jerry-thomas 1.0.3__py3-none-any.whl → 2.0.1__py3-none-any.whl

datapipeline/pipeline/pipelines.py

@@ -1,9 +1,9 @@
 import heapq
+from collections import defaultdict
 from collections.abc import Iterator, Sequence
 from typing import Any
 from itertools import tee
 
-from datapipeline.domain.sample import Sample
 from datapipeline.domain.vector import Vector
 from datapipeline.pipeline.utils.keygen import group_key_for
 from datapipeline.pipeline.utils.memory_sort import batch_sort
@@ -12,8 +12,9 @@ from datapipeline.pipeline.stages import (
     open_source_stream,
     build_record_stream,
     apply_record_operations,
+    order_record_stream,
     build_feature_stream,
-    regularize_feature_stream,
+    apply_stream_operations,
     apply_feature_transforms,
     vector_assemble_stage,
     sample_assemble_stage,
@@ -21,15 +22,61 @@ from datapipeline.pipeline.stages import (
     window_keys,
 )
 from datapipeline.pipeline.context import PipelineContext
+from datapipeline.pipeline.utils.spool_cache import SpoolCache
 
 
-def build_feature_pipeline(
+def _time_then_id(item: Any):
+    rec = getattr(item, "record", None)
+    if rec is not None:
+        t = getattr(rec, "time", None)
+    else:
+        recs = getattr(item, "records", None)
+        t = getattr(recs[0], "time", None) if recs else None
+    return (t, getattr(item, "id", None))
+
+
+def _build_feature_from_records(
     context: PipelineContext,
+    records: Iterator[Any],
     cfg: FeatureRecordConfig,
     stage: int | None = None,
+    batch_size: int | None = None,
+    partition_by: str | None = None,
 ) -> Iterator[Any]:
     runtime = context.runtime
-    record_stream_id = cfg.record_stream
+
+    if partition_by is None:
+        partition_by = runtime.registries.partition_by.get(cfg.record_stream)
+
+    features = build_feature_stream(
+        records,
+        base_feature_id=cfg.id,
+        field=cfg.field,
+        partition_by=partition_by,
+    )
+    if stage == 5:
+        return features
+
+    transformed = apply_feature_transforms(
+        context, features, cfg.scale, cfg.sequence)
+    if stage == 6:
+        return transformed
+
+    if batch_size is None:
+        batch_size = runtime.registries.sort_batch_size.get(cfg.record_stream)
+    sorted_for_grouping = batch_sort(
+        transformed, batch_size=batch_size, key=_time_then_id
+    )
+    return sorted_for_grouping
+
+
+def build_record_pipeline(
+    context: PipelineContext,
+    record_stream_id: str,
+    stage: int | None = None,
+) -> Iterator[Any]:
+    """Build a canonical record stream through stream transforms."""
+    runtime = context.runtime
 
     dtos = open_source_stream(context, record_stream_id)
     if stage == 0:
@@ -43,35 +90,41 @@ def build_feature_pipeline(
     if stage == 2:
         return records
 
-    partition_by = runtime.registries.partition_by.get(record_stream_id)
-    features = build_feature_stream(records, cfg.id, partition_by)
+    batch_size = runtime.registries.sort_batch_size.get(record_stream_id)
+    records = order_record_stream(
+        context, records, record_stream_id, batch_size)
     if stage == 3:
-        return features
+        return records
 
-    batch_size = runtime.registries.sort_batch_size.get(record_stream_id)
-    regularized = regularize_feature_stream(
-        context, features, record_stream_id, batch_size)
+    records = apply_stream_operations(context, records, record_stream_id)
     if stage == 4:
-        return regularized
+        return records
 
-    transformed = apply_feature_transforms(
-        context, regularized, cfg.scale, cfg.sequence)
-    if stage == 5:
-        return transformed
+    return records
 
-    def _time_then_id(item: Any):
-        rec = getattr(item, "record", None)
-        if rec is not None:
-            t = getattr(rec, "time", None)
-        else:
-            recs = getattr(item, "records", None)
-            t = getattr(recs[0], "time", None) if recs else None
-        return (t, getattr(item, "id", None))
 
-    sorted_for_grouping = batch_sort(
-        transformed, batch_size=batch_size, key=_time_then_id
+def build_feature_pipeline(
+    context: PipelineContext,
+    cfg: FeatureRecordConfig,
+    stage: int | None = None,
+) -> Iterator[Any]:
+    runtime = context.runtime
+    record_stream_id = cfg.record_stream
+
+    records = build_record_pipeline(context, record_stream_id, stage=stage)
+    if stage is not None and stage <= 4:
+        return records
+
+    batch_size = runtime.registries.sort_batch_size.get(record_stream_id)
+    partition_by = runtime.registries.partition_by.get(record_stream_id)
+    return _build_feature_from_records(
+        context,
+        records,
+        cfg,
+        stage=stage,
+        batch_size=batch_size,
+        partition_by=partition_by,
     )
-    return sorted_for_grouping
 
 
 def build_vector_pipeline(
@@ -130,14 +183,45 @@ def _assemble_vectors(
 ) -> Iterator[tuple[tuple, Vector]]:
     if not configs:
         return iter(())
-    streams = [
-        build_feature_pipeline(
-            context,
-            cfg,
-        )
-        for cfg in configs
-    ]
+
+    runtime = context.runtime
+    grouped: dict[str, list[FeatureRecordConfig]] = defaultdict(list)
+    for cfg in configs:
+        grouped[cfg.record_stream].append(cfg)
+
+    streams: list[Iterator[Any]] = []
+    caches: list[SpoolCache] = []
+    for record_stream_id, cfgs in grouped.items():
+        records = build_record_pipeline(context, record_stream_id, stage=4)
+        if len(cfgs) == 1:
+            record_iters = (records,)
+        else:
+            cache = SpoolCache(records, name=record_stream_id)
+            caches.append(cache)
+            record_iters = tuple(cache.reader() for _ in cfgs)
+        batch_size = runtime.registries.sort_batch_size.get(record_stream_id)
+        partition_by = runtime.registries.partition_by.get(record_stream_id)
+
+        for cfg, rec_iter in zip(cfgs, record_iters):
+            streams.append(
+                _build_feature_from_records(
+                    context,
+                    rec_iter,
+                    cfg,
+                    batch_size=batch_size,
+                    partition_by=partition_by,
+                )
+            )
+
     merged = heapq.merge(
         *streams, key=lambda fr: group_key_for(fr, group_by_cadence)
     )
-    return vector_assemble_stage(merged, group_by_cadence)
+
+    def _with_cleanup() -> Iterator[tuple[tuple, Vector]]:
+        try:
+            yield from vector_assemble_stage(merged, group_by_cadence)
+        finally:
+            for cache in caches:
+                cache.close()
+
+    return _with_cleanup()

datapipeline/pipeline/stages.py

@@ -20,11 +20,23 @@ from datapipeline.sources.models.source import Source
 from datapipeline.transforms.vector import VectorEnsureSchemaTransform
 from datapipeline.config.dataset.normalize import floor_time_to_bucket
 from datapipeline.utils.time import parse_timecode
+from datapipeline.transforms.utils import get_field, partition_key
 
 
 def open_source_stream(context: PipelineContext, stream_alias: str) -> Source:
     runtime = context.runtime
-    return runtime.registries.stream_sources.get(stream_alias).stream()
+    registry = runtime.registries.stream_sources
+    try:
+        source = registry.get(stream_alias)
+    except KeyError as exc:
+        available = sorted(registry.keys())
+        available_text = ", ".join(available) if available else "(none)"
+        raise KeyError(
+            "Unknown record_stream "
+            f"'{stream_alias}'. Check dataset.yaml and contracts/ ids. "
+            f"Available streams: {available_text}"
+        ) from exc
+    return source.stream()
 
 
 def build_record_stream(
@@ -49,45 +61,66 @@ def apply_record_operations(
     return records
 
 
+def _record_has_field(record: Any, field: str) -> bool:
+    if isinstance(record, dict):
+        return field in record
+    return hasattr(record, field)
+
+
 def build_feature_stream(
     record_stream: Iterable[TemporalRecord],
     base_feature_id: str,
+    field: str,
     partition_by: Any | None = None,
 ) -> Iterator[FeatureRecord]:
-
     keygen = FeatureIdGenerator(partition_by)
 
     for rec in record_stream:
+        if not _record_has_field(rec, field):
+            raise KeyError(
+                f"Record field '{field}' not found on {type(rec).__name__}")
         yield FeatureRecord(
             record=rec,
             id=keygen.generate(base_feature_id, rec),
+            value=get_field(rec, field),
         )
 
 
-def regularize_feature_stream(
+def order_record_stream(
     context: PipelineContext,
-    feature_stream: Iterable[FeatureRecord],
+    record_stream: Iterable[TemporalRecord],
     stream_id: str,
     batch_size: int,
-) -> Iterator[FeatureRecord]:
-    """Apply feature transforms defined in contract policies in order."""
-    # Sort by (id, time) to satisfy stream transforms (ensure_cadence/fill)
-    sorted = batch_sort(
-        feature_stream,
+) -> Iterator[TemporalRecord]:
+    """Return records sorted by (partition_key, time)."""
+    partition_by = context.runtime.registries.partition_by.get(stream_id)
+    return batch_sort(
+        record_stream,
         batch_size=batch_size,
-        key=lambda fr: (fr.id, fr.record.time),
+        key=lambda rec: (partition_key(rec, partition_by), rec.time),
     )
+
+
+def apply_stream_operations(
+    context: PipelineContext,
+    record_stream: Iterable[TemporalRecord],
+    stream_id: str,
+) -> Iterator[TemporalRecord]:
+    """Apply stream/debug transforms (expects input sorted by partition_key + time)."""
+    partition_by = context.runtime.registries.partition_by.get(stream_id)
     transformed = apply_transforms(
-        sorted,
+        record_stream,
         STREAM_TRANFORMS_EP,
         context.runtime.registries.stream_operations.get(stream_id),
         context,
+        extra_kwargs={"partition_by": partition_by},
     )
     transformed = apply_transforms(
         transformed,
         DEBUG_TRANSFORMS_EP,
         context.runtime.registries.debug_operations.get(stream_id),
         context,
+        extra_kwargs={"partition_by": partition_by},
     )
     return transformed
 
@@ -135,10 +168,9 @@ def vector_assemble_stage(
         feature_map = defaultdict(list)
         for fr in group:
             if isinstance(fr, FeatureRecordSequence):
-                records = fr.records
+                feature_map[fr.id].extend(fr.values)
             else:
-                records = [fr.record]
-            feature_map[fr.id].extend(records)
+                feature_map[fr.id].append(fr.value)
         vector = vectorize_record_group(feature_map)
         yield group_key, vector
 
@@ -242,16 +274,19 @@ def _apply_vector_schema(
 
         if not feature_entries:
             if context.schema_required:
-                raise RuntimeError("Schema missing for payload 'features'. Run `jerry build` to materialize schema.json.")
+                raise RuntimeError(
+                    "Schema missing for payload 'features'. Run `jerry build` to materialize schema.json.")
             feature_stream = stream
         else:
-            feature_schema = VectorEnsureSchemaTransform(on_missing="fill", on_extra="drop")
+            feature_schema = VectorEnsureSchemaTransform(
+                on_missing="fill", on_extra="drop")
             feature_schema.bind_context(context)
             feature_stream = feature_schema(stream)
 
         def _apply_targets(upstream: Iterator[Sample]) -> Iterator[Sample]:
             if target_entries:
-                target_schema = VectorEnsureSchemaTransform(payload="targets", on_missing="fill", on_extra="drop")
+                target_schema = VectorEnsureSchemaTransform(
+                    payload="targets", on_missing="fill", on_extra="drop")
                 target_schema.bind_context(context)
                 return target_schema(upstream)
             if not context.schema_required:
@@ -264,6 +299,7 @@ def _apply_vector_schema(
                 return iter(())
             if first.targets is None:
                 return chain([first], iterator)
-            raise RuntimeError("Schema missing for payload 'targets'. Run `jerry build` to materialize schema.json.")
+            raise RuntimeError(
+                "Schema missing for payload 'targets'. Run `jerry build` to materialize schema.json.")
 
         return _apply_targets(feature_stream)

datapipeline/pipeline/utils/spool_cache.py

@@ -0,0 +1,142 @@
+import pickle
+import tempfile
+import weakref
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Iterator, Any
+
+
+_LEN_BYTES = 8
+
+
+def _encode_len(size: int) -> bytes:
+    return int(size).to_bytes(_LEN_BYTES, "little", signed=False)
+
+
+def _decode_len(raw: bytes) -> int:
+    return int.from_bytes(raw, "little", signed=False)
+
+
+@dataclass
+class _SpoolState:
+    writer: Any
+    path: Path
+    offsets: list[int]
+    source: Iterator[Any]
+    done: bool = False
+
+    def close(self) -> None:
+        try:
+            self.writer.close()
+        except Exception:
+            pass
+
+
+class SpoolCache:
+    """Disk-backed cache for iterators with multiple sequential readers."""
+
+    def __init__(self, source: Iterator[Any], *, name: str | None = None) -> None:
+        tmp = tempfile.NamedTemporaryFile(
+            prefix=f"dp-spool-{name or 'stream'}-",
+            suffix=".pkl",
+            delete=False,
+        )
+        path = Path(tmp.name)
+        self._state = _SpoolState(
+            writer=tmp,
+            path=path,
+            offsets=[],
+            source=iter(source),
+        )
+        self._finalizer = weakref.finalize(self, _cleanup, path, tmp)
+
+    @property
+    def path(self) -> Path:
+        return self._state.path
+
+    def close(self) -> None:
+        """Close writer and remove the spool file."""
+        if self._finalizer.alive:
+            self._finalizer()
+
+    def __enter__(self) -> "SpoolCache":
+        return self
+
+    def __exit__(self, exc_type, exc, tb) -> None:
+        self.close()
+
+    def reader(self) -> Iterator[Any]:
+        return _SpoolReader(self)
+
+    def _append_next(self) -> bool:
+        if self._state.done:
+            return False
+        try:
+            item = next(self._state.source)
+        except StopIteration:
+            self._state.done = True
+            self._state.writer.flush()
+            return False
+        try:
+            data = pickle.dumps(item, protocol=pickle.HIGHEST_PROTOCOL)
+        except Exception as exc:  # pragma: no cover - defensive
+            raise TypeError(
+                "SpoolCache requires picklable records for multi-feature fanout."
+            ) from exc
+        offset = self._state.writer.tell()
+        self._state.writer.write(_encode_len(len(data)))
+        self._state.writer.write(data)
+        self._state.writer.flush()
+        self._state.offsets.append(offset)
+        return True
+
+    def _ensure_index(self, index: int) -> None:
+        while len(self._state.offsets) <= index:
+            if not self._append_next():
+                break
+
+
+class _SpoolReader:
+    def __init__(self, cache: SpoolCache) -> None:
+        self._cache = cache
+        self._index = 0
+        self._fh = open(cache.path, "rb")
+
+    def __iter__(self) -> "_SpoolReader":
+        return self
+
+    def __next__(self) -> Any:
+        self._cache._ensure_index(self._index)
+        if self._index >= len(self._cache._state.offsets):
+            self._close()
+            raise StopIteration
+        offset = self._cache._state.offsets[self._index]
+        self._index += 1
+        self._fh.seek(offset)
+        raw = self._fh.read(_LEN_BYTES)
+        if not raw:
+            self._close()
+            raise StopIteration
+        size = _decode_len(raw)
+        payload = self._fh.read(size)
+        return pickle.loads(payload)
+
+    def _close(self) -> None:
+        try:
+            self._fh.close()
+        except Exception:
+            pass
+
+    def __del__(self) -> None:
+        self._close()
+
+
+def _cleanup(path: Path, writer: Any) -> None:
+    try:
+        writer.close()
+    except Exception:
+        pass
+    try:
+        path.unlink(missing_ok=True)
+    except Exception:
+        pass

datapipeline/pipeline/utils/transform_utils.py

@@ -41,17 +41,35 @@ def _split_params(params: Any) -> Tuple[Tuple[Any, ...], dict[str, Any]]:
     return (params,), {}
 
 
+def _merge_extra_kwargs(
+    fn: Callable[..., Any],
+    kwargs: dict[str, Any],
+    extra_kwargs: Mapping[str, Any] | None,
+) -> dict[str, Any]:
+    if not extra_kwargs:
+        return kwargs
+    merged = dict(kwargs)
+    for key, value in extra_kwargs.items():
+        if key in merged:
+            continue
+        if _supports_parameter(fn, key):
+            merged[key] = value
+    return merged
+
+
 def _call_with_params(
     fn: Callable,
     stream: Iterator[Any],
     params: Any,
     context: Optional[PipelineContext],
+    extra_kwargs: Mapping[str, Any] | None = None,
 ) -> Iterator[Any]:
     """Invoke an entry-point callable with optional params semantics."""
 
     args, kwargs = _split_params(params)
     if context and _supports_parameter(fn, "context") and "context" not in kwargs:
         kwargs["context"] = context
+    kwargs = _merge_extra_kwargs(fn, kwargs, extra_kwargs)
     return fn(stream, *args, **kwargs)
 
 
@@ -59,12 +77,14 @@ def _instantiate_entry_point(
     cls: Callable[..., Any],
     params: Any,
     context: Optional[PipelineContext],
+    extra_kwargs: Mapping[str, Any] | None = None,
 ) -> Any:
     """Instantiate a transform class with parameters from the config."""
 
     args, kwargs = _split_params(params)
     if context and _supports_parameter(cls.__init__, "context") and "context" not in kwargs:
         kwargs["context"] = context
+    kwargs = _merge_extra_kwargs(cls.__init__, kwargs, extra_kwargs)
     return cls(*args, **kwargs)
 
 
@@ -83,6 +103,7 @@ def apply_transforms(
     context: Optional[PipelineContext] = None,
     observer: Callable[[TransformEvent], None] | None = None,
     observer_registry: ObserverRegistry | None = None,
+    extra_kwargs: Mapping[str, Any] | None = None,
 ) -> Iterator[Any]:
     """Instantiate and apply configured transforms in order."""
 
@@ -97,7 +118,9 @@ def apply_transforms(
             name, params = _extract_single_pair(transform, "Transform")
             ep = load_ep(group=group, name=name)
             if isclass(ep):
-                inst = _instantiate_entry_point(ep, params, context)
+                inst = _instantiate_entry_point(
+                    ep, params, context, extra_kwargs=extra_kwargs
+                )
                 _bind_context(inst, context)
                 eff_observer = observer
                 if eff_observer is None and registry:
@@ -107,7 +130,9 @@ def apply_transforms(
                 _attach_observer(inst, eff_observer)
                 stream = inst(stream)
             else:
-                stream = _call_with_params(ep, stream, params, context)
+                stream = _call_with_params(
+                    ep, stream, params, context, extra_kwargs=extra_kwargs
+                )
     return stream
 
 

datapipeline/services/artifacts.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 from dataclasses import dataclass
 import json
 from pathlib import Path
@@ -80,8 +78,7 @@ class ArtifactManager:
         except FileNotFoundError as exc:
             message = (
                 f"Artifact file not found: {path}. "
-                "Run `jerry build --project <project.yaml>` (preferred) or "
-                "`jerry inspect expected --project <project.yaml>` to regenerate it."
+                "Run `jerry build --project <project.yaml>` to regenerate it."
             )
             raise RuntimeError(message) from exc
 

datapipeline/services/constants.py

@@ -12,6 +12,7 @@ LOADERS_GROUP = "loaders"
 MAPPERS_GROUP = "mappers"
 FILTERS_GROUP = "filters"
 DEFAULT_IO_LOADER_EP = "core.io"
+DEFAULT_SYNTHETIC_LOADER_EP = "core.synthetic.ticks"
 
 # POSTPROCESS_GLOBAL_KEY = "__global__"
 POSTPROCESS_TRANSFORMS = "transforms"

datapipeline/services/factories.py

@@ -6,8 +6,7 @@ from datapipeline.mappers.noop import identity
 from datapipeline.utils.placeholders import normalize_args
 from datapipeline.sources.models.base import SourceInterface
 from datapipeline.pipeline.context import PipelineContext
-from datapipeline.config.dataset.feature import FeatureRecordConfig
-from datapipeline.pipeline.pipelines import build_feature_pipeline
+from datapipeline.pipeline.pipelines import build_record_pipeline
 from datapipeline.pipeline.utils.transform_utils import _supports_parameter
 from inspect import isclass
 from typing import Iterator, Any, Optional
@@ -52,7 +51,7 @@ class _ComposedSource(SourceInterface):
 
         # Build aligned/aux iterators (unwrap FeatureRecord -> record for aligned)
         aligned_iters: dict[str, Iterator[Any]] = {
-            k: (fr.record for fr in v["iter"])  # stage>=3 yields FeatureRecord
+            k: (getattr(item, "record", item) for item in v["iter"])
             for k, v in aligned.items()
         }
         aux_iters: dict[str, Iterator[Any]] = {
@@ -111,7 +110,7 @@ class _ComposedSource(SourceInterface):
         """Parse and resolve composed inputs into iterators.
 
         Grammar: "[alias=]stream_id" only. All inputs are built to stage 4
-        and are alignable (FeatureRecord -> domain record unwrapped).
+        and are alignable (domain records with stream transforms applied).
         """
         runtime = context.runtime
         known_streams = set(runtime.registries.stream_sources.keys())
@@ -123,8 +122,7 @@ class _ComposedSource(SourceInterface):
                 raise ValueError(
                     f"Unknown input stream '{ref}'. Known streams: {sorted(known_streams)}"
                 )
-            cfg = FeatureRecordConfig(record_stream=ref, id=alias)
-            it = build_feature_pipeline(context, cfg, stage=4)
+            it = build_record_pipeline(context, ref, stage=4)
             out[alias] = {"iter": it, "aligned": True}
 
         return out

datapipeline/services/paths.py

@@ -8,7 +8,16 @@ def pkg_root(start: Optional[Path] = None) -> tuple[Path, str, Path]:
     for d in [here, *here.parents]:
         pyproject = d / "pyproject.toml"
         if pyproject.exists():
-            return d, d.name, pyproject
+            pkg_name = d.name
+            src_dir = d / "src"
+            if src_dir.exists():
+                candidates = [
+                    p for p in src_dir.iterdir()
+                    if p.is_dir() and (p / "__init__.py").exists()
+                ]
+                if len(candidates) == 1:
+                    pkg_name = candidates[0].name
+            return d, pkg_name, pyproject
     print("[error] pyproject.toml not found (searched current and parent dirs)", file=sys.stderr)
     raise SystemExit(1)
 

datapipeline/services/project_paths.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 from pathlib import Path
 from typing import Optional
 

datapipeline/services/runs.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 from dataclasses import dataclass, asdict
 from datetime import datetime, timezone
 from pathlib import Path