jinns 1.2.0__py3-none-any.whl → 1.4.0__py3-none-any.whl

jinns/loss/_LossPDE.py

@@ -1,16 +1,16 @@
-# pylint: disable=unsubscriptable-object, no-member
 """
 Main module to implement a PDE loss in jinns
 """
+
 from __future__ import (
     annotations,
 )  # https://docs.python.org/3/library/typing.html#constant
 
 import abc
-from dataclasses import InitVar, fields
-from typing import TYPE_CHECKING, Dict, Callable
+from dataclasses import InitVar
+from typing import TYPE_CHECKING, Callable, TypedDict
+from types import EllipsisType
 import warnings
-import jax
 import jax.numpy as jnp
 import equinox as eqx
 from jaxtyping import Float, Array, Key, Int
@@ -20,9 +20,7 @@ from jinns.loss._loss_utils import (
     normalization_loss_apply,
     observations_loss_apply,
     initial_condition_apply,
-    constraints_system_loss_apply,
 )
-from jinns.data._DataGenerators import append_obs_batch
 from jinns.parameters._params import (
     _get_vmap_in_axes_params,
     _update_eq_params_dict,
@@ -32,19 +30,30 @@ from jinns.parameters._derivative_keys import (
     DerivativeKeysPDEStatio,
     DerivativeKeysPDENonStatio,
 )
+from jinns.loss._abstract_loss import AbstractLoss
 from jinns.loss._loss_weights import (
     LossWeightsPDEStatio,
     LossWeightsPDENonStatio,
-    LossWeightsPDEDict,
 )
-from jinns.loss._DynamicLossAbstract import PDEStatio, PDENonStatio
-from jinns.utils._pinn import PINN
-from jinns.utils._spinn import SPINN
 from jinns.data._Batchs import PDEStatioBatch, PDENonStatioBatch
 
 
 if TYPE_CHECKING:
-    from jinns.utils._types import *
+    # imports for type hints only
+    from jinns.parameters._params import Params
+    from jinns.nn._abstract_pinn import AbstractPINN
+    from jinns.loss import PDENonStatio, PDEStatio
+    from jinns.utils._types import BoundaryConditionFun
+
+    class LossDictPDEStatio(TypedDict):
+        dyn_loss: Float[Array, " "]
+        norm_loss: Float[Array, " "]
+        boundary_loss: Float[Array, " "]
+        observations: Float[Array, " "]
+
+    class LossDictPDENonStatio(LossDictPDEStatio):
+        initial_condition: Float[Array, " "]
+
 
 _IMPLEMENTED_BOUNDARY_CONDITIONS = [
     "dirichlet",
@@ -53,8 +62,8 @@ _IMPLEMENTED_BOUNDARY_CONDITIONS = [
 ]
 
 
-class _LossPDEAbstract(eqx.Module):
-    """
+class _LossPDEAbstract(AbstractLoss):
+    r"""
     Parameters
     ----------
 
@@ -69,11 +78,11 @@ class _LossPDEAbstract(eqx.Module):
         Fields can be "nn_params", "eq_params" or "both". Those that should not
         be updated will have a `jax.lax.stop_gradient` called on them. Default
         is `"nn_params"` for each composant of the loss.
-    omega_boundary_fun : Callable | Dict[str, Callable], default=None
+    omega_boundary_fun : BoundaryConditionFun | dict[str, BoundaryConditionFun], default=None
          The function to be matched in the border condition (can be None) or a
          dictionary of such functions as values and keys as described
          in `omega_boundary_condition`.
-    omega_boundary_condition : str | Dict[str, str], default=None
+    omega_boundary_condition : str | dict[str, str], default=None
         Either None (no condition, by default), or a string defining
         the boundary condition (Dirichlet or Von Neumann),
         or a dictionary with such strings as values. In this case,
@@ -84,24 +93,29 @@ class _LossPDEAbstract(eqx.Module):
         a particular boundary condition on this facet.
         The facet called “xmin”, resp. “xmax” etc., in 2D,
         refers to the set of 2D points with fixed “xmin”, resp. “xmax”, etc.
-    omega_boundary_dim : slice | Dict[str, slice], default=None
+    omega_boundary_dim : slice | dict[str, slice], default=None
         Either None, or a slice object or a dictionary of slice objects as
         values and keys as described in `omega_boundary_condition`.
         `omega_boundary_dim` indicates which dimension(s) of the PINN
         will be forced to match the boundary condition.
         Note that it must be a slice and not an integer
         (but a preprocessing of the user provided argument takes care of it)
-    norm_samples : Float[Array, "nb_norm_samples dimension"], default=None
-        Fixed sample point in the space over which to compute the
+    norm_samples : Float[Array, " nb_norm_samples dimension"], default=None
+        Monte-Carlo sample points for computing the
         normalization constant. Default is None.
-    norm_int_length : float, default=None
-        A float. Must be provided if `norm_samples` is provided. The domain area
-        (or interval length in 1D) upon which we perform the numerical
-        integration. Default None
-    obs_slice : slice, default=None
+    norm_weights : Float[Array, " nb_norm_samples"] | float | int, default=None
+        The importance sampling weights for Monte-Carlo integration of the
+        normalization constant. Must be provided if `norm_samples` is provided.
+        `norm_weights` should be broadcastble to
+        `norm_samples`.
+        Alternatively, the user can pass a float or an integer that will be
+        made broadcastable to `norm_samples`.
+        These corresponds to the weights $w_k = \frac{1}{q(x_k)}$ where
+        $q(\cdot)$ is the proposal p.d.f. and $x_k$ are the Monte-Carlo samples.
+    obs_slice : EllipsisType | slice, default=None
         slice object specifying the begininning/ending of the PINN output
         that is observed (this is then useful for multidim PINN). Default is None.
-    params : InitVar[Params], default=None
+    params : InitVar[Params[Array]], default=None
         The main Params object of the problem needed to instanciate the
         DerivativeKeysODE if the latter is not specified.
     """
@@ -115,24 +129,28 @@ class _LossPDEAbstract(eqx.Module):
     loss_weights: LossWeightsPDEStatio | LossWeightsPDENonStatio | None = eqx.field(
         kw_only=True, default=None
     )
-    omega_boundary_fun: Callable | Dict[str, Callable] | None = eqx.field(
+    omega_boundary_fun: (
+        BoundaryConditionFun | dict[str, BoundaryConditionFun] | None
+    ) = eqx.field(kw_only=True, default=None, static=True)
+    omega_boundary_condition: str | dict[str, str] | None = eqx.field(
         kw_only=True, default=None, static=True
     )
-    omega_boundary_condition: str | Dict[str, str] | None = eqx.field(
+    omega_boundary_dim: slice | dict[str, slice] | None = eqx.field(
         kw_only=True, default=None, static=True
     )
-    omega_boundary_dim: slice | Dict[str, slice] | None = eqx.field(
-        kw_only=True, default=None, static=True
+    norm_samples: Float[Array, " nb_norm_samples dimension"] | None = eqx.field(
+        kw_only=True, default=None
     )
-    norm_samples: Float[Array, "nb_norm_samples dimension"] | None = eqx.field(
+    norm_weights: Float[Array, " nb_norm_samples"] | float | int | None = eqx.field(
         kw_only=True, default=None
     )
-    norm_int_length: float | None = eqx.field(kw_only=True, default=None)
-    obs_slice: slice | None = eqx.field(kw_only=True, default=None, static=True)
+    obs_slice: EllipsisType | slice | None = eqx.field(
+        kw_only=True, default=None, static=True
+    )
 
-    params: InitVar[Params] = eqx.field(kw_only=True, default=None)
+    params: InitVar[Params[Array]] = eqx.field(kw_only=True, default=None)
 
-    def __post_init__(self, params=None):
+    def __post_init__(self, params: Params[Array] | None = None):
         """
         Note that neither __init__ or __post_init__ are called when udating a
         Module with eqx.tree_at
@@ -222,6 +240,11 @@ class _LossPDEAbstract(eqx.Module):
                         )
 
         if isinstance(self.omega_boundary_fun, dict):
+            if not isinstance(self.omega_boundary_dim, dict):
+                raise ValueError(
+                    "If omega_boundary_fun is a dict then"
+                    " omega_boundary_dim should also be a dict"
+                )
             if self.omega_boundary_dim is None:
                 self.omega_boundary_dim = {
                     k: jnp.s_[::] for k in self.omega_boundary_fun.keys()
@@ -251,15 +274,34 @@ class _LossPDEAbstract(eqx.Module):
             if not isinstance(self.omega_boundary_dim, slice):
                 raise ValueError("self.omega_boundary_dim must be a jnp.s_ object")
 
-        if self.norm_samples is not None and self.norm_int_length is None:
-            raise ValueError("self.norm_samples and norm_int_length must be provided")
+        if self.norm_samples is not None:
+            if self.norm_weights is None:
+                raise ValueError(
+                    "`norm_weights` must be provided when `norm_samples` is used!"
+                )
+            if isinstance(self.norm_weights, (int, float)):
+                self.norm_weights = self.norm_weights * jnp.ones(
+                    (self.norm_samples.shape[0],)
+                )
+            if isinstance(self.norm_weights, Array):
+                if not (self.norm_weights.shape[0] == self.norm_samples.shape[0]):
+                    raise ValueError(
+                        "self.norm_weights and "
+                        "self.norm_samples must have the same leading dimension"
+                    )
+            else:
+                raise ValueError("Wrong type for self.norm_weights")
+
+    @abc.abstractmethod
+    def __call__(self, *_, **__):
+        pass
 
     @abc.abstractmethod
     def evaluate(
         self: eqx.Module,
-        params: Params,
+        params: Params[Array],
         batch: PDEStatioBatch | PDENonStatioBatch,
-    ) -> tuple[Float, dict]:
+    ) -> tuple[Float[Array, " "], LossDictPDEStatio | LossDictPDENonStatio]:
         raise NotImplementedError
 
 
@@ -276,9 +318,9 @@ class LossPDEStatio(_LossPDEAbstract):
 
     Parameters
     ----------
-    u : eqx.Module
+    u : AbstractPINN
         the PINN
-    dynamic_loss : DynamicLoss
+    dynamic_loss : PDEStatio
         the stationary PDE dynamic part of the loss, basically the differential
         operator $\mathcal{N}[u](x)$. Should implement a method
         `dynamic_loss.evaluate(x, u, params)`.
@@ -301,11 +343,11 @@ class LossPDEStatio(_LossPDEAbstract):
         Fields can be "nn_params", "eq_params" or "both". Those that should not
         be updated will have a `jax.lax.stop_gradient` called on them. Default
         is `"nn_params"` for each composant of the loss.
-    omega_boundary_fun : Callable | Dict[str, Callable], default=None
+    omega_boundary_fun : BoundaryConditionFun | dict[str, BoundaryConditionFun], default=None
          The function to be matched in the border condition (can be None) or a
          dictionary of such functions as values and keys as described
          in `omega_boundary_condition`.
-    omega_boundary_condition : str | Dict[str, str], default=None
+    omega_boundary_condition : str | dict[str, str], default=None
         Either None (no condition, by default), or a string defining
         the boundary condition (Dirichlet or Von Neumann),
         or a dictionary with such strings as values. In this case,
@@ -316,24 +358,28 @@ class LossPDEStatio(_LossPDEAbstract):
         a particular boundary condition on this facet.
         The facet called “xmin”, resp. “xmax” etc., in 2D,
         refers to the set of 2D points with fixed “xmin”, resp. “xmax”, etc.
-    omega_boundary_dim : slice | Dict[str, slice], default=None
+    omega_boundary_dim : slice | dict[str, slice], default=None
         Either None, or a slice object or a dictionary of slice objects as
         values and keys as described in `omega_boundary_condition`.
         `omega_boundary_dim` indicates which dimension(s) of the PINN
         will be forced to match the boundary condition.
         Note that it must be a slice and not an integer
         (but a preprocessing of the user provided argument takes care of it)
-    norm_samples : Float[Array, "nb_norm_samples dimension"], default=None
-        Fixed sample point in the space over which to compute the
+    norm_samples : Float[Array, " nb_norm_samples dimension"], default=None
+        Monte-Carlo sample points for computing the
         normalization constant. Default is None.
-    norm_int_length : float, default=None
-        A float. Must be provided if `norm_samples` is provided. The domain area
-        (or interval length in 1D) upon which we perform the numerical
-        integration. Default None
+    norm_weights : Float[Array, " nb_norm_samples"] | float | int, default=None
+        The importance sampling weights for Monte-Carlo integration of the
+        normalization constant. Must be provided if `norm_samples` is provided.
+        `norm_weights` should have the same leading dimension as
+        `norm_samples`.
+        Alternatively, the user can pass a float or an integer.
+        These corresponds to the weights $w_k = \frac{1}{q(x_k)}$ where
+        $q(\cdot)$ is the proposal p.d.f. and $x_k$ are the Monte-Carlo samples.
     obs_slice : slice, default=None
         slice object specifying the begininning/ending of the PINN output
         that is observed (this is then useful for multidim PINN). Default is None.
-    params : InitVar[Params], default=None
+    params : InitVar[Params[Array]], default=None
         The main Params object of the problem needed to instanciate the
         DerivativeKeysODE if the latter is not specified.
 
@@ -348,13 +394,13 @@ class LossPDEStatio(_LossPDEAbstract):
     # NOTE static=True only for leaf attributes that are not valid JAX types
     # (ie. jax.Array cannot be static) and that we do not expect to change
 
-    u: eqx.Module
-    dynamic_loss: DynamicLoss | None
+    u: AbstractPINN
+    dynamic_loss: PDEStatio | None
     key: Key | None = eqx.field(kw_only=True, default=None)
 
     vmap_in_axes: tuple[Int] = eqx.field(init=False, static=True)
 
-    def __post_init__(self, params=None):
+    def __post_init__(self, params: Params[Array] | None = None):
         """
         Note that neither __init__ or __post_init__ are called when udating a
         Module with eqx.tree_at!
@@ -368,25 +414,27 @@ class LossPDEStatio(_LossPDEAbstract):
 
     def _get_dynamic_loss_batch(
         self, batch: PDEStatioBatch
-    ) -> Float[Array, "batch_size dimension"]:
+    ) -> Float[Array, " batch_size dimension"]:
         return batch.domain_batch
 
     def _get_normalization_loss_batch(
         self, _
-    ) -> Float[Array, "nb_norm_samples dimension"]:
-        return (self.norm_samples,)
+    ) -> tuple[Float[Array, " nb_norm_samples dimension"]]:
+        return (self.norm_samples,)  # type: ignore -> cannot narrow a class attr
+
+    # we could have used typing.cast though
 
     def _get_observations_loss_batch(
         self, batch: PDEStatioBatch
-    ) -> Float[Array, "batch_size obs_dim"]:
-        return (batch.obs_batch_dict["pinn_in"],)
+    ) -> Float[Array, " batch_size obs_dim"]:
+        return batch.obs_batch_dict["pinn_in"]
 
     def __call__(self, *args, **kwargs):
         return self.evaluate(*args, **kwargs)
 
     def evaluate(
-        self, params: Params, batch: PDEStatioBatch
-    ) -> tuple[Float[Array, "1"], dict[str, float]]:
+        self, params: Params[Array], batch: PDEStatioBatch
+    ) -> tuple[Float[Array, " "], LossDictPDEStatio]:
         """
         Evaluate the loss function at a batch of points for given parameters.
 
@@ -417,9 +465,9 @@ class LossPDEStatio(_LossPDEAbstract):
                 self.dynamic_loss.evaluate,
                 self.u,
                 self._get_dynamic_loss_batch(batch),
-                _set_derivatives(params, self.derivative_keys.dyn_loss),
+                _set_derivatives(params, self.derivative_keys.dyn_loss),  # type: ignore
                 self.vmap_in_axes + vmap_in_axes_params,
-                self.loss_weights.dyn_loss,
+                self.loss_weights.dyn_loss,  # type: ignore
             )
         else:
             mse_dyn_loss = jnp.array(0.0)
@@ -429,24 +477,28 @@ class LossPDEStatio(_LossPDEAbstract):
             mse_norm_loss = normalization_loss_apply(
                 self.u,
                 self._get_normalization_loss_batch(batch),
-                _set_derivatives(params, self.derivative_keys.norm_loss),
+                _set_derivatives(params, self.derivative_keys.norm_loss),  # type: ignore
                 vmap_in_axes_params,
-                self.norm_int_length,
-                self.loss_weights.norm_loss,
+                self.norm_weights,  # type: ignore -> can't get the __post_init__ narrowing here
+                self.loss_weights.norm_loss,  # type: ignore
             )
         else:
             mse_norm_loss = jnp.array(0.0)
 
         # boundary part
-        if self.omega_boundary_condition is not None:
+        if (
+            self.omega_boundary_condition is not None
+            and self.omega_boundary_dim is not None
+            and self.omega_boundary_fun is not None
+        ):  # pyright cannot narrow down the three None otherwise as it is class attribute
             mse_boundary_loss = boundary_condition_apply(
                 self.u,
                 batch,
-                _set_derivatives(params, self.derivative_keys.boundary_loss),
+                _set_derivatives(params, self.derivative_keys.boundary_loss),  # type: ignore
                 self.omega_boundary_fun,
                 self.omega_boundary_condition,
                 self.omega_boundary_dim,
-                self.loss_weights.boundary_loss,
+                self.loss_weights.boundary_loss,  # type: ignore
             )
         else:
             mse_boundary_loss = jnp.array(0.0)
@@ -459,10 +511,10 @@ class LossPDEStatio(_LossPDEAbstract):
             mse_observation_loss = observations_loss_apply(
                 self.u,
                 self._get_observations_loss_batch(batch),
-                _set_derivatives(params, self.derivative_keys.observations),
+                _set_derivatives(params, self.derivative_keys.observations),  # type: ignore
                 self.vmap_in_axes + vmap_in_axes_params,
                 batch.obs_batch_dict["val"],
-                self.loss_weights.observations,
+                self.loss_weights.observations,  # type: ignore
                 self.obs_slice,
             )
         else:
@@ -478,8 +530,6 @@ class LossPDEStatio(_LossPDEAbstract):
                 "norm_loss": mse_norm_loss,
                 "boundary_loss": mse_boundary_loss,
                 "observations": mse_observation_loss,
-                "initial_condition": jnp.array(0.0),  # for compatibility in the
-                # tree_map of SystemLoss
             }
         )
 
@@ -500,9 +550,9 @@ class LossPDENonStatio(LossPDEStatio):
 
     Parameters
     ----------
-    u : eqx.Module
+    u : AbstractPINN
         the PINN
-    dynamic_loss : DynamicLoss
+    dynamic_loss : PDENonStatio
         the non stationary PDE dynamic part of the loss, basically the differential
         operator $\mathcal{N}[u](t, x)$. Should implement a method
         `dynamic_loss.evaluate(t, x, u, params)`.
@@ -526,11 +576,11 @@ class LossPDENonStatio(LossPDEStatio):
         Fields can be "nn_params", "eq_params" or "both". Those that should not
         be updated will have a `jax.lax.stop_gradient` called on them. Default
         is `"nn_params"` for each composant of the loss.
-    omega_boundary_fun : Callable | Dict[str, Callable], default=None
+    omega_boundary_fun : BoundaryConditionFun | dict[str, BoundaryConditionFun], default=None
          The function to be matched in the border condition (can be None) or a
          dictionary of such functions as values and keys as described
          in `omega_boundary_condition`.
-    omega_boundary_condition : str | Dict[str, str], default=None
+    omega_boundary_condition : str | dict[str, str], default=None
         Either None (no condition, by default), or a string defining
         the boundary condition (Dirichlet or Von Neumann),
         or a dictionary with such strings as values. In this case,
@@ -541,37 +591,46 @@ class LossPDENonStatio(LossPDEStatio):
         a particular boundary condition on this facet.
         The facet called “xmin”, resp. “xmax” etc., in 2D,
         refers to the set of 2D points with fixed “xmin”, resp. “xmax”, etc.
-    omega_boundary_dim : slice | Dict[str, slice], default=None
+    omega_boundary_dim : slice | dict[str, slice], default=None
         Either None, or a slice object or a dictionary of slice objects as
         values and keys as described in `omega_boundary_condition`.
         `omega_boundary_dim` indicates which dimension(s) of the PINN
         will be forced to match the boundary condition.
         Note that it must be a slice and not an integer
         (but a preprocessing of the user provided argument takes care of it)
-    norm_samples : Float[Array, "nb_norm_samples dimension"], default=None
-        Fixed sample point in the space over which to compute the
+    norm_samples : Float[Array, " nb_norm_samples dimension"], default=None
+        Monte-Carlo sample points for computing the
         normalization constant. Default is None.
-    norm_int_length : float, default=None
-        A float. Must be provided if `norm_samples` is provided. The domain area
-        (or interval length in 1D) upon which we perform the numerical
-        integration. Default None
+    norm_weights : Float[Array, " nb_norm_samples"] | float | int, default=None
+        The importance sampling weights for Monte-Carlo integration of the
+        normalization constant. Must be provided if `norm_samples` is provided.
+        `norm_weights` should have the same leading dimension as
+        `norm_samples`.
+        Alternatively, the user can pass a float or an integer.
+        These corresponds to the weights $w_k = \frac{1}{q(x_k)}$ where
+        $q(\cdot)$ is the proposal p.d.f. and $x_k$ are the Monte-Carlo samples.
     obs_slice : slice, default=None
         slice object specifying the begininning/ending of the PINN output
         that is observed (this is then useful for multidim PINN). Default is None.
+    t0 : float | Float[Array, " 1"], default=None
+        The time at which to apply the initial condition. If None, the time
+        is set to `0` by default.
     initial_condition_fun : Callable, default=None
-        A function representing the temporal initial condition. If None
-        (default) then no initial condition is applied
-    params : InitVar[Params], default=None
-        The main Params object of the problem needed to instanciate the
-        DerivativeKeysODE if the latter is not specified.
+        A function representing the initial condition at `t0`. If None
+        (default) then no initial condition is applied.
+    params : InitVar[Params[Array]], default=None
+        The main `Params` object of the problem needed to instanciate the
+        `DerivativeKeysODE` if the latter is not specified.
 
     """
 
+    dynamic_loss: PDENonStatio | None
     # NOTE static=True only for leaf attributes that are not valid JAX types
     # (ie. jax.Array cannot be static) and that we do not expect to change
     initial_condition_fun: Callable | None = eqx.field(
         kw_only=True, default=None, static=True
     )
+    t0: float | Float[Array, " 1"] | None = eqx.field(kw_only=True, default=None)
 
     _max_norm_samples_omega: Int = eqx.field(init=False, static=True)
     _max_norm_time_slices: Int = eqx.field(init=False, static=True)
@@ -593,6 +652,21 @@ class LossPDENonStatio(LossPDEStatio):
                 "Initial condition wasn't provided. Be sure to cover for that"
                 "case (e.g by. hardcoding it into the PINN output)."
             )
+        # some checks for t0
+        if isinstance(self.t0, Array):
+            if not self.t0.shape:  # e.g. user input: jnp.array(0.)
+                self.t0 = jnp.array([self.t0])
+            elif self.t0.shape != (1,):
+                raise ValueError(
+                    f"Wrong self.t0 input. It should be"
+                    f"a float or an array of shape (1,). Got shape: {self.t0.shape}"
+                )
+        elif isinstance(self.t0, float):  # e.g. user input: 0
+            self.t0 = jnp.array([self.t0])
+        elif self.t0 is None:
+            self.t0 = jnp.array([0])
+        else:
+            raise ValueError("Wrong value for t0")
 
         # witht the variables below we avoid memory overflow since a cartesian
         # product is taken
@@ -601,28 +675,30 @@ class LossPDENonStatio(LossPDEStatio):
 
     def _get_dynamic_loss_batch(
         self, batch: PDENonStatioBatch
-    ) -> Float[Array, "batch_size 1+dimension"]:
+    ) -> Float[Array, " batch_size 1+dimension"]:
         return batch.domain_batch
 
     def _get_normalization_loss_batch(
         self, batch: PDENonStatioBatch
-    ) -> Float[Array, "nb_norm_time_slices nb_norm_samples dimension"]:
+    ) -> tuple[
+        Float[Array, " nb_norm_time_slices 1"], Float[Array, " nb_norm_samples dim"]
+    ]:
         return (
             batch.domain_batch[: self._max_norm_time_slices, 0:1],
-            self.norm_samples[: self._max_norm_samples_omega],
+            self.norm_samples[: self._max_norm_samples_omega],  # type: ignore -> cannot narrow a class attr
         )
 
     def _get_observations_loss_batch(
         self, batch: PDENonStatioBatch
-    ) -> tuple[Float[Array, "batch_size 1"], Float[Array, "batch_size dimension"]]:
-        return (batch.obs_batch_dict["pinn_in"],)
+    ) -> Float[Array, " batch_size 1+dim"]:
+        return batch.obs_batch_dict["pinn_in"]
 
     def __call__(self, *args, **kwargs):
         return self.evaluate(*args, **kwargs)
 
     def evaluate(
-        self, params: Params, batch: PDENonStatioBatch
-    ) -> tuple[Float[Array, "1"], dict[str, float]]:
+        self, params: Params[Array], batch: PDENonStatioBatch
+    ) -> tuple[Float[Array, " "], LossDictPDENonStatio]:
         """
         Evaluate the loss function at a batch of points for given parameters.
 
@@ -639,6 +715,7 @@ class LossPDENonStatio(LossPDEStatio):
             inputs/outputs/parameters
         """
         omega_batch = batch.initial_batch
+        assert omega_batch is not None
 
         # Retrieve the optional eq_params_batch
         # and update eq_params with the latter
@@ -651,17 +728,19 @@ class LossPDENonStatio(LossPDEStatio):
 
         # For mse_dyn_loss, mse_norm_loss, mse_boundary_loss,
         # mse_observation_loss we use the evaluate from parent class
-        partial_mse, partial_mse_terms = super().evaluate(params, batch)
+        partial_mse, partial_mse_terms = super().evaluate(params, batch)  # type: ignore
+        # ignore because batch is not PDEStatioBatch. We could use typing.cast though
 
         # initial condition
         if self.initial_condition_fun is not None:
             mse_initial_condition = initial_condition_apply(
                 self.u,
                 omega_batch,
-                _set_derivatives(params, self.derivative_keys.initial_condition),
+                _set_derivatives(params, self.derivative_keys.initial_condition),  # type: ignore
                 (0,) + vmap_in_axes_params,
                 self.initial_condition_fun,
-                self.loss_weights.initial_condition,
+                self.t0,  # type: ignore can't get the narrowing in __post_init__
+                self.loss_weights.initial_condition,  # type: ignore
             )
         else:
             mse_initial_condition = jnp.array(0.0)
@@ -673,419 +752,3 @@ class LossPDENonStatio(LossPDEStatio):
             **partial_mse_terms,
             "initial_condition": mse_initial_condition,
         }
-
-
-class SystemLossPDE(eqx.Module):
-    r"""
-    Class to implement a system of PDEs.
-    The goal is to give maximum freedom to the user. The class is created with
-    a dict of dynamic loss, and dictionaries of all the objects that are used
-    in LossPDENonStatio and LossPDEStatio. When then iterate
-    over the dynamic losses that compose the system. All the PINNs with all the
-    parameter dictionaries are passed as arguments to each dynamic loss
-    evaluate functions; it is inside the dynamic loss that specification are
-    performed.
-
-    **Note:** All the dictionaries (except `dynamic_loss_dict`) must have the same keys.
-    Indeed, these dictionaries (except `dynamic_loss_dict`) are tied to one
-    solution.
-
-    Parameters
-    ----------
-    u_dict : Dict[str, eqx.Module]
-        dict of PINNs
-    loss_weights : LossWeightsPDEDict
-        A dictionary of LossWeightsODE
-    derivative_keys_dict : Dict[str, DerivativeKeysPDEStatio | DerivativeKeysPDENonStatio], default=None
-        A dictionnary of DerivativeKeysPDEStatio or DerivativeKeysPDENonStatio
-        specifying what field of `params`
-        should be used during gradient computations for each of the terms of
-        the total loss, for each of the loss in the system. Default is
-        `"nn_params`" everywhere.
-    dynamic_loss_dict : Dict[str, PDEStatio | PDENonStatio]
-        A dict of dynamic part of the loss, basically the differential
-        operator $\mathcal{N}[u](t, x)$ or $\mathcal{N}[u](x)$.
-    key_dict : Dict[str, Key], default=None
-        A dictionary of JAX PRNG keys. The dictionary keys of key_dict must
-        match that of u_dict. See LossPDEStatio or LossPDENonStatio for
-        more details.
-    omega_boundary_fun_dict : Dict[str, Callable | Dict[str, Callable] | None], default=None
-        A dict of of function or of dict of functions or of None
-        (see doc for `omega_boundary_fun` in
-        LossPDEStatio or LossPDENonStatio). Default is None.
-        Must share the keys of `u_dict`.
-    omega_boundary_condition_dict : Dict[str, str | Dict[str, str] | None], default=None
-        A dict of strings or of dict of strings or of None
-        (see doc for `omega_boundary_condition_dict` in
-        LossPDEStatio or LossPDENonStatio). Default is None.
-        Must share the keys of `u_dict`
-    omega_boundary_dim_dict : Dict[str, slice | Dict[str, slice] | None], default=None
-        A dict of slices or of dict of slices or of None
-        (see doc for `omega_boundary_dim` in
-        LossPDEStatio or LossPDENonStatio). Default is None.
-        Must share the keys of `u_dict`
-    initial_condition_fun_dict : Dict[str, Callable | None], default=None
-        A dict of functions representing the temporal initial condition (None
-        value is possible). If None
-        (default) then no temporal boundary condition is applied
-        Must share the keys of `u_dict`
-    norm_samples_dict : Dict[str, Float[Array, "nb_norm_samples dimension"] | None, default=None
-        A dict of fixed sample point in the space over which to compute the
-        normalization constant. Default is None
-        Must share the keys of `u_dict`
-    norm_int_length_dict : Dict[str, float | None] | None, default=None
-        A dict of Float. The domain area
-        (or interval length in 1D) upon which we perform the numerical
-        integration for each element of u_dict.
-        Default is None
-        Must share the keys of `u_dict`
-    obs_slice_dict : Dict[str, slice | None] | None, default=None
-        dict of obs_slice, with keys from `u_dict` to designate the
-        output(s) channels that are forced to observed values, for each
-        PINNs. Default is None. But if a value is given, all the entries of
-        `u_dict` must be represented here with default value `jnp.s_[...]`
-        if no particular slice is to be given
-    params : InitVar[ParamsDict], default=None
-        The main Params object of the problem needed to instanciate the
-        DerivativeKeysODE if the latter is not specified.
-
-    """
-
-    # NOTE static=True only for leaf attributes that are not valid JAX types
-    # (ie. jax.Array cannot be static) and that we do not expect to change
-    u_dict: Dict[str, eqx.Module]
-    dynamic_loss_dict: Dict[str, PDEStatio | PDENonStatio]
-    key_dict: Dict[str, Key] | None = eqx.field(kw_only=True, default=None)
-    derivative_keys_dict: Dict[
-        str, DerivativeKeysPDEStatio | DerivativeKeysPDENonStatio | None
-    ] = eqx.field(kw_only=True, default=None)
-    omega_boundary_fun_dict: Dict[str, Callable | Dict[str, Callable] | None] | None = (
-        eqx.field(kw_only=True, default=None, static=True)
-    )
-    omega_boundary_condition_dict: Dict[str, str | Dict[str, str] | None] | None = (
-        eqx.field(kw_only=True, default=None, static=True)
-    )
-    omega_boundary_dim_dict: Dict[str, slice | Dict[str, slice] | None] | None = (
-        eqx.field(kw_only=True, default=None, static=True)
-    )
-    initial_condition_fun_dict: Dict[str, Callable | None] | None = eqx.field(
-        kw_only=True, default=None, static=True
-    )
-    norm_samples_dict: Dict[str, Float[Array, "nb_norm_samples dimension"]] | None = (
-        eqx.field(kw_only=True, default=None)
-    )
-    norm_int_length_dict: Dict[str, float | None] | None = eqx.field(
-        kw_only=True, default=None
-    )
-    obs_slice_dict: Dict[str, slice | None] | None = eqx.field(
-        kw_only=True, default=None, static=True
-    )
-
-    # For the user loss_weights are passed as a LossWeightsPDEDict (with internal
-    # dictionary having keys in u_dict and / or dynamic_loss_dict)
-    loss_weights: InitVar[LossWeightsPDEDict | None] = eqx.field(
-        kw_only=True, default=None
-    )
-    params_dict: InitVar[ParamsDict] = eqx.field(kw_only=True, default=None)
-
-    # following have init=False and are set in the __post_init__
-    u_constraints_dict: Dict[str, LossPDEStatio | LossPDENonStatio] = eqx.field(
-        init=False
-    )
-    derivative_keys_dyn_loss: DerivativeKeysPDEStatio | DerivativeKeysPDENonStatio = (
-        eqx.field(init=False)
-    )
-    u_dict_with_none: Dict[str, None] = eqx.field(init=False)
-    # internally the loss weights are handled with a dictionary
-    _loss_weights: Dict[str, dict] = eqx.field(init=False)
-
-    def __post_init__(self, loss_weights=None, params_dict=None):
-        # a dictionary that will be useful at different places
-        self.u_dict_with_none = {k: None for k in self.u_dict.keys()}
-        # First, for all the optional dict,
-        # if the user did not provide at all this optional argument,
-        # we make sure there is a null ponderating loss_weight and we
-        # create a dummy dict with the required keys and all the values to
-        # None
-        if self.key_dict is None:
-            self.key_dict = self.u_dict_with_none
-        if self.omega_boundary_fun_dict is None:
-            self.omega_boundary_fun_dict = self.u_dict_with_none
-        if self.omega_boundary_condition_dict is None:
-            self.omega_boundary_condition_dict = self.u_dict_with_none
-        if self.omega_boundary_dim_dict is None:
-            self.omega_boundary_dim_dict = self.u_dict_with_none
-        if self.initial_condition_fun_dict is None:
-            self.initial_condition_fun_dict = self.u_dict_with_none
-        if self.norm_samples_dict is None:
-            self.norm_samples_dict = self.u_dict_with_none
-        if self.norm_int_length_dict is None:
-            self.norm_int_length_dict = self.u_dict_with_none
-        if self.obs_slice_dict is None:
-            self.obs_slice_dict = {k: jnp.s_[...] for k in self.u_dict.keys()}
-            if self.u_dict.keys() != self.obs_slice_dict.keys():
-                raise ValueError("obs_slice_dict should have same keys as u_dict")
-        if self.derivative_keys_dict is None:
-            self.derivative_keys_dict = {
-                k: None
-                for k in set(
-                    list(self.dynamic_loss_dict.keys()) + list(self.u_dict.keys())
-                )
-            }
-            # set() because we can have duplicate entries and in this case we
-            # say it corresponds to the same derivative_keys_dict entry
-            # we need both because the constraints (all but dyn_loss) will be
-            # done by iterating on u_dict while the dyn_loss will be by
-            # iterating on dynamic_loss_dict. So each time we will require dome
-            # derivative_keys_dict
-
-        # derivative keys for the u_constraints. Note that we create missing
-        # DerivativeKeysODE around a Params object and not ParamsDict
-        # this works because u_dict.keys == params_dict.nn_params.keys()
-        for k in self.u_dict.keys():
-            if self.derivative_keys_dict[k] is None:
-                if self.u_dict[k].eq_type == "statio_PDE":
-                    self.derivative_keys_dict[k] = DerivativeKeysPDEStatio(
-                        params=params_dict.extract_params(k)
-                    )
-                else:
-                    self.derivative_keys_dict[k] = DerivativeKeysPDENonStatio(
-                        params=params_dict.extract_params(k)
-                    )
-
-        # Second we make sure that all the dicts (except dynamic_loss_dict) have the same keys
-        if (
-            self.u_dict.keys() != self.key_dict.keys()
-            or self.u_dict.keys() != self.omega_boundary_fun_dict.keys()
-            or self.u_dict.keys() != self.omega_boundary_condition_dict.keys()
-            or self.u_dict.keys() != self.omega_boundary_dim_dict.keys()
-            or self.u_dict.keys() != self.initial_condition_fun_dict.keys()
-            or self.u_dict.keys() != self.norm_samples_dict.keys()
-            or self.u_dict.keys() != self.norm_int_length_dict.keys()
-        ):
-            raise ValueError("All the dicts concerning the PINNs should have same keys")
-
-        self._loss_weights = self.set_loss_weights(loss_weights)
-
-        # Third, in order not to benefit from LossPDEStatio and
-        # LossPDENonStatio and in order to factorize code, we create internally
-        # some losses object to implement the constraints on the solutions.
-        # We will not use the dynamic loss term
-        self.u_constraints_dict = {}
-        for i in self.u_dict.keys():
-            if self.u_dict[i].eq_type == "statio_PDE":
-                self.u_constraints_dict[i] = LossPDEStatio(
-                    u=self.u_dict[i],
-                    loss_weights=LossWeightsPDENonStatio(
-                        dyn_loss=0.0,
-                        norm_loss=1.0,
-                        boundary_loss=1.0,
-                        observations=1.0,
-                        initial_condition=1.0,
-                    ),
-                    dynamic_loss=None,
-                    key=self.key_dict[i],
-                    derivative_keys=self.derivative_keys_dict[i],
-                    omega_boundary_fun=self.omega_boundary_fun_dict[i],
-                    omega_boundary_condition=self.omega_boundary_condition_dict[i],
-                    omega_boundary_dim=self.omega_boundary_dim_dict[i],
-                    norm_samples=self.norm_samples_dict[i],
-                    norm_int_length=self.norm_int_length_dict[i],
-                    obs_slice=self.obs_slice_dict[i],
-                )
-            elif self.u_dict[i].eq_type == "nonstatio_PDE":
-                self.u_constraints_dict[i] = LossPDENonStatio(
-                    u=self.u_dict[i],
-                    loss_weights=LossWeightsPDENonStatio(
-                        dyn_loss=0.0,
-                        norm_loss=1.0,
-                        boundary_loss=1.0,
-                        observations=1.0,
-                        initial_condition=1.0,
-                    ),
-                    dynamic_loss=None,
-                    key=self.key_dict[i],
-                    derivative_keys=self.derivative_keys_dict[i],
-                    omega_boundary_fun=self.omega_boundary_fun_dict[i],
-                    omega_boundary_condition=self.omega_boundary_condition_dict[i],
-                    omega_boundary_dim=self.omega_boundary_dim_dict[i],
-                    initial_condition_fun=self.initial_condition_fun_dict[i],
-                    norm_samples=self.norm_samples_dict[i],
-                    norm_int_length=self.norm_int_length_dict[i],
-                    obs_slice=self.obs_slice_dict[i],
-                )
-            else:
-                raise ValueError(
-                    "Wrong value for self.u_dict[i].eq_type[i], "
-                    f"got {self.u_dict[i].eq_type[i]}"
-                )
-
-        # derivative keys for the dynamic loss. Note that we create a
-        # DerivativeKeysODE around a ParamsDict object because a whole
-        # params_dict is feed to DynamicLoss.evaluate functions (extract_params
-        # happen inside it)
-        self.derivative_keys_dyn_loss = DerivativeKeysPDENonStatio(params=params_dict)
-
-        # also make sure we only have PINNs or SPINNs
-        if not (
-            all(isinstance(value, PINN) for value in self.u_dict.values())
-            or all(isinstance(value, SPINN) for value in self.u_dict.values())
-        ):
-            raise ValueError(
-                "We only accept dictionary of PINNs or dictionary of SPINNs"
-            )
-
-    def set_loss_weights(
-        self, loss_weights_init: LossWeightsPDEDict
-    ) -> dict[str, dict]:
-        """
-        This rather complex function enables the user to specify a simple
-        loss_weights=LossWeightsPDEDict(dyn_loss=1., initial_condition=Tmax)
-        for ponderating values being applied to all the equations of the
-        system... So all the transformations are handled here
-        """
-        _loss_weights = {}
-        for k in fields(loss_weights_init):
-            v = getattr(loss_weights_init, k.name)
-            if isinstance(v, dict):
-                for vv in v.keys():
-                    if not isinstance(vv, (int, float)) and not (
-                        isinstance(vv, Array)
-                        and ((vv.shape == (1,) or len(vv.shape) == 0))
-                    ):
-                        # TODO improve that
-                        raise ValueError(
-                            f"loss values cannot be vectorial here, got {vv}"
-                        )
-                if k.name == "dyn_loss":
-                    if v.keys() == self.dynamic_loss_dict.keys():
-                        _loss_weights[k.name] = v
-                    else:
-                        raise ValueError(
-                            "Keys in nested dictionary of loss_weights"
-                            " do not match dynamic_loss_dict keys"
-                        )
-                else:
-                    if v.keys() == self.u_dict.keys():
-                        _loss_weights[k.name] = v
-                    else:
-                        raise ValueError(
-                            "Keys in nested dictionary of loss_weights"
-                            " do not match u_dict keys"
-                        )
-            if v is None:
-                _loss_weights[k.name] = {kk: 0 for kk in self.u_dict.keys()}
-            else:
-                if not isinstance(v, (int, float)) and not (
-                    isinstance(v, Array) and ((v.shape == (1,) or len(v.shape) == 0))
-                ):
-                    # TODO improve that
-                    raise ValueError(f"loss values cannot be vectorial here, got {v}")
-                if k.name == "dyn_loss":
-                    _loss_weights[k.name] = {
-                        kk: v for kk in self.dynamic_loss_dict.keys()
-                    }
-                else:
-                    _loss_weights[k.name] = {kk: v for kk in self.u_dict.keys()}
-        return _loss_weights
-
-    def __call__(self, *args, **kwargs):
-        return self.evaluate(*args, **kwargs)
-
-    def evaluate(
-        self,
-        params_dict: ParamsDict,
-        batch: PDEStatioBatch | PDENonStatioBatch,
-    ) -> tuple[Float[Array, "1"], dict[str, float]]:
-        """
-        Evaluate the loss function at a batch of points for given parameters.
-
-
-        Parameters
-        ---------
-        params_dict
-            Parameters at which the losses of the system are evaluated
-        batch
-            Such named tuples are composed of  batch of points in the
-            domain, a batch of points in the domain
-            border, (a batch of time points a for PDENonStatioBatch) and an
-            optional additional batch of parameters (eg. for metamodeling)
-            and an optional additional batch of observed
-            inputs/outputs/parameters
-        """
-        if self.u_dict.keys() != params_dict.nn_params.keys():
-            raise ValueError("u_dict and params_dict[nn_params] should have same keys ")
-
-        vmap_in_axes = (0,)
-
-        # Retrieve the optional eq_params_batch
-        # and update eq_params with the latter
-        # and update vmap_in_axes
-        if batch.param_batch_dict is not None:
-            eq_params_batch_dict = batch.param_batch_dict
-
-            # feed the eq_params with the batch
-            for k in eq_params_batch_dict.keys():
-                params_dict.eq_params[k] = eq_params_batch_dict[k]
-
-        vmap_in_axes_params = _get_vmap_in_axes_params(
-            batch.param_batch_dict, params_dict
-        )
-
-        def dyn_loss_for_one_key(dyn_loss, loss_weight):
-            """The function used in tree_map"""
-            return dynamic_loss_apply(
-                dyn_loss.evaluate,
-                self.u_dict,
-                (
-                    batch.domain_batch
-                    if isinstance(batch, PDEStatioBatch)
-                    else batch.domain_batch
-                ),
-                _set_derivatives(params_dict, self.derivative_keys_dyn_loss.dyn_loss),
-                vmap_in_axes + vmap_in_axes_params,
-                loss_weight,
-                u_type=type(list(self.u_dict.values())[0]),
-            )
-
-        dyn_loss_mse_dict = jax.tree_util.tree_map(
-            dyn_loss_for_one_key,
-            self.dynamic_loss_dict,
-            self._loss_weights["dyn_loss"],
-            is_leaf=lambda x: isinstance(
-                x, (PDEStatio, PDENonStatio)
-            ),  # before when dynamic losses
-            # where plain (unregister pytree) node classes, we could not traverse
-            # this level. Now that dynamic losses are eqx.Module they can be
-            # traversed by tree map recursion. Hence we need to specify to that
-            # we want to stop at this level
-        )
-        mse_dyn_loss = jax.tree_util.tree_reduce(
-            lambda x, y: x + y, jax.tree_util.tree_leaves(dyn_loss_mse_dict)
-        )
-
-        # boundary conditions, normalization conditions, observation_loss,
-        # initial condition... loss this is done via the internal
-        # LossPDEStatio and NonStatio
-        loss_weight_struct = {
-            "dyn_loss": "*",
-            "norm_loss": "*",
-            "boundary_loss": "*",
-            "observations": "*",
-            "initial_condition": "*",
-        }
-        # we need to do the following for the tree_mapping to work
-        if batch.obs_batch_dict is None:
-            batch = append_obs_batch(batch, self.u_dict_with_none)
-        total_loss, res_dict = constraints_system_loss_apply(
-            self.u_constraints_dict,
-            batch,
-            params_dict,
-            self._loss_weights,
-            loss_weight_struct,
-        )
-
-        # Add the mse_dyn_loss from the previous computations
-        total_loss += mse_dyn_loss
-        res_dict["dyn_loss"] += mse_dyn_loss
-        return total_loss, res_dict