Skip to content

Cumulitive EPDE updates #37

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Maslyaev merged 5 commits into from
Sep 15, 2025
Merged

Cumulitive EPDE updates #37

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6d07fd2
minmax normalization for odes added
Gromwud Sep 15, 2025
b9d4d7a
right_part_selection.py rst properties bugfix
Gromwud Sep 15, 2025
931c551
lr calculation optimized for bigger datasets
Gromwud Sep 15, 2025
0fd6216
Merge remote-tracking branch 'origin/main'
Gromwud Sep 15, 2025
e88c2f2
Merge branch 'ITMO-NSS-team:main' into main
Gromwud Sep 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
77 changes: 31 additions & 46 deletions epde/operators/common/fitness.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,6 +170,7 @@ def apply(self, objective: Equation, arguments: dict, force_out_of_place: bool =
# print(aic)
# print(len([_ for _ in objective.weights_final if _ !=0]))
# print(objective.aic)
assert objective.simplified, 'Trying to evaluate not simplified equation.'

# Calculate r-loss
data_shape = global_var.grid_cache.g_func.shape
Expand All @@ -189,17 +190,21 @@ def apply(self, objective: Equation, arguments: dict, force_out_of_place: bool =
if target_vals.ndim == 1:
window_size = len(target_vals) // 2
num_horizons = len(target_vals) - window_size + 1
if window_size < 15:
step_size = 1
else:
step_size = num_horizons // 30
eq_window_weights = []
# Compute coefficients and collect statistics over horizons
if len(features_vals) == 0:
for start_idx in range(num_horizons):
for start_idx in range(0, num_horizons, step_size):
end_idx = start_idx + window_size
target_window = target_vals[start_idx:end_idx]
eq_window_weights.append(np.abs(np.std(target_window) / np.mean(target_window)))
lr = np.mean(eq_window_weights)
else:
features = self.feature_reshape(features_vals)
for start_idx in range(num_horizons):
for start_idx in range(0, num_horizons, step_size):
end_idx = start_idx + window_size
target_window = target_vals[start_idx:end_idx]
feature_window = features[start_idx:end_idx, :]
Expand All @@ -216,9 +221,13 @@ def apply(self, objective: Equation, arguments: dict, force_out_of_place: bool =
eq_window_weights = []
window_size = target_vals.shape[dim] // 2
num_horizons = target_vals.shape[dim] - window_size + 1
if window_size < 15:
step_size = 1
else:
step_size = num_horizons // 30
# Compute coefficients and collect statistics over horizons
if len(features_vals) == 0:
for start_idx in range(num_horizons):
for start_idx in range(0, num_horizons, step_size):
end_idx = start_idx + window_size
if dim == 0:
target_window = target_vals[start_idx:end_idx, :].reshape(-1)
Expand All @@ -228,7 +237,7 @@ def apply(self, objective: Equation, arguments: dict, force_out_of_place: bool =
lr += np.mean(eq_window_weights)
else:
features = self.feature_reshape(features_vals)
for start_idx in range(num_horizons):
for start_idx in range(0, num_horizons, step_size):
end_idx = start_idx + window_size
estimator = LinearRegression(fit_intercept=False)
if dim == 0:
Expand Down Expand Up @@ -454,23 +463,18 @@ def apply(self, objective: SoEq, arguments: dict, force_out_of_place: bool = Fal
print(f'solution shape {solution.shape}')
print(f'solution[..., eq_idx] {solution[..., eq_idx].shape}, eq_idx {eq_idx}')
referential_data = global_var.tensor_cache.get((eq.main_var_to_explain, (1.0,)))
# initial_data = global_var.tensor_cache.get(('u', (1.0,))).reshape(solution[..., eq_idx].shape)
#
# sol_pinn = solution[..., eq_idx]
# sol_ann = referential_data.reshape(solution[..., eq_idx].shape)
# sol_pinn_normalized = (sol_pinn - min(initial_data)) / (max(initial_data) - min(initial_data))
# sol_ann_normalized = (sol_ann - min(initial_data)) / (max(initial_data) - min(initial_data))
#
# discr = sol_pinn_normalized - sol_ann_normalized
discr = (solution[..., eq_idx] - referential_data.reshape(solution[..., eq_idx].shape)) # Default
maximum = np.max([referential_data.max(axis=0), solution[..., eq_idx].max(axis=0)])
minimum = np.min([referential_data.min(axis=0), solution[..., eq_idx].min(axis=0)])
discr = ((solution[..., eq_idx] - minimum) - (referential_data - minimum)) / (maximum - minimum) # Normalized
# discr = (solution[..., eq_idx] - referential_data.reshape(solution[..., eq_idx].shape)) # Default
discr = np.multiply(discr, self.g_fun_vals.reshape(discr.shape))
rl_error = np.linalg.norm(discr, ord=2)

print(f'fitness error is {rl_error}, while loss addition is {float(loss_add)}')
lp = rl_error + self.params['pinn_loss_mult'] * float(
loss_add) # TODO: make pinn_loss_mult case dependent
if np.sum(eq.weights_final) == 0:
lp /= self.params['penalty_coeff']
loss_add) * 0 # TODO: make pinn_loss_mult case dependent
# if np.sum(eq.weights_final) == 0:
# lp /= self.params['penalty_coeff']

ssr = np.sum(discr ** 2)
n = len(discr)
Expand All @@ -488,6 +492,7 @@ def apply(self, objective: SoEq, arguments: dict, force_out_of_place: bool = Fal
eq.aic_calculated = True

# Calculate r-loss
data_shape = global_var.grid_cache.g_func.shape
target = eq.structure[eq.target_idx]
target_vals = target.evaluate(False).reshape(*data_shape)
features_vals = []
Expand Down Expand Up @@ -548,20 +553,12 @@ def apply(self, objective: SoEq, arguments: dict, force_out_of_place: bool = Fal
estimator = LinearRegression(fit_intercept=False)
if dim == 0:
target_window = target_vals[start_idx:end_idx, :].reshape(-1)
feature_window = features.reshape(*data_shape, -1)[start_idx:end_idx, :].reshape(-1,
features.shape[
-1])
estimator.fit(feature_window, target_window,
sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[start_idx:end_idx,
:].reshape(-1))
feature_window = features.reshape(*data_shape, -1)[start_idx:end_idx, :].reshape(-1, features.shape[-1])
estimator.fit(feature_window, target_window, sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[start_idx:end_idx, :].reshape(-1))
else:
target_window = target_vals[:, start_idx:end_idx].reshape(-1)
feature_window = features.reshape(*data_shape, -1)[:, start_idx:end_idx].reshape(-1,
features.shape[
-1])
estimator.fit(feature_window, target_window,
sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:,
start_idx:end_idx].reshape(-1))
feature_window = features.reshape(*data_shape, -1)[:, start_idx:end_idx].reshape(-1, features.shape[-1])
estimator.fit(feature_window, target_window, sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:, start_idx:end_idx].reshape(-1))
valuable_weights = estimator.coef_[:-1]
eq_window_weights.append(valuable_weights)
eq_cv = np.array([np.abs(np.std(_) / np.mean(_)) for _ in zip(*eq_window_weights)])
Expand Down Expand Up @@ -592,28 +589,16 @@ def apply(self, objective: SoEq, arguments: dict, force_out_of_place: bool = Fal
estimator = LinearRegression(fit_intercept=False)
if dim == 0:
target_window = target_vals[start_idx:end_idx, :, :].reshape(-1)
feature_window = features.reshape(*data_shape, -1)[start_idx:end_idx, :, :].reshape(-1,
features.shape[
-1])
estimator.fit(feature_window, target_window,
sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[start_idx:end_idx, :,
:].reshape(-1))
feature_window = features.reshape(*data_shape, -1)[start_idx:end_idx, :, :].reshape(-1, features.shape[-1])
estimator.fit(feature_window, target_window, sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[start_idx:end_idx, :, :].reshape(-1))
elif dim == 1:
target_window = target_vals[:, start_idx:end_idx, :].reshape(-1)
feature_window = features.reshape(*data_shape, -1)[:, start_idx:end_idx, :].reshape(-1,
features.shape[
-1])
estimator.fit(feature_window, target_window,
sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:, start_idx:end_idx,
:].reshape(-1))
feature_window = features.reshape(*data_shape, -1)[:, start_idx:end_idx, :].reshape(-1, features.shape[-1])
estimator.fit(feature_window, target_window, sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:, start_idx:end_idx, :].reshape(-1))
elif dim == 2:
target_window = target_vals[:, :, start_idx:end_idx].reshape(-1)
feature_window = features.reshape(*data_shape, -1)[:, :, start_idx:end_idx].reshape(-1,
features.shape[
-1])
estimator.fit(feature_window, target_window,
sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:, :,
start_idx:end_idx].reshape(-1))
feature_window = features.reshape(*data_shape, -1)[:, :, start_idx:end_idx].reshape(-1, features.shape[-1])
estimator.fit(feature_window, target_window, sample_weight=self.g_fun_vals.reshape(*data_shape, -1)[:, :, start_idx:end_idx].reshape(-1))
valuable_weights = estimator.coef_[:-1]
eq_window_weights.append(valuable_weights)
eq_cv = np.array([np.abs(np.std(_) / np.mean(_)) for _ in zip(*eq_window_weights)])
Expand Down
8 changes: 3 additions & 5 deletions epde/operators/common/right_part_selection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,13 +52,11 @@ def apply(self, objective : Equation, arguments : dict):
min_fitness = np.inf
weights_internal = np.zeros_like(objective.structure)
min_idx = 0
if not objective.contains_deriv(objective.main_var_to_explain):
objective.restore_property(deriv = True)
if not objective.contains_variable(objective.main_var_to_explain):
objective.restore_property(mandatory_family = objective.main_var_to_explain)
if not any(term.contains_variable(objective.main_var_to_explain) and term.contains_deriv(objective.main_var_to_explain) for term in objective.structure):
objective.restore_property(mandatory_family=objective.main_var_to_explain, deriv=True)

for target_idx, target_term in enumerate(objective.structure):
if not objective.structure[target_idx].contains_deriv(objective.main_var_to_explain):
if not (objective.structure[target_idx].contains_variable(objective.main_var_to_explain) and objective.structure[target_idx].contains_deriv(objective.main_var_to_explain)):
continue
objective.target_idx = target_idx
fitness = self.suboperators['fitness_calculation'].apply(objective,
Expand Down
21 changes: 14 additions & 7 deletions epde/structure/main_structures.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,8 +215,8 @@ def evaluate(self, structural, grids=None):
self.prev_normalized = normalize
value = super().evaluate(structural)
if normalize:
value = np.ones_like(value)
if np.ndim(value) != 1:
value = np.ones_like(value)
for factor in self.structure:
temp = factor.evaluate()
# value *= normalize_ts(temp)
Expand All @@ -226,10 +226,14 @@ def evaluate(self, structural, grids=None):
# # value = normalize_ts(value)
# value = minmax_normalize(value)
else:
if np.std(value) != 0:
value = (value - np.mean(value)) / np.std(value)
else:
value = (value - np.mean(value))
# if np.std(value) != 0:
# value = (value - np.mean(value)) / np.std(value)
# else:
# value = (value - np.mean(value))
for factor in self.structure:
temp = factor.evaluate()
# value *= normalize_ts(temp)
value *= (temp - np.mean(temp) - np.min(temp)) / (np.max(temp) - np.min(temp))
if np.all([len(factor.params) == 1 for factor in self.structure]) and grids is None:
# Место возможных проблем: сохранение/загрузка нормализованных данных
self.saved[normalize] = global_var.tensor_cache.add(self.cache_label, value, normalized=normalize)
Expand Down Expand Up @@ -508,10 +512,13 @@ def restore_property(self, deriv: bool = False, mandatory_family: bool = False):
mf_marker = mandatory_family if mandatory_family else None
temp = Term(self.pool, mandatory_family=mf_marker,
max_factors_in_term=self.metaparameters['max_factors_in_term']['value'])
if deriv and temp.contains_deriv():
if deriv and mandatory_family and temp.contains_deriv() and temp.contains_variable(self.main_var_to_explain):
self.structure[replacement_idx] = temp
break
elif deriv and temp.contains_deriv() and not mandatory_family:
self.structure[replacement_idx] = temp
break
elif mandatory_family and temp.contains_variable(self.main_var_to_explain):
elif mandatory_family and temp.contains_variable(self.main_var_to_explain) and not deriv:
self.structure[replacement_idx] = temp
break
else:
Expand Down