You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
338 lines
9.7 KiB
338 lines
9.7 KiB
# MIPLearn: Extensible Framework for Learning-Enhanced Mixed-Integer Optimization
|
|
# Copyright (C) 2020-2021, UChicago Argonne, LLC. All rights reserved.
|
|
# Released under the modified BSD license. See COPYING.md for more details.
|
|
|
|
from abc import ABC, abstractmethod
|
|
from copy import deepcopy
|
|
from typing import Dict, Optional, Any, Union, List, Tuple, cast, Set
|
|
|
|
import h5py
|
|
import numpy as np
|
|
from h5py import Dataset
|
|
from overrides import overrides
|
|
|
|
Bytes = Union[bytes, bytearray]
|
|
Scalar = Union[None, bool, str, int, float]
|
|
Vector = Union[
|
|
None,
|
|
List[bool],
|
|
List[str],
|
|
List[int],
|
|
List[float],
|
|
List[Optional[str]],
|
|
np.ndarray,
|
|
]
|
|
VectorList = Union[
|
|
List[List[bool]],
|
|
List[List[str]],
|
|
List[List[int]],
|
|
List[List[float]],
|
|
List[Optional[List[bool]]],
|
|
List[Optional[List[str]]],
|
|
List[Optional[List[int]]],
|
|
List[Optional[List[float]]],
|
|
]
|
|
|
|
|
|
class Sample(ABC):
|
|
"""Abstract dictionary-like class that stores training data."""
|
|
|
|
@abstractmethod
|
|
def get_bytes(self, key: str) -> Optional[Bytes]:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def put_bytes(self, key: str, value: Bytes) -> None:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def get_scalar(self, key: str) -> Optional[Any]:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def put_scalar(self, key: str, value: Scalar) -> None:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def get_vector(self, key: str) -> Optional[Any]:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def put_vector(self, key: str, value: Vector) -> None:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def get_vector_list(self, key: str) -> Optional[Any]:
|
|
pass
|
|
|
|
@abstractmethod
|
|
def put_vector_list(self, key: str, value: VectorList) -> None:
|
|
pass
|
|
|
|
def get_set(self, key: str) -> Set:
|
|
v = self.get_vector(key)
|
|
if v:
|
|
return set(v)
|
|
else:
|
|
return set()
|
|
|
|
def put_set(self, key: str, value: Set) -> None:
|
|
v = list(value)
|
|
self.put_vector(key, v)
|
|
|
|
def _assert_is_scalar(self, value: Any) -> None:
|
|
if value is None:
|
|
return
|
|
if isinstance(value, (str, bool, int, float)):
|
|
return
|
|
assert False, f"scalar expected; found instead: {value}"
|
|
|
|
def _assert_is_vector(self, value: Any) -> None:
|
|
assert isinstance(
|
|
value, (list, np.ndarray)
|
|
), f"list or numpy array expected; found instead: {value}"
|
|
for v in value:
|
|
self._assert_is_scalar(v)
|
|
|
|
def _assert_is_vector_list(self, value: Any) -> None:
|
|
assert isinstance(
|
|
value, (list, np.ndarray)
|
|
), f"list or numpy array expected; found instead: {value}"
|
|
for v in value:
|
|
if v is None:
|
|
continue
|
|
self._assert_is_vector(v)
|
|
|
|
|
|
class MemorySample(Sample):
|
|
"""Dictionary-like class that stores training data in-memory."""
|
|
|
|
def __init__(
|
|
self,
|
|
data: Optional[Dict[str, Any]] = None,
|
|
) -> None:
|
|
if data is None:
|
|
data = {}
|
|
self._data: Dict[str, Any] = data
|
|
|
|
@overrides
|
|
def get_bytes(self, key: str) -> Optional[Bytes]:
|
|
return self._get(key)
|
|
|
|
@overrides
|
|
def get_scalar(self, key: str) -> Optional[Any]:
|
|
return self._get(key)
|
|
|
|
@overrides
|
|
def get_vector(self, key: str) -> Optional[Any]:
|
|
return self._get(key)
|
|
|
|
@overrides
|
|
def get_vector_list(self, key: str) -> Optional[Any]:
|
|
return self._get(key)
|
|
|
|
@overrides
|
|
def put_bytes(self, key: str, value: Bytes) -> None:
|
|
assert isinstance(
|
|
value, (bytes, bytearray)
|
|
), f"bytes expected; found: {value}" # type: ignore
|
|
self._put(key, value)
|
|
|
|
@overrides
|
|
def put_scalar(self, key: str, value: Scalar) -> None:
|
|
if value is None:
|
|
return
|
|
self._assert_is_scalar(value)
|
|
self._put(key, value)
|
|
|
|
@overrides
|
|
def put_vector(self, key: str, value: Vector) -> None:
|
|
if value is None:
|
|
return
|
|
self._assert_is_vector(value)
|
|
self._put(key, value)
|
|
|
|
@overrides
|
|
def put_vector_list(self, key: str, value: VectorList) -> None:
|
|
self._assert_is_vector_list(value)
|
|
self._put(key, value)
|
|
|
|
def _get(self, key: str) -> Optional[Any]:
|
|
if key in self._data:
|
|
return self._data[key]
|
|
else:
|
|
return None
|
|
|
|
def _put(self, key: str, value: Any) -> None:
|
|
self._data[key] = value
|
|
|
|
|
|
class Hdf5Sample(Sample):
|
|
"""
|
|
Dictionary-like class that stores training data in an HDF5 file.
|
|
|
|
Unlike MemorySample, this class only loads to memory the parts of the data set that
|
|
are actually accessed, and therefore it is more scalable.
|
|
"""
|
|
|
|
def __init__(self, filename: str, mode: str = "r+") -> None:
|
|
self.file = h5py.File(filename, mode, libver="latest")
|
|
|
|
@overrides
|
|
def get_bytes(self, key: str) -> Optional[Bytes]:
|
|
if key not in self.file:
|
|
return None
|
|
ds = self.file[key]
|
|
assert (
|
|
len(ds.shape) == 1
|
|
), f"1-dimensional array expected; found shape {ds.shape}"
|
|
return ds[()].tobytes()
|
|
|
|
@overrides
|
|
def get_scalar(self, key: str) -> Optional[Any]:
|
|
if key not in self.file:
|
|
return None
|
|
ds = self.file[key]
|
|
assert (
|
|
len(ds.shape) == 0
|
|
), f"0-dimensional array expected; found shape {ds.shape}"
|
|
if h5py.check_string_dtype(ds.dtype):
|
|
return ds.asstr()[()]
|
|
else:
|
|
return ds[()].tolist()
|
|
|
|
@overrides
|
|
def get_vector(self, key: str) -> Optional[Any]:
|
|
if key not in self.file:
|
|
return None
|
|
ds = self.file[key]
|
|
assert (
|
|
len(ds.shape) == 1
|
|
), f"1-dimensional array expected; found shape {ds.shape}"
|
|
if h5py.check_string_dtype(ds.dtype):
|
|
result = ds.asstr()[:].tolist()
|
|
result = [r if len(r) > 0 else None for r in result]
|
|
return result
|
|
else:
|
|
return ds[:].tolist()
|
|
|
|
@overrides
|
|
def get_vector_list(self, key: str) -> Optional[Any]:
|
|
if key not in self.file:
|
|
return None
|
|
ds = self.file[key]
|
|
lens = self.get_vector(f"{key}_lengths")
|
|
if h5py.check_string_dtype(ds.dtype):
|
|
padded = ds.asstr()[:].tolist()
|
|
else:
|
|
padded = ds[:].tolist()
|
|
return _crop(padded, lens)
|
|
|
|
@overrides
|
|
def put_bytes(self, key: str, value: Bytes) -> None:
|
|
assert isinstance(
|
|
value, (bytes, bytearray)
|
|
), f"bytes expected; found: {value}" # type: ignore
|
|
self._put(key, np.frombuffer(value, dtype="uint8"), compress=True)
|
|
|
|
@overrides
|
|
def put_scalar(self, key: str, value: Any) -> None:
|
|
if value is None:
|
|
return
|
|
self._assert_is_scalar(value)
|
|
self._put(key, value)
|
|
|
|
@overrides
|
|
def put_vector(self, key: str, value: Vector) -> None:
|
|
if value is None:
|
|
return
|
|
self._assert_is_vector(value)
|
|
|
|
for v in value:
|
|
# Convert strings to bytes
|
|
if isinstance(v, str):
|
|
value = np.array(
|
|
[u if u is not None else b"" for u in value],
|
|
dtype="S",
|
|
)
|
|
break
|
|
|
|
# Convert all floating point numbers to half-precision
|
|
if isinstance(v, float):
|
|
value = np.array(value, dtype=np.dtype("f2"))
|
|
break
|
|
|
|
self._put(key, value, compress=True)
|
|
|
|
@overrides
|
|
def put_vector_list(self, key: str, value: VectorList) -> None:
|
|
self._assert_is_vector_list(value)
|
|
padded, lens = _pad(value)
|
|
self.put_vector(f"{key}_lengths", lens)
|
|
data = None
|
|
for v in value:
|
|
if v is None or len(v) == 0:
|
|
continue
|
|
if isinstance(v[0], str):
|
|
data = np.array(padded, dtype="S")
|
|
elif isinstance(v[0], float):
|
|
data = np.array(padded, dtype=np.dtype("f2"))
|
|
elif isinstance(v[0], bool):
|
|
data = np.array(padded, dtype=bool)
|
|
else:
|
|
data = np.array(padded)
|
|
break
|
|
if data is None:
|
|
data = np.array(padded)
|
|
self._put(key, data, compress=True)
|
|
|
|
def _put(self, key: str, value: Any, compress: bool = False) -> Dataset:
|
|
if key in self.file:
|
|
del self.file[key]
|
|
if compress:
|
|
ds = self.file.create_dataset(key, data=value, compression="gzip")
|
|
else:
|
|
ds = self.file.create_dataset(key, data=value)
|
|
return ds
|
|
|
|
|
|
def _pad(veclist: VectorList) -> Tuple[VectorList, List[int]]:
|
|
veclist = deepcopy(veclist)
|
|
lens = [len(v) if v is not None else -1 for v in veclist]
|
|
maxlen = max(lens)
|
|
|
|
# Find appropriate constant to pad the vectors
|
|
constant: Union[int, float, str] = 0
|
|
for v in veclist:
|
|
if v is None or len(v) == 0:
|
|
continue
|
|
if isinstance(v[0], int):
|
|
constant = 0
|
|
elif isinstance(v[0], float):
|
|
constant = 0.0
|
|
elif isinstance(v[0], str):
|
|
constant = ""
|
|
else:
|
|
assert False, f"unsupported data type: {v[0]}"
|
|
|
|
# Pad vectors
|
|
for (i, vi) in enumerate(veclist):
|
|
if vi is None:
|
|
vi = veclist[i] = []
|
|
assert isinstance(vi, list), f"list expected; found: {vi}"
|
|
for k in range(len(vi), maxlen):
|
|
vi.append(constant)
|
|
|
|
return veclist, lens
|
|
|
|
|
|
def _crop(veclist: VectorList, lens: List[int]) -> VectorList:
|
|
result: VectorList = cast(VectorList, [])
|
|
for (i, v) in enumerate(veclist):
|
|
if lens[i] < 0:
|
|
result.append(None) # type: ignore
|
|
else:
|
|
assert isinstance(v, list)
|
|
result.append(v[: lens[i]])
|
|
return result
|