Skip to content

GitLab

Commit 5c989b56 authored by Mike Maslyaev's avatar Mike Maslyaev
Browse files

Better fixes for existing issues, copy to reserve branch

parent 132bbc84
system_discovery
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing
with 114 additions and 98 deletions
+114 -98
epde/operators/fitness.py
View file @ 5c989b56
......@@ -18,7 +18,7 @@ from matplotlib import cm
from epde.structure.main_structures import SoEq, Equation
from epde.operators.template import CompoundOperator
import epde.globals as global_var
from TEDEouS.solver import point_sort_shift_solver
# from TEDEouS.solver import point_sort_shift_solver
class L2Fitness(CompoundOperator):
......@@ -89,99 +89,99 @@ class L2Fitness(CompoundOperator):
self._tags = {'fitness evaluation', 'gene level', 'contains suboperators', 'inplace'}
class SolverBasedFitness(CompoundOperator):
def __init__(self, param_keys : list, model_architecture = None):
super().__init__(param_keys)
if model_architecture is None:
try:
dim = global_var.tensor_cache.get(label = None).ndim
except ValueError:
dim = global_var.dimensionality
self.model_architecture = torch.nn.Sequential(
torch.nn.Linear(dim, 256),
torch.nn.Tanh(),
torch.nn.Linear(256, 64),
torch.nn.Tanh(),
torch.nn.Linear(64, 64),
torch.nn.Tanh(),
torch.nn.Linear(64, 1024),
torch.nn.Tanh(),
torch.nn.Linear(1024, 1)
)
else:
self.model_architecture = model_architecture
self.training_grid_set = False
# class SolverBasedFitness(CompoundOperator):
# def __init__(self, param_keys : list, model_architecture = None):
# super().__init__(param_keys)
# if model_architecture is None:
# try:
# dim = global_var.tensor_cache.get(label = None).ndim
# except ValueError:
# dim = global_var.dimensionality
# self.model_architecture = torch.nn.Sequential(
# torch.nn.Linear(dim, 256),
# torch.nn.Tanh(),
# torch.nn.Linear(256, 64),
# torch.nn.Tanh(),
# torch.nn.Linear(64, 64),
# torch.nn.Tanh(),
# torch.nn.Linear(64, 1024),
# torch.nn.Tanh(),
# torch.nn.Linear(1024, 1)
# )
# else:
# self.model_architecture = model_architecture
# self.training_grid_set = False
def apply(self, equation, weights_internal = None, weights_final = None):
raise NotImplementedError('Solver evaluation approach is a subject to be done.')
if not self.training_grid_set:
self.set_training_grid()
# def apply(self, equation, weights_internal = None, weights_final = None):
# raise NotImplementedError('Solver evaluation approach is a subject to be done.')
# if not self.training_grid_set:
# self.set_training_grid()
if weights_internal is None:
self.suboperators['sparsity'].apply(equation)
else:
equation.weights_internal = weights_internal
equation.weights_internal_evald = True
if weights_final is None:
self.suboperators['coeff_calc'].apply(equation)
else:
equation.weights_final = weights_final
equation.weights_final_evald = True
# if weights_internal is None:
# self.suboperators['sparsity'].apply(equation)
# else:
# equation.weights_internal = weights_internal
# equation.weights_internal_evald = True
# if weights_final is None:
# self.suboperators['coeff_calc'].apply(equation)
# else:
# equation.weights_final = weights_final
# equation.weights_final_evald = True
architecture = deepcopy(self.model_architecture)
# architecture = deepcopy(self.model_architecture)
eq_bc = equation.boundary_conditions()
eq_form = equation.solver_form()
# eq_bc = equation.boundary_conditions()
# eq_form = equation.solver_form()
equation.model = point_sort_shift_solver(self.training_grid, architecture, eq_form, eq_bc,
lambda_bound = self.params['lambda_bound'], verbose = False,
learning_rate = self.params['learning_rate'], eps = self.params['eps'],
tmin = self.params['tmin'], tmax = self.params['tmax'], use_cache=True,
print_every=None,
cache_dir='../cache/')
# equation.model = point_sort_shift_solver(self.training_grid, architecture, eq_form, eq_bc,
# lambda_bound = self.params['lambda_bound'], verbose = False,
# learning_rate = self.params['learning_rate'], eps = self.params['eps'],
# tmin = self.params['tmin'], tmax = self.params['tmax'], use_cache=True,
# print_every=None,
# cache_dir='../cache/')
main_var_key = ('u', (1.0,))
u_modeled = equation.model(self.training_grid).detach().numpy().reshape(global_var.tensor_cache.get(main_var_key).shape)
rl_error = np.linalg.norm(u_modeled - global_var.tensor_cache.get(main_var_key), ord = 2)
# main_var_key = ('u', (1.0,))
# u_modeled = equation.model(self.training_grid).detach().numpy().reshape(global_var.tensor_cache.get(main_var_key).shape)
# rl_error = np.linalg.norm(u_modeled - global_var.tensor_cache.get(main_var_key), ord = 2)
if self.params['verbose']:
self.plot_data_vs_solution(global_var.tensor_cache.get(main_var_key), u_modeled)
# if self.params['verbose']:
# self.plot_data_vs_solution(global_var.tensor_cache.get(main_var_key), u_modeled)
if rl_error == 0:
fitness_value = np.inf
print('infinite fitness!', equation.text_form)
else:
fitness_value = 1 / (rl_error)
# if rl_error == 0:
# fitness_value = np.inf
# print('infinite fitness!', equation.text_form)
# else:
# fitness_value = 1 / (rl_error)
equation.fitness_calculated = True
equation.fitness_value = fitness_value
equation.clear_after_solver()
# equation.fitness_calculated = True
# equation.fitness_value = fitness_value
# equation.clear_after_solver()
def plot_data_vs_solution(self, data, solution):
# def plot_data_vs_solution(self, data, solution):
if self.training_grid.shape[1]==2:
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.plot_trisurf(self.training_grid[:,0].reshape(-1), self.training_grid[:,1].reshape(-1),
solution.reshape(-1), cmap=cm.jet, linewidth=0.2)
ax.set_xlabel("x1")
ax.set_ylabel("x2")
plt.show()
if self.training_grid.shape[1]==1:
fig = plt.figure()
plt.scatter(self.training_grid.reshape(-1), solution.reshape(-1), color = 'r')
plt.scatter(self.training_grid.reshape(-1), data.reshape(-1), color = 'k')
plt.show()
# if self.training_grid.shape[1]==2:
# fig = plt.figure()
# ax = fig.add_subplot(111, projection='3d')
# ax.plot_trisurf(self.training_grid[:,0].reshape(-1), self.training_grid[:,1].reshape(-1),
# solution.reshape(-1), cmap=cm.jet, linewidth=0.2)
# ax.set_xlabel("x1")
# ax.set_ylabel("x2")
# plt.show()
# if self.training_grid.shape[1]==1:
# fig = plt.figure()
# plt.scatter(self.training_grid.reshape(-1), solution.reshape(-1), color = 'r')
# plt.scatter(self.training_grid.reshape(-1), data.reshape(-1), color = 'k')
# plt.show()
def set_training_grid(self):
keys, training_grid = global_var.grid_cache.get_all()
assert len(keys) == training_grid[0].ndim, 'Mismatching dimensionalities'
training_grid = np.array(training_grid).reshape((len(training_grid), -1))
self.training_grid = torch.from_numpy(training_grid).T.type(torch.FloatTensor)
self.device = torch.device('cpu')
self.training_grid.to(self.device)
self.training_grid_set = True
# Возможная проблема, когда подаётся тензор со значениями коэфф-тов перед производными
# def set_training_grid(self):
# keys, training_grid = global_var.grid_cache.get_all()
# assert len(keys) == training_grid[0].ndim, 'Mismatching dimensionalities'
# training_grid = np.array(training_grid).reshape((len(training_grid), -1))
# self.training_grid = torch.from_numpy(training_grid).T.type(torch.FloatTensor)
# self.device = torch.device('cpu')
# self.training_grid.to(self.device)
# self.training_grid_set = True
# # Возможная проблема, когда подаётся тензор со значениями коэфф-тов перед производными
def use_default_tags(self):
self._tags = {'fitness evaluation', 'gene level', 'contains suboperators', 'inplace'}
\ No newline at end of file
# def use_default_tags(self):
# self._tags = {'fitness evaluation', 'gene level', 'contains suboperators', 'inplace'}
epde/operators/moeadd_specific.py
View file @ 5c989b56
......@@ -403,17 +403,19 @@ class InitialParetoLevelSorting(CompoundOperator):
self_args, subop_args = self.parse_suboperator_args(arguments = arguments)
if len(objective.population) == 0:
for candidate in objective.unplaced_candidates:
self.suboperators['right_part_selector'].apply(objective = candidate,
arguments = subop_args['right_part_selector'])
self.suboperators['chromosome_fitness'].apply(objective = candidate,
for idx, candidate in enumerate(objective.unplaced_candidates):
while True:
temp_candidate = copy.deepcopy(candidate)
self.suboperators['right_part_selector'].apply(objective = temp_candidate,
arguments = subop_args['right_part_selector'])
if all([temp_candidate != solution for solution in objective.unplaced_candidates[:idx] +
objective.unplaced_candidates[idx+1:]]):
objective.unplaced_candidates[idx] = temp_candidate
break
self.suboperators['chromosome_fitness'].apply(objective = objective.unplaced_candidates[idx],
arguments = subop_args['chromosome_fitness'])
# print(f'candidate: {candidate.text_form}')
objective.initial_placing()
# print(f'unplaced candidates: {objective.unplaced_candidates}')
# print(f'placed candidates: {[solution.text_form for solution in objective.population]}')
return objective
def get_initial_sorter(right_part_selector : CompoundOperator,
......@@ -424,4 +426,4 @@ def get_initial_sorter(right_part_selector : CompoundOperator,
add_kwarg_to_updater(operator = sorter, labeled_base_val = {})
sorter.set_suboperators(operators = {'right_part_selector' : right_part_selector,
'chromosome_fitness' : chromosome_fitness})
return sorter
\ No newline at end of file
return sorter
epde/operators/selections.py
View file @ 5c989b56
......@@ -69,13 +69,18 @@ class MOEADDSelection(CompoundOperator):
candidate_solution_domains = list(map(lambda x: x.get_domain(self_args['weights']), [candidate for candidate in
objective.population]))
solution_mask = [(objective.population[solution_idx].get_domain(self_args['weights']) in selected_regions_idxs)
for solution_idx in candidate_solution_domains]
try:
solution_mask = [(objective.population[solution_idx].get_domain(self_args['weights']) in selected_regions_idxs)
for solution_idx in candidate_solution_domains]
except IndexError:
print(f'Indexes are: {[solution_idx for solution_idx in candidate_solution_domains]}')
print(len(objective.population), len(candidate_solution_domains))
raise IndexError('list index out of range')
available_in_proximity = sum(solution_mask)
parent_idxs = np.random.choice([idx for idx in np.arange(len(objective.population)) if solution_mask[idx]],
size = min(available_in_proximity, parents_number),
replace = False)
if available_in_proximity < parents_number: # <
if available_in_proximity < parents_number:
parent_idxs_additional = np.random.choice([idx for idx in np.arange(len(objective.population))
if not solution_mask[idx]],
size = parents_number - available_in_proximity,
......@@ -203,4 +208,4 @@ class SelectionConstraintProcesser(object):
return selected_idxs
def use_default_tags(self):
self._tags = {'constraints', 'selection', 'custom level', 'auxilary'}
\ No newline at end of file
self._tags = {'constraints', 'selection', 'custom level', 'auxilary'}
requirements.txt 0 → 100644
View file @ 5c989b56
psutil==5.8.0
scipy==1.7.1
numpy==1.19.5
matplotlib==3.3.4
epde==1.1.9
torch==1.9.0
dataclasses==0.8
guppy==0.1.11
scikit_learn==1.1.3
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment