pixeltable 0.1.2__py3-none-any.whl → 0.2.1__py3-none-any.whl

pixeltable/exprs/comparison.py

@@ -0,0 +1,77 @@
+from __future__ import annotations
+from typing import Optional, List, Any, Dict, Tuple
+
+import sqlalchemy as sql
+
+from .globals import ComparisonOperator
+from .expr import Expr
+from .predicate import Predicate
+from .data_row import DataRow
+from .row_builder import RowBuilder
+import pixeltable.catalog as catalog
+
+
+class Comparison(Predicate):
+    def __init__(self, operator: ComparisonOperator, op1: Expr, op2: Expr):
+        super().__init__()
+        self.operator = operator
+        self.components = [op1, op2]
+        self.id = self._create_id()
+
+    def __str__(self) -> str:
+        return f'{self._op1} {self.operator} {self._op2}'
+
+    def _equals(self, other: Comparison) -> bool:
+        return self.operator == other.operator
+
+    def _id_attrs(self) -> List[Tuple[str, Any]]:
+        return super()._id_attrs() + [('operator', self.operator.value)]
+
+    @property
+    def _op1(self) -> Expr:
+        return self.components[0]
+
+    @property
+    def _op2(self) -> Expr:
+        return self.components[1]
+
+    def sql_expr(self) -> Optional[sql.ClauseElement]:
+        left = self._op1.sql_expr()
+        right = self._op2.sql_expr()
+        if left is None or right is None:
+            return None
+        if self.operator == ComparisonOperator.LT:
+            return left < right
+        if self.operator == ComparisonOperator.LE:
+            return left <= right
+        if self.operator == ComparisonOperator.EQ:
+            return left == right
+        if self.operator == ComparisonOperator.NE:
+            return left != right
+        if self.operator == ComparisonOperator.GT:
+            return left > right
+        if self.operator == ComparisonOperator.GE:
+            return left >= right
+
+    def eval(self, data_row: DataRow, row_builder: RowBuilder) -> None:
+        if self.operator == ComparisonOperator.LT:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] < data_row[self._op2.slot_idx]
+        elif self.operator == ComparisonOperator.LE:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] <= data_row[self._op2.slot_idx]
+        elif self.operator == ComparisonOperator.EQ:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] == data_row[self._op2.slot_idx]
+        elif self.operator == ComparisonOperator.NE:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] != data_row[self._op2.slot_idx]
+        elif self.operator == ComparisonOperator.GT:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] > data_row[self._op2.slot_idx]
+        elif self.operator == ComparisonOperator.GE:
+            data_row[self.slot_idx] = data_row[self._op1.slot_idx] >= data_row[self._op2.slot_idx]
+
+    def _as_dict(self) -> Dict:
+        return {'operator': self.operator.value, **super()._as_dict()}
+
+    @classmethod
+    def _from_dict(cls, d: Dict, components: List[Expr]) -> Expr:
+        assert 'operator' in d
+        return cls(ComparisonOperator(d['operator']), components[0], components[1])
+

pixeltable/exprs/compound_predicate.py

@@ -0,0 +1,98 @@
+from __future__ import annotations
+from typing import Optional, List, Any, Dict, Tuple, Callable
+import operator
+
+import sqlalchemy as sql
+
+from .expr import Expr
+from .globals import LogicalOperator
+from .predicate import Predicate
+from .data_row import DataRow
+from .row_builder import RowBuilder
+import pixeltable.catalog as catalog
+
+
+class CompoundPredicate(Predicate):
+    def __init__(self, operator: LogicalOperator, operands: List[Predicate]):
+        super().__init__()
+        self.operator = operator
+        # operands are stored in self.components
+        if self.operator == LogicalOperator.NOT:
+            assert len(operands) == 1
+            self.components = operands
+        else:
+            assert len(operands) > 1
+            self.operands: List[Predicate] = []
+            for operand in operands:
+                self._merge_operand(operand)
+
+        self.id = self._create_id()
+
+    def __str__(self) -> str:
+        if self.operator == LogicalOperator.NOT:
+            return f'~({self.components[0]})'
+        return f' {self.operator} '.join([f'({e})' for e in self.components])
+
+    @classmethod
+    def make_conjunction(cls, operands: List[Predicate]) -> Optional[Predicate]:
+        if len(operands) == 0:
+            return None
+        if len(operands) == 1:
+            return operands[0]
+        return CompoundPredicate(LogicalOperator.AND, operands)
+
+    def _merge_operand(self, operand: Predicate) -> None:
+        """
+        Merge this operand, if possible, otherwise simply record it.
+        """
+        if isinstance(operand, CompoundPredicate) and operand.operator == self.operator:
+            # this can be merged
+            for child_op in operand.components:
+                self._merge_operand(child_op)
+        else:
+            self.components.append(operand)
+
+    def _equals(self, other: CompoundPredicate) -> bool:
+        return self.operator == other.operator
+
+    def _id_attrs(self) -> List[Tuple[str, Any]]:
+        return super()._id_attrs() + [('operator', self.operator.value)]
+
+    def split_conjuncts(
+            self, condition: Callable[[Predicate], bool]) -> Tuple[List[Predicate], Optional[Predicate]]:
+        if self.operator == LogicalOperator.OR or self.operator == LogicalOperator.NOT:
+            return super().split_conjuncts(condition)
+        matches = [op for op in self.components if condition(op)]
+        non_matches = [op for op in self.components if not condition(op)]
+        return (matches, self.make_conjunction(non_matches))
+
+    def sql_expr(self) -> Optional[sql.ClauseElement]:
+        sql_exprs = [op.sql_expr() for op in self.components]
+        if any(e is None for e in sql_exprs):
+            return None
+        if self.operator == LogicalOperator.NOT:
+            assert len(sql_exprs) == 1
+            return sql.not_(sql_exprs[0])
+        assert len(sql_exprs) > 1
+        operator = sql.and_ if self.operator == LogicalOperator.AND else sql.or_
+        combined = operator(*sql_exprs)
+        return combined
+
+    def eval(self, data_row: DataRow, row_builder: RowBuilder) -> None:
+        if self.operator == LogicalOperator.NOT:
+            data_row[self.slot_idx] = not data_row[self.components[0].slot_idx]
+        else:
+            val = True if self.operator == LogicalOperator.AND else False
+            op_function = operator.and_ if self.operator == LogicalOperator.AND else operator.or_
+            for op in self.components:
+                val = op_function(val, data_row[op.slot_idx])
+            data_row[self.slot_idx] = val
+
+    def _as_dict(self) -> Dict:
+        return {'operator': self.operator.value, **super()._as_dict()}
+
+    @classmethod
+    def _from_dict(cls, d: Dict, components: List[Expr]) -> Expr:
+        assert 'operator' in d
+        return cls(LogicalOperator(d['operator']), components)
+

pixeltable/exprs/data_row.py

@@ -0,0 +1,187 @@
+from __future__ import annotations
+from typing import Optional, List, Any, Tuple
+import io
+import urllib
+
+import PIL
+import numpy as np
+
+
+class DataRow:
+    """
+    Encapsulates all data and execution state needed by RowBuilder and DataRowBatch:
+    - state for in-memory computation
+    - state for storing the data
+    This is not meant to be a black-box abstraction.
+
+    In-memory representations by column type:
+    - StringType: str
+    - IntType: int
+    - FloatType: float
+    - BoolType: bool
+    - TimestampType: datetime.datetime
+    - JsonType: json-serializable object
+    - ArrayType: numpy.ndarray
+    - ImageType: PIL.Image.Image
+    - VideoType: local path if available, otherwise url
+    """
+    def __init__(self, size: int, img_slot_idxs: List[int], media_slot_idxs: List[int], array_slot_idxs: List[int]):
+        self.vals: List[Any] = [None] * size  # either cell values or exceptions
+        self.has_val = [False] * size
+        self.excs: List[Optional[Exception]] = [None] * size
+
+        # control structures that are shared across all DataRows in a batch
+        self.img_slot_idxs = img_slot_idxs
+        self.media_slot_idxs = media_slot_idxs  # all media types aside from image
+        self.array_slot_idxs = array_slot_idxs
+
+        # the primary key of a store row is a sequence of ints (the number is different for table vs view)
+        self.pk: Optional[Tuple[int, ...]] = None
+
+        # file_urls:
+        # - stored url of file for media in vals[i]
+        # - None if vals[i] is not media type
+        # - not None if file_paths[i] is not None
+        self.file_urls: List[Optional[str]] = [None] * size
+
+        # file_paths:
+        # - local path of media file in vals[i]; points to the file cache if file_urls[i] is remote
+        # - None if vals[i] is not a media type or if there is no local file yet for file_urls[i]
+        self.file_paths: List[Optional[str]] = [None] * size
+
+    def clear(self) -> None:
+        size = len(self.vals)
+        self.vals = [None] * size
+        self.has_val = [False] * size
+        self.excs = [None] * size
+        self.pk = None
+        self.file_urls = [None] * size
+        self.file_paths = [None] * size
+
+    def copy(self, target: DataRow) -> None:
+        """Create a copy of the contents of this DataRow in target
+        The copy shares the cell values, but not the control structures (eg, self.has_val), because these
+        need to be independently updateable.
+        """
+        target.vals = self.vals.copy()
+        target.has_val = self.has_val.copy()
+        target.excs = self.excs.copy()
+        target.pk = self.pk
+        target.file_urls = self.file_urls.copy()
+        target.file_paths = self.file_paths.copy()
+
+    def set_pk(self, pk: Tuple[int, ...]) -> None:
+        self.pk = pk
+
+    def has_exc(self, slot_idx: int) -> bool:
+        return self.excs[slot_idx] is not None
+
+    def get_exc(self, slot_idx: int) -> Exception:
+        assert self.has_val[slot_idx] is False
+        assert self.excs[slot_idx] is not None
+        return self.excs[slot_idx]
+
+    def set_exc(self, slot_idx: int, exc: Exception) -> None:
+        assert self.excs[slot_idx] is None
+        self.excs[slot_idx] = exc
+
+        if self.has_val[slot_idx]:
+            # eg. during validation, where contents of file is found invalid
+            self.has_val[slot_idx] = False
+            self.vals[slot_idx] = None
+            self.file_paths[slot_idx] = None
+            self.file_urls[slot_idx] = None
+
+    def __getitem__(self, index: object) -> Any:
+        """Returns in-memory value, ie, what is needed for expr evaluation"""
+        if not self.has_val[index]:
+            # for debugging purposes
+            pass
+        assert self.has_val[index], index
+
+        if self.file_urls[index] is not None and index in self.img_slot_idxs:
+            # if we need to load this from a file, it should have been materialized locally
+            assert self.file_paths[index] is not None
+            if self.vals[index] is None:
+                self.vals[index] = PIL.Image.open(self.file_paths[index])
+
+        return self.vals[index]
+
+    def get_stored_val(self, index: object) -> Any:
+        """Return the value that gets stored in the db"""
+        assert self.excs[index] is None
+        if not self.has_val[index]:
+            # for debugging purposes
+            pass
+        assert self.has_val[index]
+
+        if self.file_urls[index] is not None and (index in self.img_slot_idxs or index in self.media_slot_idxs):
+            # if this is an image or other media type we want to store, we should have a url
+            return self.file_urls[index]
+
+        if self.vals[index] is not None and index in self.array_slot_idxs:
+            assert isinstance(self.vals[index], np.ndarray)
+            np_array = self.vals[index]
+            buffer = io.BytesIO()
+            np.save(buffer, np_array)
+            return buffer.getvalue()
+
+        return self.vals[index]
+
+    def __setitem__(self, idx: object, val: Any) -> None:
+        """Assign in-memory cell value
+        This allows overwriting
+        """
+        assert self.excs[idx] is None
+
+        if (idx in self.img_slot_idxs or idx in self.media_slot_idxs) and isinstance(val, str):
+            # this is either a local file path or a URL
+            parsed = urllib.parse.urlparse(val)
+            if parsed.scheme == '' or parsed.scheme == 'file':
+                # local file path
+                assert self.file_urls[idx] is None and self.file_paths[idx] is None
+                if parsed.scheme == '':
+                    self.file_urls[idx] = urllib.parse.urljoin('file:', urllib.request.pathname2url(parsed.path))
+                else:
+                    self.file_urls[idx] = val
+                self.file_paths[idx] = urllib.parse.unquote(parsed.path)
+            else:
+                # URL
+                assert self.file_urls[idx] is None
+                self.file_urls[idx] = val
+
+            if idx in self.media_slot_idxs:
+                self.vals[idx] = self.file_paths[idx] if self.file_paths[idx] is not None else self.file_urls[idx]
+        elif idx in self.array_slot_idxs and isinstance(val, bytes):
+            self.vals[idx] = np.load(io.BytesIO(val))
+        else:
+            self.vals[idx] = val
+        self.has_val[idx] = True
+
+    def set_file_path(self, idx: object, path: str) -> None:
+        """Augment an existing url with a local file path"""
+        assert self.has_val[idx]
+        assert idx in self.img_slot_idxs or idx in self.media_slot_idxs
+        self.file_paths[idx] = path
+        if idx in self.media_slot_idxs:
+            self.vals[idx] = path
+
+    def flush_img(self, index: object, filepath: Optional[str] = None) -> None:
+        """Discard the in-memory value and save it to a local file, if filepath is not None"""
+        if self.vals[index] is None:
+            return
+        assert self.excs[index] is None
+        if self.file_paths[index] is None:
+            if filepath is not None:
+                # we want to save this to a file
+                self.file_paths[index] = filepath
+                self.file_urls[index] = urllib.parse.urljoin('file:', urllib.request.pathname2url(filepath))
+                self.vals[index].save(filepath, format='JPEG')
+            else:
+                # we discard the content of this cell
+                self.has_val[index] = False
+        else:
+            # we already have a file for this image, nothing left to do
+            pass
+        self.vals[index] = None
+