading code
sample_vector = config_space.sample_configuration().get_array()
#Corresponding Array:
#[ 0.40941199 2. ]
Now convert back to the Configuration:
sample = Configuration(config_space, vector=sample_vector)
Works nicely:
batch_size, Value: 24
sequence_size, Value: '512'*
Add some noise to the vector (BOHB does that when sampling from the model,
it treats ordinal, integer and continuous parameters in the same way)
sample_vector[0] += 0.1
sample_vector[1] += 0.1
#Corresponding Array:
#[ 0.50941199 2.1 ]
Now convert back to the Configuration:
sample = Configuration(config_space, vector=sample_vector)
For some reason sequence_size disappears.
Configuration:
batch_size, Value: 32
Is that intended? Shouldn't ordinal hyperparameter have the same representation as integer hyperparameter when transformed to an array? Now it has representation that is similar to categorical hyperparameter.
Running SMAC3 on the algorithm SparrowToRiss, I get sometimes the following error:
File "/work/ws/nemo/fr_tl1023-aclib-0/aclib2-master/configurators/smac3_warm/smac/optimizer/smbo.py", line 294, in _get_next_by_local_search
configuration, acq_val = self.acq_optimizer.maximize(start_point)
File "/work/ws/nemo/fr_tl1023-aclib-0/aclib2-master/configurators/smac3_warm/smac/optimizer/local_search.py", line 94, in maximize
for neighbor in all_neighbors:
File "/home/fr/fr_fr/fr_tl1023/anaconda3/lib/python3.5/site-packages/ConfigSpace/util.py", line 247, in get_one_exchange_neighbourhood
new_configuration.is_valid_configuration()
File "/home/fr/fr_fr/fr_tl1023/anaconda3/lib/python3.5/site-packages/ConfigSpace/configuration_space.py", line 1043, in is_valid_configuration
self._vector, allow_inactive_with_values=self.allow_inactive_with_values)
File "/home/fr/fr_fr/fr_tl1023/anaconda3/lib/python3.5/site-packages/ConfigSpace/configuration_space.py", line 665, in _check_configuration
hyperparameter))
ValueError: Hyperparameter instantiation '1.00000000001' (type: <class 'numpy.float64'>) is illegal for hyperparameter rate-limit, Type: UniformInteger, Range: [900000000, 18000000000], Default: 9000000000, on log-scale
The corresponding parameter is
rate-limit [900000000,18000000000][9000000000]il
Please note that I was able to run SMAC3 on this target algorithm previously (roughly 4 months ago), so the bug was introduced recently.
I have an issue with installing the new configspace version on Nemo (using anaconda).
Even using a complete newly installed anaconda, I get the following error.
Any ideas?
$ python
Python 3.6.3 |Anaconda, Inc.| (default, Oct 13 2017, 12:02:49)
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import ConfigSpace
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/fr/fr_fr/fr_tl1023/anaconda3/lib/python3.6/site-packages/ConfigSpace/__init__.py", line 33, in <module>
from ConfigSpace.configuration_space import Configuration, \
ImportError: /home/fr/fr_fr/fr_tl1023/anaconda3/lib/python3.6/site-packages/ConfigSpace/configuration_space.cpython-36m-x86_64-linux-gnu.so: undefined symbol: __intel_sse2_strchr
One function that I commonly use within the OpenML spheres is 'filter setups according to config space'. For this I utilize the ConfigSpace library. I download a list of setups, give it a config space object and only keep the setups that are within this config space. I created a (non-perfect) version of it yesterday in the openml-python-contrib library:
https://github.com/openml/openml-python-contrib/blob/master/openmlcontrib/setups/functions.py
One thing that I couldn't figure out in the documentation of ConfigSpace: how to take into account hyperparameter conditionals. i.e., the libsvm_svc degree is only relevant when the poly kernel is selected. This knowledge is usually available in the Config Space, I just don't know how to extract it.
Hi
I really like the concept of configspace. However I find it is hard to use it and further extend it.
I only find the examples at https://automl.github.io/SMAC3
Is there better documentation before I jump into the source code?
Thanks
This is not really an issue but rather an "feature" request..
Could you write any documentation about how we could use this module?
My Goal:
I hope i can use this module combined with auto-sklearn to export the (module + parameter) configuration (preferably to an database). So i can load the configuration (from the Database) and train a model more dynamically.
as reported by @cicero34 in automl/auto-sklearn#384 (comment)
Is there anywhere documentation or examples how to do this? Currently I am getting the following error:
Traceback (most recent call last):
File "/home/janvanrijn/projects/sklearn-bot/examples/run_on_study.py", line 54, in <module>
run()
File "/home/janvanrijn/projects/sklearn-bot/examples/run_on_study.py", line 38, in run
configuration_space = sklearnbot.config_spaces.get_config_space(args.classifier_name, None)
File "/home/janvanrijn/projects/sklearn-bot/sklearnbot/config_spaces/bootstrap.py", line 46, in get_config_space
return getattr(sklearnbot.config_spaces, classifier_name).get_hyperparameter_search_space(seed)
File "/home/janvanrijn/projects/sklearn-bot/sklearnbot/config_spaces/neural_network.py", line 104, in get_hyperparameter_search_space
cs.add_condition(nesterovs_momentum_confition_momentum)
File "ConfigSpace/configuration_space.py", line 187, in ConfigSpace.configuration_space.ConfigurationSpace.add_condition
File "ConfigSpace/configuration_space.py", line 273, in ConfigSpace.configuration_space.ConfigurationSpace._check_condition
ValueError: Adding a second condition (different) for a hyperparameter is ambigouos and therefore forbidden. Add a conjunction instead!
Already inserted: mlpclassifier__nesterovs_momentum | mlpclassifier__solver == 'sgd'
New one: mlpclassifier__nesterovs_momentum | mlpclassifier__momentum > 0
Whenever i set the default value lesser than 1e-10 for UniformFloatHyperparameter it raises ValueError
`ValueError Traceback (most recent call last)
in ()
25 # For example, gamma can be either "auto" or a fixed float
26 gamma = CategoricalHyperparameter("gamma", ["auto", "value"], default_value="auto") # only rbf, poly, sigmoid
---> 27 gamma_value = UniformFloatHyperparameter("gamma_value", 1e-15, 8, default_value=1e-11)
28 cs.add_hyperparameters([gamma, gamma_value])
29
ConfigSpace/hyperparameters.pyx in ConfigSpace.hyperparameters.UniformFloatHyperparameter.init()
ConfigSpace/hyperparameters.pyx in ConfigSpace.hyperparameters.UniformFloatHyperparameter.check_default()
ValueError: Illegal default value 0.0`
Not sure if i am doing something wrong
Hi,
the ConfigSpace throws an unexpected error for random configurations. Here an example script (using the spear_qcp example in SMAC3:
import sys
from ConfigSpace.configuration_space import ConfigSpace
from ConfigSpace.io.pcs import read
# see SMAC3 spear_qcp example
with open("target_algorithm/spear-python/spear-params-mixed.pcs") as fp:
pcs = fp.readlines()
cs = read(pcs_string=pcs)
for i in range(10000):
config = cs.sample_configuration()
#print(config)
for p in config:
try:
print(config[p])
except KeyError:
print(p)
print(config)
sys.exit(1)
The output is:
KeyError
sp-clause-inversion
Configuration:
sp-clause-activity-inc, Value: 0.5245793579385227
sp-clause-decay, Value: 1.793762045021577
sp-clause-del-heur, Value: '2'
sp-first-restart, Value: 124
sp-learned-clause-sort-heur, Value: '7'
sp-learned-clauses-inc, Value: 1.172033997030427
sp-learned-size-factor, Value: 0.7146238837315566
sp-max-res-lit-inc, Value: 1.93385606282595
sp-max-res-runs, Value: 25
sp-orig-clause-sort-heur, Value: '15'
sp-phase-dec-heur, Value: '5'
sp-rand-phase-dec-freq, Value: '0.05'
sp-rand-phase-scaling, Value: 0.6488652050952217
sp-rand-var-dec-freq, Value: '0'
sp-res-cutoff-cls, Value: 3
sp-res-cutoff-lits, Value: 180
sp-res-order-heur, Value: '8'
sp-resolution, Value: '1'
sp-restart-inc, Value: 1.3726376426746159
sp-update-dec-queue, Value: '0'
sp-use-pure-literal-rule, Value: '0'
sp-var-activity-inc, Value: 0.512447739865641
sp-var-dec-heur, Value: '16'
sp-variable-decay, Value: 1.4303654243518888
Traceback (most recent call last):
File "/home/lindauer/git/SMAC3/examples/spear_qcp/test_pcs.py", line 18, in <module>
print(p, config[p])
File "/home/lindauer/bin/anaconda3/lib/python3.5/site-packages/ConfigSpace-0.3.7-py3.5.egg/ConfigSpace/configuration_space.py", line 1053, in __getitem__
raise KeyError()
KeyError
The weird think is that the error is not raised if I first print the configuration (see commented print within for-loop)
The function should return np.inf if at least one float hyperparameter is in the configuration space, otherwise it should count the number of possible configurations.
Currently, the type of a hyperparameter has to be correct when creating a configuration from a dictionary, for example it needs to be a float when creating a float hyperparameter. This is especially cumbersome for Constants (their type can't be easily determined by isinstance). The Configuration object should convert the strings to the correct types.
from ConfigSpace.configuration_space import ConfigurationSpace
from ConfigSpace.hyperparameters import CategoricalHyperparameter
from ConfigSpace.conditions import InCondition
from ConfigSpace.read_and_write.pcs_new import write
cs = ConfigurationSpace(seed=12345)
p1 = CategoricalHyperparameter(name="p1", choices=[True, False], default_value=True)
p2 = CategoricalHyperparameter(name="p2", choices=[True, False], default_value=True)
cs.add_hyperparameters([p1,p2])
c1 = InCondition(child=p2, parent=p1, values=[True])
cs.add_condition(c1)
with open("test.pcs", "w") as fh:
fh.write(write(cs))
Error:
Traceback (most recent call last):
File "test.py", line 16, in <module>
fh.write(write(cs))
File "/home/lindauer/git/ConfigSpace/ConfigSpace/read_and_write/pcs_new.py", line 462, in write
condition_lines.write(build_condition(condition))
File "/home/lindauer/git/ConfigSpace/ConfigSpace/read_and_write/pcs_new.py", line 145, in build_condition
", ".join(condition.values))
TypeError: sequence item 0: expected str instance, bool found
In response to automl/SMAC3#442
Constants can not be a boolean, neither can they be None. Is there any reason for this? I think there are many cases where we want them to be either of these.
Traceback (most recent call last):
File "/home/janvanrijn/projects/sklearn-bot/examples/run_on_task.py", line 39, in <module>
run()
File "/home/janvanrijn/projects/sklearn-bot/examples/run_on_task.py", line 32, in run
configuration_space = sklearnbot.config_spaces.get_config_space(args.classifier, None)
File "/home/janvanrijn/projects/sklearn-bot/sklearnbot/config_spaces/bootstrap.py", line 19, in get_config_space
return sklearnbot.config_spaces.decision_tree.get_hyperparameter_search_space(seed)
File "/home/janvanrijn/projects/sklearn-bot/sklearnbot/config_spaces/decision_tree.py", line 21, in get_hyperparameter_search_space
'max_leaf_nodes', None, meta={'component': 'decisiontreeclassifier'})
File "ConfigSpace/hyperparameters.pyx", line 122, in ConfigSpace.hyperparameters.Constant.__init__
TypeError: Constant value is of type <class 'NoneType'>, but only the following types are allowed: (<class 'int'>, <class 'float'>, <class 'str'>)
Hi,
Is there a way to add a further value to an already existing parameter?
With some hacks, it was possible before 0.4.0; but after the introduction of Cython, I cannot access all attributes of a parameter anymore.
Best,
Marius
The current code suggests that something like
x | y > 5 in {1, 2, 3 && z in 4, 5, 6}
which should be forbidden according to the specs.
whenever it is changed, so other applications can easily and cheaply keep track of the current state of the configuration space (calling hash() every time a configuration space comes by might be too expensive).
Hi,
Is it possible to combine two configuration?
in get_one_exchange_neighbourhood when ForbiddenValueError is caught instead of ValueError most of searchspace tests fail.
Hi,
I see that the code contains large amount of type hints.
This is useful for static analyzers, but as far as I know
python doesn't enforce anything runtime.
You might consider to use PyContracts.
https://pypi.python.org/pypi/PyContracts
In most simple cases, it can use the same type hints from,
and check them runtime.
On the other hand, pycontracts is quite powerful,
and you can define more fine grained conditions,
like a list must have at least 3 item long, or
a file path must exist, etc.
If it slows the code down, (it should not),
then all the checks can be disabled with a single call.
Hi,
Is it possible to reset the default values of a ConfigSpace?
Overview of missing parts of the documentation:
guys,
Why does configspace.io does not work?
If I use get_parents_of() of the ConfigSpace on @1:counter-bump from /data/aad/aclib/target_algorithms/sat/clasp-cssc14/clasp-sat-params-nat.pcs,
I get the following exception:
File "/home/lindauer/.local/lib/python3.5/site-packages/ConfigSpace-0.2.0.dev0-py3.5.egg/ConfigSpace/configuration_space.py", line 393, in get_parents_of
parents.append(condition.parent)
AttributeError: 'AndConjunction' object has no attribute 'parent'
Using the deactivate_inactive_hyperparameters() fails in the following example:
from ConfigSpace.io import pcs
from ConfigSpace.util import deactivate_inactive_hyperparameters
with open("/data/aad/aclib2/target_algorithms/sat/satenstein/params.pcs") as fp:
pcs_str = fp.readlines()
cs = pcs.read(pcs_str)
config = {'CONadaptpromwalkprob': '0', 'CONadaptwalkprob': 0, 'CONdp': '0.05', 'CONpromdp': '0.05', 'CONpromwp': '0.01', 'CONselectclause': '1', 'CONwp': '0.01', 'DLSalpha': '1.189', 'DLSc': '0.01', 'DLSmaxinc': '10', 'DLSpflat': '0.15', 'DLSps': '0.1', 'DLSrho': '0.5', 'DLSs': '0.01', 'DLSsapsthresh': '-0.1', 'DLSscoringmeasure': '1', 'DLStiebreaking': '1', 'DLSwp': '0.03', 'adaptive': '1', 'adaptivenoisescheme': '1', 'adaptiveprom': '0', 'adaptpromwalkprob': '0', 'adaptwalkprob': '0', 'c': '0.01', 'clausepen': '1', 'decreasingvariable': '3', 'dp': '0.05', 'heuristic': '2', 'novnoise': '0.5', 'performalternatenovelty': '1', 'performrandomwalk': '0', 'phi': '5', 'promdp': '0.05', 'promisinglist': '1', 'promnovnoise': '0.5', 'promphi': '5', 'promtheta': '6', 'promwp': '0.01', 'ps': '0.4', 'randomwalk': '1', 'rdp': '0.05', 'rfp': '0.01', 'rwp': '0.01', 's': '0.01', 'selectclause': '1', 'singleclause': '1', 'smoothingscheme': '1', 'tabu': '1', 'tabusearch': '0', 'theta': '6', 'tiebreaking': '1', 'updateschemepromlist': '3', 'wp': '0.01', 'wpwalk': '0.5'}
config = deactivate_inactive_hyperparameters(config,cs)
with the following message:
Traceback (most recent call last):
File "test.py", line 10, in <module>
config = deactivate_inactive_hyperparameters(config,cs)
File "/home/lindauer/anaconda3/lib/python3.6/site-packages/ConfigSpace/util.py", line 381, in deactivate_inactive_hyperparameters
if not condition.evaluate(configuration):
File "/home/lindauer/anaconda3/lib/python3.6/site-packages/ConfigSpace/conditions.py", line 214, in evaluate
"'%s'" % condition.parent.name)
ValueError: Evaluate must be called with all instanstatiated parent hyperparameters in the conjunction; you are (at least) missing 'CONadaptwalkprob'
I don't understand why since CONadaptwalkprob is part of the config.
by generating a random number for the configuration space on instantiation. When a condition is added, the configuration space will save it in the condition object, and error if there already is a different random number. Thereby, the bug of having a condition in two configuration spaces is effectively prevented.
Would it be possible to create a tag of this repository of the revision which is python 2 compatible? Ideally, this would be a version compatible with the python 2 compatible revision of autosklearn.
Does this functionality exist?
I stumbled over the following and I wondered whether this is expected behaviour:
When I write a ConfigSpace-object to a file using the pcs.write() and load it again with pcs.read(), cs == cs_reloaded evaluates to False. From some debugging I get that the attributes _idx_to_hyperparameter, _parents and _hyperparameter_idx are responsible (possibly others too).
I am not very familiar with the ConfigSpace-class, but intuitively this feels like a bug rather than a feature.
Constants cannot be defined as a Boolean, categoricals can:
from ConfigSpace.hyperparameters import CategoricalHyperparameter, Constant
# works
a = CategoricalHyperparameter("a", choices(True, False))
# does not work
b = Constant("b", value=True)
Latest Master branch on Python 3.6 in a jupyter-notebook (although I don't think that matters)
import ConfigSpace as CS
config_space = CS.ConfigurationSpace()
config_space.add_hyperparameter(CS.UniformFloatHyperparameter('momentum', lower=0, upper=1, default_value=0.9));
print(config_space.sample_configuration().get_dictionary())
print(config_space.sample_configuration())
yields:
{'momentum': 0.7084377317027317}
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-5-d4fde6c7e983> in <module>()
17
18 print(config_space.sample_configuration().get_dictionary())
---> 19 print(config_space.sample_configuration())
~/repositories/github/ConfigSpace/ConfigSpace/configuration_space.cpython-36m-x86_64-linux-gnu.so in ConfigSpace.configuration_space.Configuration.__repr__()
AttributeError: module 'ConfigSpace.io' has no attribute 'StringIO'
So the config space is working, but the print fails. It also fails if I try to print the config space.
Any idea what's going on?
From PEP 8:
Package and Module Names
Modules should have short, all-lowercase names. Underscores can be used in the module name if it improves readability. Python packages should also have short, all-lowercase names, although the use of underscores is discouraged.
Does that mean the module should be 'configspace' rather than 'ConfigSpace'?
What is the conceptual difference between Constant and UnParametrizedHyperparameter ?
Hi,
I tried the following mini example for the new pcs format:
x1 real [0,10] [0]
x2 categorical {0,1,2,3,4,5,6,7,8,9,10} [0]
x3 integer [0,10] [0]
x4 ordinal [0,10] [0]
x1 | x2 in {5,6,7,8,9,10} && x3 > 5
x4 | x1 > 0 || x3 < 1 && x4 < 1
However, it complains
File "/home/lindauer/anaconda3/lib/python3.5/site-packages/ConfigSpace-0.2.2-py3.5.egg/ConfigSpace/conditions.py", line 239, in __init__
(child.name, value, parent.name))
ValueError: Hyperparameter 'x1' is conditional on the illegal value '5' of its parent hyperparameter 'x3'
The order of values for a categorical (=choices) hyperparameter matters and causes two configuration spaces to be unequal although they are equal.
Would it be helpful to throw an assertion error at the top of the function?
@hmendozap raised this issue in the auto-sklearn issue tracker:
When adding a UniformFloat or Constant hyperparameter to the configuration space of any component, the string conversion to the 'parameter configuration space file' has only 10 decimal digits precision that mismatches the native float precision (53 bits) when running auto-sklearn and setting the hyperparameter for a component.
I believe this is specific to the pcs writer of SMAC. @mlindauer is this a critical issue that we should fix?
Would it be possible to add a dict-field to each hyperparameter, that can be optionally left None? In order to keep using ConfigSpace in my projects, I sometimes find the need to store relevant meta-data about a hyperparameter (e.g., OpenML input id) and there is nowhere to store it.
When installing the package without cython being installed this happens:
pip install ConfigSpace
Collecting ConfigSpace
Downloading ConfigSpace-0.4.4.tar.gz (841kB)
100% |████████████████████████████████| 849kB 316kB/s
Complete output from command python setup.py egg_info:
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "/tmp/pip-build-6w71v_pm/ConfigSpace/setup.py", line 4, in <module>
from Cython.Build import cythonize
ModuleNotFoundError: No module named 'Cython'
Shouldn't there be a way to list cython as a dependency such that pip can install that first?
Maybe I am doing something wrong but this is happening. If I set the lower bound of num_layers to 1 then the problem is fixed. Theoretically it should work even with 2.
Error thrown:
ValueError: Hyperparameter 'num_units_1' is conditional on the illegal value '1' of its parent hyperparameter 'num_layers'
MWE
import ConfigSpace
cs = ConfigSpace.ConfigurationSpace()
max_nr_layers = 6
num_layers = ConfigSpace.UniformIntegerHyperparameter(
"num_layers",
lower=2,
upper=max_nr_layers,
default_value=2)
cs.add_hyperparameter(num_layers)
# since the configuration will actually be sampled
# we have to check for max number of layers
for i in range(1, max_nr_layers + 1):
n_units = ConfigSpace.UniformIntegerHyperparameter(
"num_units_%d" % i,
lower=16,
upper=128,
default_value=64,
log=True
)
cs.add_hyperparameter(n_units)
cs.add_condition(
ConfigSpace.OrConjunction(
ConfigSpace.GreaterThanCondition(
n_units,
num_layers,
i
),
ConfigSpace.EqualsCondition(
n_units,
num_layers,
i
)
)
)
Hi,
I'm reading in configurations from a file. Those files contain values for inactive hyperparameters. I stumbled upon this comment, which states that a method to remove inactive hyperparameters from a Configuration was, at some point, a ToDo.
I'd be glad to implement it, but I need a few pointers. To check if a parameter is active, I'd do the same as checking the configuration. Is that about right?
This would be extremely helpful for me and I'd be happy to try and add a method.
UPDATE: I found a method to correct a sampled array. I think this is already doing what I intended.
Using add_configuration_space with a configspace containing an AbstractConjunction results in a TypeError: 'module' object is not callable in:
ConfigSpace/ConfigSpace/conditions.pyx
Lines 472 to 473 in 58321ea
copy is not callable and should be used similar to:
ConfigSpace/ConfigSpace/conditions.pyx
Lines 197 to 202 in 58321ea
Several Python 2 vs 3 issues still in the code:
File "build/bdist.linux-x86_64/egg/ConfigSpace/io/pyll.py", line 60
exec pyll_string
^
SyntaxError: Missing parentheses in call to 'exec'
File "build/bdist.linux-x86_64/egg/ConfigSpace/convert.py", line 70
print "Reading searchspace..."
^
SyntaxError: Missing parentheses in call to 'print'
Extracting ConfigSpace-0.1.1-py3.5.egg to /tmp/venv/lib/python3.5/site-packages
File "/tmp/venv/lib/python3.5/site-packages/ConfigSpace-0.1.1-py3.5.egg/ConfigSpace/convert.py", line 70
print "Reading searchspace..."
^
SyntaxError: Missing parentheses in call to 'print'
File "/tmp/venv/lib/python3.5/site-packages/ConfigSpace-0.1.1-py3.5.egg/ConfigSpace/io/pyll.py", line 60
exec pyll_string
^
SyntaxError: Missing parentheses in call to 'exec'
This would allow in a searchspace like Auto-sklearn to combine the preprocessor and the classifier or two good configurations into a new, even better one.
get_one_exchange_neighborhood to work with the configuration vector instead of the configuration dictionary to determine if a hyperparameter is active or not.is_forbidden work with an array of configurations.is_forbidden in Cythonevaluate-function for condition checking in Cython.Hi, unfortunately, the pcs_new.write() writes weird conditions, e.g.:
(@1:counter-bump | @1:S:counterCond == 'yes' && @1:counter-bump | @1:0:restarts in {'F', 'D', 'L', 'x', '+'})
pcs_new.read() also cannot read these conditions and complains.
I think correct would be:
@1:counter-bump | @1:S:counterCond == 'yes' && @1:0:restarts in {'F', 'D', 'L', 'x', '+'}
I hope it is easy to fix.
I attach both the original pcs file (clasp in the old pcs format) and the pcs file generated by pcs_new.write()
Relevant stack trace (don't get more .. ):
ValueError: Trying to set illegal value '2077.1950945298586' forhyperparameter C, Type: UniformFloat, Range: [0.03125, 32768.0], Default: 1.0, on log-scale.
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.