pixeltable 0.2.18__py3-none-any.whl → 0.2.20__py3-none-any.whl

pixeltable/functions/vision.py

@@ -19,12 +19,9 @@ from typing import Any, Optional, Union
 import numpy as np
 import PIL.Image
 
-import pixeltable.func as func
-import pixeltable.type_system as ts
+import pixeltable as pxt
 from pixeltable.utils.code import local_public_names
 
-# TODO: figure out a better submodule structure
-
 
 # the following function has been adapted from MMEval
 # (sources at https://github.com/open-mmlab/mmeval)
@@ -161,25 +158,41 @@ def __calculate_image_tpfp(
     return tp, fp
 
 
-@func.udf(
-    return_type=ts.JsonType(nullable=False),
-    param_types=[
-        ts.JsonType(nullable=False),
-        ts.JsonType(nullable=False),
-        ts.JsonType(nullable=False),
-        ts.JsonType(nullable=False),
-        ts.JsonType(nullable=False),
-    ],
-)
+@pxt.udf
 def eval_detections(
     pred_bboxes: list[list[int]],
     pred_labels: list[int],
     pred_scores: list[float],
     gt_bboxes: list[list[int]],
     gt_labels: list[int],
+    min_iou: float = 0.5,
 ) -> list[dict]:
     """
     Evaluates the performance of a set of predicted bounding boxes against a set of ground truth bounding boxes.
+
+    Args:
+        pred_bboxes: List of predicted bounding boxes, each represented as [xmin, ymin, xmax, ymax].
+        pred_labels: List of predicted labels.
+        pred_scores: List of predicted scores.
+        gt_bboxes: List of ground truth bounding boxes, each represented as [xmin, ymin, xmax, ymax].
+        gt_labels: List of ground truth labels.
+        min_iou: Minimum intersection-over-union (IoU) threshold for a predicted bounding box to be
+            considered a true positive.
+
+    Returns:
+        A list of dictionaries, one per label class, with the following structure:
+        ```python
+        {
+            'min_iou': float,  # The value of `min_iou` used for the detections
+            'class': int,  # The label class
+            'tp': list[int],  # List of 1's and 0's indicating true positives for each
+                              # predicted bounding box of this class
+            'fp': list[int],  # List of 1's and 0's indicating false positives for each
+                              # predicted bounding box of this class; `fp[n] == 1 - tp[n]`
+            'scores': list[float],  # List of predicted scores for each bounding box of this class
+            'num_gts': int,  # Number of ground truth bounding boxes of this class
+        }
+        ```
     """
     class_idxs = list(set(pred_labels + gt_labels))
     result: list[dict] = []
@@ -192,11 +205,11 @@ def eval_detections(
         pred_filter = pred_classes_arr == class_idx
         gt_filter = gt_classes_arr == class_idx
         class_pred_scores = pred_scores_arr[pred_filter]
-        tp, fp = __calculate_image_tpfp(pred_bboxes_arr[pred_filter], class_pred_scores, gt_bboxes_arr[gt_filter], 0.5)
+        tp, fp = __calculate_image_tpfp(pred_bboxes_arr[pred_filter], class_pred_scores, gt_bboxes_arr[gt_filter], min_iou)
         ordered_class_pred_scores = -np.sort(-class_pred_scores)
         result.append(
             {
-                'min_iou': 0.5,
+                'min_iou': min_iou,
                 'class': class_idx,
                 'tp': tp.tolist(),
                 'fp': fp.tolist(),
@@ -207,11 +220,20 @@ def eval_detections(
     return result
 
 
-@func.uda(update_types=[ts.JsonType()], value_type=ts.JsonType(), allows_std_agg=True, allows_window=False)
-class mean_ap(func.Aggregator):
+@pxt.uda(update_types=[pxt.JsonType()], value_type=pxt.JsonType(), allows_std_agg=True, allows_window=False)
+class mean_ap(pxt.Aggregator):
     """
     Calculates the mean average precision (mAP) over
     [`eval_detections()`][pixeltable.functions.vision.eval_detections] results.
+
+    __Parameters:__
+
+    - `eval_dicts` (list[dict]): List of dictionaries as returned by
+        [`eval_detections()`][pixeltable.functions.vision.eval_detections].
+
+    __Returns:__
+
+    - A `dict[int, float]` mapping each label class to an average precision (AP) value for that class.
     """
     def __init__(self):
         self.class_tpfp: dict[int, list[dict]] = defaultdict(list)
@@ -246,7 +268,7 @@ class mean_ap(func.Aggregator):
         return result
 
 
-def _create_label_colors(labels: list[Any]) -> dict[Any, str]:
+def __create_label_colors(labels: list[Any]) -> dict[Any, str]:
     """
     Create random colors for labels such that a particular label always gets the same color.
 
@@ -265,7 +287,7 @@ def _create_label_colors(labels: list[Any]) -> dict[Any, str]:
     return result
 
 
-@func.udf
+@pxt.udf
 def draw_bounding_boxes(
         img: PIL.Image.Image,
         boxes: list[list[int]],
@@ -324,7 +346,7 @@ def draw_bounding_boxes(
         if color is not None:
             box_colors = [color] * num_boxes
         else:
-            label_colors = _create_label_colors(labels)
+            label_colors = __create_label_colors(labels)
             box_colors = [label_colors[label] for label in labels]
 
     from PIL import ImageColor, ImageDraw, ImageFont

pixeltable/functions/whisper.py

@@ -9,6 +9,7 @@ first `pip install openai-whisper`.
 from typing import TYPE_CHECKING, Optional
 
 import pixeltable as pxt
+from pixeltable.env import Env
 
 if TYPE_CHECKING:
     from whisper import Whisper  # type: ignore[import-untyped]
@@ -71,6 +72,8 @@ def transcribe(
 
         >>> tbl['result'] = transcribe(tbl.audio, model='base.en')
     """
+    Env.get().require_package('whisper')
+    Env.get().require_package('torch')
     import torch
 
     if decode_options is None:

pixeltable/globals.py

@@ -1,6 +1,6 @@
 import dataclasses
 import logging
-from typing import Any, Optional, Union
+from typing import Any, Iterable, Optional, Union
 from uuid import UUID
 
 import pandas as pd
@@ -16,6 +16,7 @@ from pixeltable.dataframe import DataFrameResultSet
 from pixeltable.env import Env
 from pixeltable.iterators import ComponentIterator
 from pixeltable.metadata import schema
+from pixeltable.utils.filecache import FileCache
 
 _logger = logging.getLogger('pixeltable')
 
@@ -193,6 +194,7 @@ def create_view(
     )
     Catalog.get().paths[path] = view
     _logger.info(f'Created view `{path_str}`.')
+    FileCache.get().emit_eviction_warnings()
     return view
 
 
@@ -487,3 +489,7 @@ def configure_logging(
         remove: comma-separated list of module names
     """
     return Env.get().configure_logging(to_stdout=to_stdout, level=level, add=add, remove=remove)
+
+
+def array(elements: Iterable) -> exprs.Expr:
+    return exprs.InlineArray(elements)

pixeltable/io/globals.py

@@ -43,7 +43,7 @@ def create_label_studio_project(
     The API key and URL for a valid Label Studio server must be specified in Pixeltable config. Either:
 
     * Set the `LABEL_STUDIO_API_KEY` and `LABEL_STUDIO_URL` environment variables; or
-    * Specify `api_key` and `url` fields in the `label-studio` section of `$PIXELTABLE_HOME/config.yaml`.
+    * Specify `api_key` and `url` fields in the `label-studio` section of `$PIXELTABLE_HOME/config.toml`.
 
     __Requirements:__
 

pixeltable/io/hf_datasets.py

@@ -34,9 +34,7 @@ _hf_to_pxt: dict[str, ts.ColumnType] = {
 }
 
 
-def _to_pixeltable_type(
-    feature_type: Union[datasets.ClassLabel, datasets.Value, datasets.Sequence],
-) -> Optional[ts.ColumnType]:
+def _to_pixeltable_type(feature_type: Any) -> Optional[ts.ColumnType]:
     """Convert a huggingface feature type to a pixeltable ColumnType if one is defined."""
     import datasets
 
@@ -51,6 +49,8 @@ def _to_pixeltable_type(
         dtype = _to_pixeltable_type(feature_type.feature)
         length = feature_type.length if feature_type.length != -1 else None
         return ts.ArrayType(shape=(length,), dtype=dtype)
+    elif isinstance(feature_type, datasets.Image):
+        return ts.ImageType(nullable=True)
     else:
         return None
 

pixeltable/iterators/document.py

@@ -166,7 +166,7 @@ class DocumentSplitter(ComponentIterator):
         return {
             'document': DocumentType(nullable=False),
             'separators': StringType(nullable=False),
-            'metadata': StringType(nullable=True),
+            'metadata': StringType(nullable=False),
             'limit': IntType(nullable=True),
             'overlap': IntType(nullable=True),
             'skip_tags': StringType(nullable=True),

pixeltable/metadata/__init__.py

@@ -10,7 +10,7 @@ import sqlalchemy.orm as orm
 from .schema import SystemInfo, SystemInfoMd
 
 # current version of the metadata; this is incremented whenever the metadata schema changes
-VERSION = 20
+VERSION = 21
 
 
 def create_system_info(engine: sql.engine.Engine) -> None:

pixeltable/metadata/converters/convert_18.py

@@ -13,7 +13,7 @@ def _(engine: sql.engine.Engine) -> None:
     )
 
 
-def __substitute_md(k: Any, v: Any) -> Optional[tuple[Any, Any]]:
+def __substitute_md(k: Optional[str], v: Any) -> Optional[tuple[Optional[str], Any]]:
     # Migrate a few changed function names
     if k == 'path' and v == 'pixeltable.functions.string.str_format':
         return 'path', 'pixeltable.functions.string.format'

pixeltable/metadata/converters/convert_20.py

@@ -0,0 +1,56 @@
+from typing import Any, Optional
+
+import sqlalchemy as sql
+
+from pixeltable.metadata import register_converter
+from pixeltable.metadata.converters.util import convert_table_md
+
+
+@register_converter(version=20)
+def _(engine: sql.engine.Engine) -> None:
+    convert_table_md(
+        engine,
+        substitution_fn=__substitute_md
+    )
+
+
+def __substitute_md(k: Optional[str], v: Any) -> Optional[tuple[Optional[str], Any]]:
+    if isinstance(v, dict) and '_classname' in v:
+        # The way InlineArray is represented changed in v20. Previously, literal values were stored
+        # directly in the Inline expr; now we store them in Literal sub-exprs. This converter
+        # constructs new Literal exprs for the literal values in InlineArray, interleaving them
+        # with non-literal exprs into the correct sequence.
+        if v['_classname'] == 'InlineArray':
+            components = v.get('components')  # Might be None, but that's ok
+            updated_components = []
+            for idx, val in v['elements']:
+                # idx >= 0, then this is a non-literal sub-expr. Otherwise, idx could be either
+                # None or -1, for legacy reasons (which are now obviated).
+                if idx is not None and idx >= 0:
+                    updated_components.append(components[idx])
+                else:
+                    updated_components.append({'val': val, '_classname': 'Literal'})
+            # InlineList was split out from InlineArray in v20. If is_json=True, then this is
+            # actually an InlineList. If is_json=False, then we assume it's an InlineArray for now,
+            # but it might actually be transformed into an InlineList when it is instantiated
+            # (unfortunately, there is no way to disambiguate at this stage; see comments in
+            # InlineArray._from_dict() for more details).
+            updated_v = {'_classname': 'InlineList' if v.get('is_json') else 'InlineArray'}
+            if len(updated_components) > 0:
+                updated_v['components'] = updated_components
+            return k, updated_v
+        if v['_classname'] == 'InlineDict':
+            components = v.get('components')
+            keys = []
+            updated_components = []
+            for key, idx, val in v['dict_items']:
+                keys.append(key)
+                if idx is not None and idx >= 0:
+                    updated_components.append(components[idx])
+                else:
+                    updated_components.append({'val': val, '_classname': 'Literal'})
+            updated_v = {'keys': keys, '_classname': 'InlineDict'}
+            if len(updated_components) > 0:
+                updated_v['components'] = updated_components
+            return k, updated_v
+    return None

pixeltable/metadata/converters/util.py

@@ -14,8 +14,22 @@ def convert_table_md(
     table_md_updater: Optional[Callable[[dict], None]] = None,
     column_md_updater: Optional[Callable[[dict], None]] = None,
     external_store_md_updater: Optional[Callable[[dict], None]] = None,
-    substitution_fn: Optional[Callable[[Any, Any], Optional[tuple[Any, Any]]]] = None
+    substitution_fn: Optional[Callable[[Optional[str], Any], Optional[tuple[Optional[str], Any]]]] = None
 ) -> None:
+    """
+    Converts table metadata based on the specified conversion functions.
+
+    Args:
+        engine: The SQLAlchemy engine.
+        table_md_updater: A function that updates the table metadata in place.
+        column_md_updater: A function that updates the column metadata in place.
+        external_store_md_updater: A function that updates the external store metadata in place.
+        substitution_fn: A function that substitutes metadata values. If specified, all metadata will be traversed
+            recursively, and `substitution_fn` will be called once for each metadata entry. If the entry appears in
+            a dict as a `(k, v)` pair, then `substitution_fn(k, v)` will be called. If the entry appears in a list,
+            then `substitution_fn(None, v)` will be called. If `substitution_fn` returns a tuple `(k', v')`, then
+            the original entry will be replaced, and the traversal will continue with `v'`.
+    """
     with engine.begin() as conn:
         for row in conn.execute(sql.select(Table)):
             id = row[0]
@@ -49,18 +63,29 @@ def __update_external_store_md(table_md: dict, external_store_md_updater: Callab
         external_store_md_updater(store_md)
 
 
-def __substitute_md_rec(md: Any, substitution_fn: Callable[[Any, Any], Optional[tuple[Any, Any]]]) -> Any:
+def __substitute_md_rec(
+    md: Any,
+    substitution_fn: Callable[[Optional[str], Any], Optional[tuple[Optional[str], Any]]]
+) -> Any:
     if isinstance(md, dict):
         updated_md = {}
         for k, v in md.items():
             substitute = substitution_fn(k, v)
             if substitute is not None:
                 updated_k, updated_v = substitute
-                updated_md[updated_k] = updated_v
+                updated_md[updated_k] = __substitute_md_rec(updated_v, substitution_fn)
             else:
                 updated_md[k] = __substitute_md_rec(v, substitution_fn)
         return updated_md
     elif isinstance(md, list):
-        return [__substitute_md_rec(v, substitution_fn) for v in md]
+        updated_md = []
+        for v in md:
+            substitute = substitution_fn(None, v)
+            if substitute is not None:
+                _, updated_v = substitute
+                updated_md.append(__substitute_md_rec(updated_v, substitution_fn))
+            else:
+                updated_md.append(__substitute_md_rec(v, substitution_fn))
+        return updated_md
     else:
         return md

pixeltable/metadata/notes.py

@@ -2,6 +2,7 @@
 # rather than as a comment, so that the existence of a description can be enforced by
 # the unit tests when new versions are added.
 VERSION_NOTES = {
+    21: 'Separate InlineArray and InlineList',
     20: 'Store DB timestamps in UTC',
     19: 'UDF renames; ImageMemberAccess removal',
     18: 'Restructured index metadata',

pixeltable/tool/create_test_db_dump.py

@@ -4,6 +4,7 @@ import logging
 import os
 import pathlib
 import subprocess
+import sys
 from typing import Any
 from zoneinfo import ZoneInfo
 
@@ -24,12 +25,18 @@ _logger = logging.getLogger('pixeltable')
 class Dumper:
 
     def __init__(self, output_dir='target', db_name='pxtdump') -> None:
+        if sys.version_info >= (3, 10):
+            raise RuntimeError(
+                'This script must be run on Python 3.9. '
+                'DB dumps are incompatible across versions due to issues with pickling anonymous UDFs.'
+            )
+
         self.output_dir = pathlib.Path(output_dir)
         shared_home = pathlib.Path(os.environ.get('PIXELTABLE_HOME', '~/.pixeltable')).expanduser()
         mock_home_dir = self.output_dir / '.pixeltable'
         mock_home_dir.mkdir(parents=True, exist_ok=True)
         os.environ['PIXELTABLE_HOME'] = str(mock_home_dir)
-        os.environ['PIXELTABLE_CONFIG'] = str(shared_home / 'config.yaml')
+        os.environ['PIXELTABLE_CONFIG'] = str(shared_home / 'config.toml')
         os.environ['PIXELTABLE_DB'] = db_name
         os.environ['PIXELTABLE_PGDATA'] = str(shared_home / 'pgdata')
 
@@ -226,9 +233,13 @@ class Dumper:
         add_column('isin_2', t.c2.isin([1, 2, 3, 4, 5]))
         add_column('isin_3', t.c2.isin(t.c6.f5))
 
-        # inline_array and inline_dict
-        add_column('inline_array_1', [[1, 2, 3], [4, 5, 6]])
-        add_column('inline_array_2', [['a', 'b', 'c'], ['d', 'e', 'f']])
+        # inline_array, inline_list, inline_dict
+        add_column('inline_array_1', pxt.array([[1, 2, 3], [4, 5, 6]]))
+        add_column('inline_array_2', pxt.array([['a', 'b', 'c'], ['d', 'e', 'f']]))
+        add_column('inline_array_exprs', pxt.array([[t.c2, t.c2 + 1], [t.c2 + 2, t.c2]]))
+        add_column('inline_array_mixed', pxt.array([[1, t.c2], [3, t.c2]]))
+        add_column('inline_list_1', [[1, 2, 3], [4, 5, 6]])
+        add_column('inline_list_2', [['a', 'b', 'c'], ['d', 'e', 'f']])
         add_column('inline_list_exprs', [t.c1, [t.c1n, t.c2]])
         add_column('inline_list_mixed', [1, 'a', t.c1, [1, 'a', t.c1n], 1, 'a'])
         add_column('inline_dict', {'int': 22, 'dict': {'key': 'val'}, 'expr': t.c1})

pixeltable/type_system.py

@@ -204,6 +204,8 @@ class ColumnType:
 
     @classmethod
     def infer_literal_type(cls, val: Any, nullable: bool = False) -> Optional[ColumnType]:
+        if val is None:
+            return InvalidType(nullable=True)
         if isinstance(val, str):
             return StringType(nullable=nullable)
         if isinstance(val, bool):
@@ -395,7 +397,7 @@ class InvalidType(ColumnType):
         assert False
 
     def print_value(self, val: Any) -> str:
-        assert False
+        return str(val)
 
     def _validate_literal(self, val: Any) -> None:
         assert False