feat: Bump version v0.4.2

CHANGELOG.md

@@ -1,3 +1,18 @@
+## v0.4.2
+
+Bugfixes before the upcoming release.
+- Better imports and lazy loading
+- Default device is now 'auto', which resolves to cuda/cpu depending on availability
+- Rich text now available as `AutoPageFormatter`
+    - Fixed bug with permuted coordinates (e0c6dc52)
+- CroppedTable now directly has `angle` property
+- CI tests, Python 3.9 support
+- More type hints
+- Light restructuring (non-breaking)
+- Internal data structure tweaks
+    - (`fctn_results` → `predictions.tatr`)
+    - (`effective_*` → `predictions.effective`)
+
 ## v0.4.0
 
 Features: 3 new table structure recognition options!

gmft/__init__.py

@@ -1,140 +1,133 @@
 """
 Currently, contains aliases for key classes and functions.
 
-Unfortunately, although at one point the ability to import classes from the top level module (ie. `from gmft import AutoTableFormatter`) was encouraged,
-it is now discouraged and may be removed in future versions. The reason being: importing through the top level module
-loads the entire library, even when you're using only a small part of it.
+Importing from the top-level module previously resulted in long load times.
+However, v0.5 introduces lazy loading, which greatly improves the situation.
 
-Instead, `gmft.auto` is now encouraged. For example, `from gmft.auto import AutoTableFormatter`.
+Now, classes may either be imported from their original locations,
+`gmft.auto`, or from here, where they will be lazy loaded.
 """
 
+# small classes are fine, but discouraged.
 from gmft.base import Rect
+from gmft.core.legacy.mirror import DeprecationMirrorMeta
 from gmft.pdf_bindings.base import BasePDFDocument, BasePage
 from gmft.detectors.base import CroppedTable, RotatedCroppedTable
 from gmft.formatters.base import FormattedTable
 
-from gmft.auto import (
-    TATRDetector as TATRTableDetectorOrig,
-    TableDetectorConfig as TableDetectorConfigOrig,
-    TableDetector as TableDetectorOrig,
-    TATRFormatConfig as TATRFormatConfigOrig,
-    TATRFormattedTable as TATRFormattedTableOrig,
-    TATRFormatter as TATRTableFormatterOrig,
-    AutoTableFormatter as AutoTableFormatterOrig,
-    AutoFormatConfig as AutoFormatConfigOrig,
-    AutoTableDetector as AutoTableDetectorOrig,
+# config-only classes specific to TATR are still discouraged.
+
+# these auto classes are lazy-loaded
+from gmft.core.auto_lazy import (
+    AutoTableFormatter,
+    AutoFormatConfig,
+    AutoTableDetector,
 )
 
-has_warned = False
+# We need to support these imports for compatibility:
+# TATRTableDetector
+# TableDetectorConfig
+# TableDetector
+# TATRFormatConfig
+# TATRFormattedTable
+# TATRTableFormatter
+# AutoTableFormatter
+# AutoFormatConfig
+# AutoTableDetector
 
 
-def _deprecation_warning(name):
-    global has_warned
-    if has_warned:
-        return
-    import warnings
+# These bulky TATR-specific detectors are discouraged, but still available for compatibility.
+class TATRTableDetector(metaclass=DeprecationMirrorMeta):
+    """
+    This import is deprecated.
 
-    msg = f"(Deprecation) While once encouraged, \
-importing {name} and other classes from the top level module is now deprecated and will break in v0.5.0. \
-Please import from gmft.auto instead."
-    warnings.warn(msg, DeprecationWarning, stacklevel=2)
-    print(msg)
-    has_warned = True
+    Please use:
+    - gmft.AutoTableDetector
+    - gmft.detectors.tatr.TATRDetector
+    """
 
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.detectors.tatr import TATRDetector as OrigCls
 
-# These small classes are fine, but still discouraged.
-# Rect
-# BasePDFDocument
-# BasePage
-# CroppedTable
-# RotatedCroppedTable
+        return OrigCls
 
 
-class TATRTableDetector(TATRTableDetectorOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
+class TableDetectorConfig(metaclass=DeprecationMirrorMeta):
     """
+    This import is deprecated.
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TATRTableDetector")
-        super().__init__(*args, **kwargs)
-
-
-class TableDetectorConfig(TableDetectorConfigOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
+    Please use:
+    - Reformat API (v0.5)
+    - gmft.detectors.tatr.TATRDetectorConfig
     """
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TableDetectorConfig")
-        super().__init__(*args, **kwargs)
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.impl.tatr.config import TATRDetectorConfig as OrigCls
 
+        return OrigCls
 
-class TableDetector(TableDetectorOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
-    """
-
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TableDetector")
-        super().__init__(*args, **kwargs)
 
-
-class TATRFormatConfig(TATRFormatConfigOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
+class TableDetector(metaclass=DeprecationMirrorMeta):
     """
+    This import is deprecated.
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TATRFormatConfig")
-        super().__init__(*args, **kwargs)
-
-
-class TATRFormattedTable(TATRFormattedTableOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
+    Please use:
+    - gmft.AutoTableDetector
+    - gmft.detectors.tatr.TATRDetector
     """
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TATRFormattedTable")
-        super().__init__(*args, **kwargs)
-
-
-class TATRTableFormatter(TATRTableFormatterOrig):
-    """
-    Deprecated. Please import from gmft.auto instead.
-    """
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.auto import TATRDetector as OrigCls
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("TATRTableFormatter")
-        super().__init__(*args, **kwargs)
+        return OrigCls
 
 
-class AutoTableFormatter(AutoTableFormatterOrig):
+class TATRFormatConfig(metaclass=DeprecationMirrorMeta):
     """
-    Deprecated. Please import from gmft.auto instead.
+    This import is deprecated.
+
+    Please use:
+    - Reformat API (v0.5)
+    - gmft.formatters.tatr.TATRFormatConfig
     """
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("AutoTableFormatter")
-        super().__init__(*args, **kwargs)
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.impl.tatr.config import TATRFormatConfig as OrigCls
+
+        return OrigCls
 
 
-class AutoFormatConfig(AutoFormatConfigOrig):
+class TATRFormattedTable(metaclass=DeprecationMirrorMeta):
     """
-    Deprecated. Please import from gmft.auto instead.
+    This import is deprecated.
+
+    Please use:
+    - Reformat API (v0.5)
+    - gmft.formatters.tatr.TATRFormattedTable
     """
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("AutoFormatConfig")
-        super().__init__(*args, **kwargs)
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.formatters.tatr import TATRFormattedTable as OrigCls
 
+        return OrigCls
 
-class AutoTableDetector(AutoTableDetectorOrig):
+
+class TATRTableFormatter(metaclass=DeprecationMirrorMeta):
     """
-    Deprecated. Please import from gmft.auto instead.
+    This import is deprecated.
+
+    Please use:
+    - gmft.auto.AutoTableFormatter
+    - gmft.formatters.tatr.TATRFormatter
     """
 
-    def __init__(self, *args, **kwargs):
-        _deprecation_warning("AutoTableDetector")
-        super().__init__(*args, **kwargs)
+    @classmethod
+    def get_mirrored_class(cls):
+        from gmft.formatters.tatr import TATRFormatter as OrigCls
+
+        return OrigCls

gmft/algorithm/structure.py

@@ -767,7 +767,7 @@ def extract_to_df(table: TATRFormattedTable, config: TATRFormatConfig = None):
 
     outliers = {}  # store table-wide information about outliers or pecularities
 
-    results = table.predictions["tatr"]
+    results = table.predictions.tatr
 
     # 1. collate identified boxes
     boxes = []
@@ -889,8 +889,9 @@ def extract_to_df(table: TATRFormattedTable, config: TATRFormatConfig = None):
         if not known_means:
             # no text was detected
             outliers["no text"] = True
-            table.predictions["effective"] = _empty_effective_predictions()
-            table.predictions["indices"] = _empty_indices_predictions()
+            table.predictions.effective = _empty_effective_predictions()
+            table.predictions.indices = _empty_indices_predictions()
+            table.predictions.status = "ready"
             table._df = pd.DataFrame()
             table.outliers = outliers
             return table._df
@@ -930,7 +931,7 @@ def extract_to_df(table: TATRFormattedTable, config: TATRFormatConfig = None):
         )
 
     # nms takes care of deduplication
-    table.predictions["effective"] = {
+    table.predictions.effective = {
         "rows": sorted_rows,
         "columns": sorted_columns,
         "headers": sorted_headers,
@@ -1071,7 +1072,8 @@ def extract_to_df(table: TATRFormattedTable, config: TATRFormatConfig = None):
         ]
         indices_preds["_projecting"] = [i for i, x in enumerate(is_projecting) if x]
 
-    table.predictions["indices"] = indices_preds
+    table.predictions.indices = indices_preds
+    table.predictions.status = "ready"
 
     # if projecting_indices:
     # insert at end

gmft/auto.py

@@ -22,41 +22,8 @@
 TATRTableFormatter = TATRFormatter
 # TATRFormatConfig = TATRFormatConfig
 
-
-class AutoTableFormatter:
-    """
-    The recommended :class:`~gmft.formatters.base.BaseFormatter`. Currently points to :class:`~gmft.formatters.tatr.TATRFormatter`.
-    Uses a TableTransformerForObjectDetection for small/medium tables, and a custom algorithm for large tables.
-
-    Using :meth:`extract`, a :class:`~gmft.formatters.base.FormattedTable` is produced, which can be exported to csv, df, etc.
-    """
-
-    def __new__(cls, *args, **kwargs):
-        from gmft.formatters.tatr import TATRFormatter
-
-        return TATRFormatter(*args, **kwargs)
-
-
-class AutoFormatConfig:
-    """
-    Configuration for the recommended :class:`~gmft.formatters.base.BaseFormatter`. Currently points to :class:`~gmft.formatters.tatr.TATRFormatConfig`.
-    """
-
-    def __new__(cls, *args, **kwargs):
-        from gmft.impl.tatr.config import TATRFormatConfig
-
-        return TATRFormatConfig(*args, **kwargs)
-
-
-class AutoTableDetector:
-    """
-    The recommended :class:`~gmft.detectors.base.BaseDetector`. Currently points to :class:`~gmft.detectors.tatr.TATRDetector`.
-    Uses TableTransformerForObjectDetection for small/medium tables, and a custom algorithm for large tables.
-
-    Using :meth:`~gmft.detectors.base.BaseDetector.extract` produces a :class:`~gmft.formatters.base.FormattedTable`, which can be exported to csv, df, etc.
-    """
-
-    def __new__(cls, *args, **kwargs):
-        from gmft.detectors.tatr import TATRDetector
-
-        return TATRDetector(*args, **kwargs)
+from gmft.core.auto_lazy import (
+    AutoTableFormatter,
+    AutoFormatConfig,
+    AutoTableDetector,
+)

gmft/core/auto_lazy.py

@@ -0,0 +1,37 @@
+class AutoTableFormatter:
+    """
+    The recommended :class:`~gmft.formatters.base.BaseFormatter`. Currently points to :class:`~gmft.formatters.tatr.TATRFormatter`.
+    Uses a TableTransformerForObjectDetection for small/medium tables, and a custom algorithm for large tables.
+
+    Using :meth:`extract`, a :class:`~gmft.formatters.base.FormattedTable` is produced, which can be exported to csv, df, etc.
+    """
+
+    def __new__(cls, *args, **kwargs):
+        from gmft.formatters.tatr import TATRFormatter
+
+        return TATRFormatter(*args, **kwargs)
+
+
+class AutoFormatConfig:
+    """
+    Configuration for the recommended :class:`~gmft.formatters.base.BaseFormatter`. Currently points to :class:`~gmft.formatters.tatr.TATRFormatConfig`.
+    """
+
+    def __new__(cls, *args, **kwargs):
+        from gmft.impl.tatr.config import TATRFormatConfig
+
+        return TATRFormatConfig(*args, **kwargs)
+
+
+class AutoTableDetector:
+    """
+    The recommended :class:`~gmft.detectors.base.BaseDetector`. Currently points to :class:`~gmft.detectors.tatr.TATRDetector`.
+    Uses TableTransformerForObjectDetection for small/medium tables, and a custom algorithm for large tables.
+
+    Using :meth:`~gmft.detectors.base.BaseDetector.extract` produces a :class:`~gmft.formatters.base.FormattedTable`, which can be exported to csv, df, etc.
+    """
+
+    def __new__(cls, *args, **kwargs):
+        from gmft.detectors.tatr import TATRDetector
+
+        return TATRDetector(*args, **kwargs)

gmft/core/io/serial/dicts.py

@@ -3,6 +3,7 @@
 from gmft.core.ml.prediction import (
     IndicesPredictions,
     RawBboxPredictions,
+    _empty_effective_predictions,
     _empty_indices_predictions,
 )
 from gmft.detectors.base import CroppedTable
@@ -57,3 +58,11 @@ def _extract_indices(d: dict) -> IndicesPredictions:
         }
 
     return _empty_indices_predictions()
+
+
+def _extract_effective(d: dict) -> IndicesPredictions:
+    # version gmft>=0.5 format
+    if "predictions.effective" in d:
+        return d["predictions.effective"]
+
+    return _empty_effective_predictions()