qgt_onthefly.py

# Copyright 2021 The NetKet Authors - All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from typing import Callable, Optional, Union
from functools import partial
from textwrap import dedent

import jax
from jax import numpy as jnp
from flax import struct

import netket.jax as nkjax
from netket.utils.types import PyTree
from netket.utils import warn_deprecation

from .qgt_onthefly_logic import mat_vec_factory, mat_vec_chunked_factory

from ..linear_operator import LinearOperator, Uninitialized


def check_valid_vector_type(x, target):
    """
    Raises a TypeError if x is complex where target is real, because it is not
    supported by QGTOnTheFly and the imaginary part would be dicscarded after
    anyhow.
    """

    def check(x, target):
        if jnp.iscomplexobj(target) and not jnp.iscomplexobj(x):
            raise TypeError(
                dedent(
                    """
                    Cannot multiply the (real part of the) QGT by a complex vector.
                    You should either take the real part of the vector, or perform 
                    the multiplication against the real and imaginary part of the
                    vector separately and then recomposing the two.
                    """
                )
            )

    jax.tree_multimap(check, x, target)


def QGTOnTheFly(vstate=None, **kwargs) -> "QGTOnTheFlyT":
    """
    Lazy representation of an S Matrix computed by performing 2 jvp
    and 1 vjp products, using the variational state's model, the
    samples that have already been computed, and the vector.

    The S matrix is not computed yet, but can be computed by calling
    :code:`to_dense`.
    The details on how the ⟨S⟩⁻¹⟨F⟩ system is solved are contaianed in
    the field `sr`.

    Args:
        vstate: The variational State.
    """
    if vstate is None:
        return partial(QGTOnTheFly, **kwargs)

    if "centered" in kwargs:
        warn_deprecation(
            "The argument `centered` is deprecated. The implementation now always behaves as if centered=False."
        )
        kwargs.pop("centered")

    # TODO: Find a better way to handle this case
    from netket.vqs import ExactState

    if isinstance(vstate, ExactState):
        raise TypeError("Only QGTJacobianPyTree works with ExactState.")

    if jnp.ndim(vstate.samples) == 2:
        samples = vstate.samples
    else:
        samples = vstate.samples.reshape((-1, vstate.samples.shape[-1]))

    chunk_size = vstate.chunk_size
    n_samples = samples.shape[0]

    if chunk_size is None or chunk_size >= n_samples:
        mv_factory = mat_vec_factory
        chunking = False
    else:
        samples, _ = nkjax.chunk(samples, chunk_size)
        mv_factory = mat_vec_chunked_factory
        chunking = True

    mat_vec = mv_factory(
        forward_fn=vstate._apply_fun,
        params=vstate.parameters,
        model_state=vstate.model_state,
        samples=samples,
    )
    return QGTOnTheFlyT(
        _mat_vec=mat_vec,
        _params=vstate.parameters,
        _chunking=chunking,
        **kwargs,
    )


@struct.dataclass
class QGTOnTheFlyT(LinearOperator):
    """
    Lazy representation of an S Matrix computed by performing 2 jvp
    and 1 vjp products, using the variational state's model, the
    samples that have already been computed, and the vector.

    The S matrix is not computed yet, but can be computed by calling
    :code:`to_dense`.
    The details on how the ⟨S⟩⁻¹⟨F⟩ system is solved are contaianed in
    the field `sr`.
    """

    _mat_vec: Callable[[PyTree, float], PyTree] = Uninitialized
    """The S matrix-vector product as generated by mat_vec_factory.
    It's a jax.Partial, so can be used as pytree_node."""

    _params: PyTree = Uninitialized
    """The first input to apply_fun (parameters of the ansatz).
    Only used as a shape placeholder."""

    _chunking: bool = struct.field(pytree_node=False, default=False)
    """Wether the implementation with chunks is used which currently does not support vmapping over it"""

    def __matmul__(self, y):
        return onthefly_mat_treevec(self, y)

    def _solve(self, solve_fun, y: PyTree, *, x0: Optional[PyTree], **kwargs) -> PyTree:
        return _solve(self, solve_fun, y, x0=x0)

    def to_dense(self) -> jnp.ndarray:
        """
        Convert the lazy matrix representation to a dense matrix representation.

        Returns:
            A dense matrix representation of this S matrix.
        """
        return _to_dense(self)

    def __repr__(self):
        return f"QGTOnTheFly(diag_shift={self.diag_shift})"


@jax.jit
def onthefly_mat_treevec(
    S: QGTOnTheFly, vec: Union[PyTree, jnp.ndarray]
) -> Union[PyTree, jnp.ndarray]:
    """
    Perform the lazy mat-vec product, where vec is either a tree with the same structure as
    params or a ravelled vector
    """

    # if hasa ndim it's an array and not a pytree
    if hasattr(vec, "ndim"):
        if not vec.ndim == 1:
            raise ValueError("Unsupported mat-vec for chunks of vectors")
        # If the input is a vector
        if not nkjax.tree_size(S._params) == vec.size:
            raise ValueError(
                """Size mismatch between number of parameters ({nkjax.tree_size(S.params)})
                                and vector size {vec.size}.
                             """
            )

        _, unravel = nkjax.tree_ravel(S._params)
        vec = unravel(vec)
        ravel_result = True
    else:
        ravel_result = False

    check_valid_vector_type(vec, S._params)

    vec = nkjax.tree_cast(vec, S._params)

    res = S._mat_vec(vec, S.diag_shift)

    if ravel_result:
        res, _ = nkjax.tree_ravel(res)

    return res


@jax.jit
def _solve(
    self: QGTOnTheFlyT, solve_fun, y: PyTree, *, x0: Optional[PyTree], **kwargs
) -> PyTree:

    check_valid_vector_type(y, self._params)

    y = nkjax.tree_cast(y, self._params)

    # we could cache this...
    if x0 is None:
        x0 = jax.tree_map(jnp.zeros_like, y)

    out, info = solve_fun(self, y, x0=x0)
    return out, info


@jax.jit
def _to_dense(self: QGTOnTheFlyT) -> jnp.ndarray:
    """
    Convert the lazy matrix representation to a dense matrix representation

    Returns:
        A dense matrix representation of this S matrix.
    """
    Npars = nkjax.tree_size(self._params)
    I = jax.numpy.eye(Npars)

    if self._chunking:
        # the linear_call in mat_vec_chunked does currently not have a jax batching rule,
        # so it cannot be vmapped but we can use scan
        # which is better for reducing the memory consumption anyway
        _, out = jax.lax.scan(lambda _, x: (None, self @ x), None, I)
    else:
        out = jax.vmap(lambda x: self @ x, in_axes=0)(I)

    if nkjax.is_complex(out):
        out = out.T

    return out