|
| 1 | +--- |
| 2 | +title: Upgrade guide |
| 3 | +--- |
| 4 | + |
| 5 | +# Upgrade guide |
| 6 | + |
| 7 | +This page explains how to upgrade an existing code base to **PLAID v1.0.0**. |
| 8 | + |
| 9 | +PLAID follows [Semantic Versioning](https://semver.org/). The `v1.0.0` release is |
| 10 | +the first major release: it consolidates the data model, removes deprecated and |
| 11 | +out-of-scope modules, and simplifies several public APIs. As a major release, it |
| 12 | +contains **breaking changes**. |
| 13 | + |
| 14 | +The guide is organized by **version jump**. Read the section that matches the |
| 15 | +version you are upgrading *from*. For the exhaustive, change-by-change history, |
| 16 | +see the [`CHANGELOG.md`](https://github.com/PLAID-lib/plaid/blob/main/CHANGELOG.md). |
| 17 | + |
| 18 | +!!! tip "Upgrade incrementally" |
| 19 | + If you are several versions behind, pin `pyplaid`, upgrade one step at a |
| 20 | + time, and run your test suite between steps. The last release of the `0.x` |
| 21 | + series is **`0.1.15`**; the sections below describe the jump from `0.1.x` to |
| 22 | + `1.0.0`. |
| 23 | + |
| 24 | +!!! info "Related documentation" |
| 25 | + This guide focuses on *what changed* and *how to migrate*. For *how the new |
| 26 | + API works*, see: |
| 27 | + |
| 28 | + - [Quickstart](quickstart.md) — the new read/write pattern in a nutshell. |
| 29 | + - [Concepts](concepts.md) — [Sample](concepts/sample.md), |
| 30 | + [Problem definition](concepts/problem_definition.md), |
| 31 | + [Infos](concepts/infos.md), [Dataset](concepts/dataset.md), |
| 32 | + [Disk format](concepts/disk_format.md). |
| 33 | + - [Conversion tutorial](tutorials/storage.md) — end-to-end storage workflow |
| 34 | + (save, load, backends, Hub, parallel I/O). |
| 35 | + - [API reference](api/index.md) — in particular |
| 36 | + [`plaid.storage`](api/storage/backend_api.md). |
| 37 | + |
| 38 | +--- |
| 39 | + |
| 40 | +## Upgrade to v1.0.0 from 0.1.x |
| 41 | + |
| 42 | +`v1.0.0` reorganizes the package. The changes most likely to affect your code are |
| 43 | +listed below, with before/after examples. |
| 44 | + |
| 45 | +### Top-level imports |
| 46 | + |
| 47 | +The single biggest change is that **the `Dataset` class has been removed**: it is |
| 48 | +no longer exported from the top-level `plaid` package *and no longer exists as a |
| 49 | +module either*. A new `Infos` object is exported, and the version string moved |
| 50 | +module. See [Removing the `Dataset` class](#removing-the-dataset-class-use-plaidstorage) |
| 51 | +below for the full migration. |
| 52 | + |
| 53 | +```python |
| 54 | +# Before (0.1.x) |
| 55 | +from plaid import Dataset, Sample, ProblemDefinition |
| 56 | +from plaid import __version__ # backed by plaid._version |
| 57 | + |
| 58 | +# After (1.0.0) |
| 59 | +from plaid import Sample, ProblemDefinition, Infos |
| 60 | +from plaid import __version__ # backed by plaid.version |
| 61 | +# `Dataset` no longer exists — there is no `plaid.containers.dataset` module. |
| 62 | +# Use the storage helpers instead: |
| 63 | +from plaid.storage import save_to_disk, init_from_disk |
| 64 | +``` |
| 65 | + |
| 66 | +The helpers `get_number_of_samples` and `get_sample_ids` are still exported from |
| 67 | +the top-level package. |
| 68 | + |
| 69 | +### Removing the `Dataset` class: use `plaid.storage` |
| 70 | + |
| 71 | +In the `0.1.x` series, `plaid.Dataset` was a **monolithic, in-memory container**: |
| 72 | +you built one `Dataset` object, appended every `Sample` to it, kept the whole |
| 73 | +collection in RAM, and called `save_to_dir` / `load` on that object. |
| 74 | + |
| 75 | +In `v1.0.0` this class is **removed entirely** — there is no public high-level |
| 76 | +dataset container class anymore, and there is no `plaid.containers.dataset` |
| 77 | +module to import from. The data model is now centered on three objects — |
| 78 | +[`Sample`](concepts/sample.md), [`ProblemDefinition`](concepts/problem_definition.md) |
| 79 | +and [`Infos`](concepts/infos.md) — plus the storage helpers in |
| 80 | +[`plaid.storage`](api/storage/backend_api.md). A dataset on disk is a shared |
| 81 | +metadata layout plus backend-specific sample payloads; loading it back gives you, |
| 82 | +**per split**, a backend dataset object and a `Converter` that materializes |
| 83 | +individual `Sample` objects lazily. |
| 84 | + |
| 85 | +This is a deliberate shift away from "load the whole dataset into one in-memory |
| 86 | +object" toward **backend-agnostic, lazy, per-sample access**, so that large |
| 87 | +datasets that do not fit in memory can be streamed sample by sample into ML |
| 88 | +pipelines. The concepts are introduced in [Quickstart](quickstart.md) and the |
| 89 | +[Dataset concept page](concepts/dataset.md); the end-to-end workflow is in the |
| 90 | +[Conversion tutorial](tutorials/storage.md). |
| 91 | + |
| 92 | +#### Writing: build-then-append → `save_to_disk(sample_constructor, ids)` |
| 93 | + |
| 94 | +Instead of building a `Dataset` and appending samples, you provide a |
| 95 | +`sample_constructor(id) -> Sample` callable plus an `ids` mapping of split names |
| 96 | +to sliceable id sequences. PLAID handles iteration, generator creation and |
| 97 | +parallel sharding internally, and writes directly to the chosen backend. |
| 98 | + |
| 99 | +```python |
| 100 | +# Before (0.1.x) — everything in memory, then dumped |
| 101 | +from plaid import Dataset, Sample |
| 102 | + |
| 103 | +dataset = Dataset() |
| 104 | +for raw in raw_items: |
| 105 | + sample = Sample() |
| 106 | + # fill the sample: add_tree, add_field, ... |
| 107 | + dataset.add_sample(sample) |
| 108 | +dataset.save_to_dir("my_plaid_dataset") |
| 109 | + |
| 110 | +# After (1.0.0) — lazy, per-sample, backend-aware |
| 111 | +from plaid import Sample |
| 112 | +from plaid.storage import save_to_disk |
| 113 | + |
| 114 | +def sample_constructor(sample_id): |
| 115 | + sample = Sample() |
| 116 | + # fill the sample: add_tree, add_field, ... |
| 117 | + return sample |
| 118 | + |
| 119 | +save_to_disk( |
| 120 | + "my_plaid_dataset", |
| 121 | + sample_constructor=sample_constructor, |
| 122 | + ids={"train": [0, 1, 2], "test": [3, 4]}, |
| 123 | + backend="zarr", # one of "hf_datasets", "cgns", "zarr" |
| 124 | +) |
| 125 | +``` |
| 126 | + |
| 127 | +See the [Conversion tutorial](tutorials/storage.md) for a complete example |
| 128 | +(including `num_proc` parallel writing and `push_to_hub`) and the |
| 129 | +[writer API](api/storage/writer.md). |
| 130 | + |
| 131 | +#### Reading: `Dataset.load(...)` → `init_from_disk(...)` + converter |
| 132 | + |
| 133 | +Loading no longer returns a single object you index into. It returns a |
| 134 | +dictionary of backend datasets and a dictionary of converters, one per split. |
| 135 | +You materialize a `Sample` on demand with `converter.to_plaid(dataset, idx)`. |
| 136 | + |
| 137 | +```python |
| 138 | +# Before (0.1.x) |
| 139 | +from plaid import Dataset |
| 140 | + |
| 141 | +dataset = Dataset() |
| 142 | +dataset.load("my_plaid_dataset") |
| 143 | +sample = dataset[0] |
| 144 | +n = len(dataset) |
| 145 | + |
| 146 | +# After (1.0.0) |
| 147 | +from plaid.storage import init_from_disk |
| 148 | + |
| 149 | +datasetdict, converterdict = init_from_disk("my_plaid_dataset") |
| 150 | +dataset = datasetdict["train"] |
| 151 | +converter = converterdict["train"] |
| 152 | + |
| 153 | +sample = converter.to_plaid(dataset, 0) # materialize one Sample lazily |
| 154 | +n = len(dataset) |
| 155 | +``` |
| 156 | + |
| 157 | +The same shape is used for the Hub (`download_from_hub`, |
| 158 | +`init_streaming_from_hub`). See the [Dataset concept page](concepts/dataset.md), |
| 159 | +the [reader API](api/storage/reader.md), and the [backend API](api/storage/backend_api.md). |
| 160 | + |
| 161 | +#### Operation-by-operation map |
| 162 | + |
| 163 | +| `0.1.x` — `Dataset` method | `1.0.0` — replacement | |
| 164 | +| --- | --- | |
| 165 | +| `Dataset()` + `add_sample` / `add_samples` / `from_list_of_samples` | `save_to_disk(sample_constructor=..., ids=...)` | |
| 166 | +| `Dataset.save_to_dir(path)` / `add_to_dir` | `save_to_disk(path, sample_constructor=..., ids=...)` | |
| 167 | +| `Dataset.load(path)` | `init_from_disk(path)` → `(datasetdict, converterdict)` | |
| 168 | +| `dataset[i]` / `get_samples()` | `converter.to_plaid(dataset, i)` | |
| 169 | +| `len(dataset)` / `get_number_of_samples()` | `len(dataset)` (per-split backend object) | |
| 170 | +| `dataset.set_infos(...)` / `get_infos()` | pass [`Infos`](concepts/infos.md) to `save_to_disk(infos=...)`; read back with `Infos.from_path(path)` | |
| 171 | +| persisting a `ProblemDefinition` with the dataset | `save_to_disk(..., pb_defs=...)`; read back with `load_problem_definitions_from_disk(path)` | |
| 172 | +| `Dataset.add_features_from_tabular` (ex-`from_tabular`) | build the corresponding `Sample` objects in `sample_constructor` | |
| 173 | +| `Dataset.extract_dataset_from_identifier` | request features at read time: `converter.to_plaid(dataset, i, features=[...])` | |
| 174 | +| `Dataset.get_tabular_from_stacked_identifiers` | gather features yourself from the materialized `Sample` objects | |
| 175 | +| `plaid.examples` | `plaid.downloadable_examples` | |
| 176 | +| change backend (e.g. CGNS → HF) | `init_from_disk` then `save_to_disk` with the new `backend` (see the [Conversion tutorial](tutorials/storage.md)) | |
| 177 | + |
| 178 | +If you only need a subset of features or spatial indices, the converter supports |
| 179 | +`features=[...]` and `indexers={...}` for partial reads on the `hf_datasets` and |
| 180 | +`zarr` backends — see the [Conversion tutorial](tutorials/storage.md#indexed-extraction-with-indexers). |
| 181 | + |
| 182 | +### Removed modules |
| 183 | + |
| 184 | +The following modules were removed from the `plaid` package in `1.0.0`. They were |
| 185 | +either out of the scope of the data model or superseded: |
| 186 | + |
| 187 | +| Removed module | What to do instead | |
| 188 | +| --- | --- | |
| 189 | +| `plaid.pipelines` (`plaid_blocks`, `sklearn_block_wrappers`) | build ML pipelines outside PLAID, on top of the data model | |
| 190 | +| `plaid.post` (`bisect`, `metrics`) | compute post-processing / metrics in your own code | |
| 191 | +| `plaid.utils.split` | manage dataset splits via `ProblemDefinition` train/test splits | |
| 192 | +| `plaid.utils.stats` | compute statistics in your own code | |
| 193 | +| `plaid.utils.interpolation` | use an external interpolation routine | |
| 194 | +| `plaid.utils.init_with_tabular` | construct samples explicitly | |
| 195 | +| `plaid.utils.deprecation`, `plaid.utils.base` | internal helpers, no public replacement | |
| 196 | + |
| 197 | +If you imported any of these, remove the import and move the corresponding logic |
| 198 | +into your own project, or rely on the supported data-model APIs. |
| 199 | + |
| 200 | +### `ProblemDefinition` |
| 201 | + |
| 202 | +`ProblemDefinition` was rewritten as a compact [pydantic](https://docs.pydantic.dev/) |
| 203 | +model with four required fields — `input_features`, `output_features`, |
| 204 | +`train_split` and `test_split`. The many `*_features_identifiers` accessors were |
| 205 | +collapsed into two methods, splits became plain model attributes, and YAML key |
| 206 | +order is now enforced on save. |
| 207 | + |
| 208 | +```python |
| 209 | +# Before (0.1.x) |
| 210 | +pb.add_in_features_identifiers([...]) |
| 211 | +pb.add_out_features_identifiers([...]) |
| 212 | +pb.set_in_features_identifiers([...]) |
| 213 | +pb.set_out_features_identifiers([...]) |
| 214 | +pb.get_in_features_identifiers() |
| 215 | +pb.get_out_features_identifiers() |
| 216 | +pb.get_split("train") # split accessors |
| 217 | + |
| 218 | +# After (1.0.0) |
| 219 | +from plaid import ProblemDefinition |
| 220 | + |
| 221 | +pb = ProblemDefinition( |
| 222 | + input_features=["Base/Zone/GridCoordinates/CoordinateX"], |
| 223 | + output_features=["Base/Zone/VertexFields/pressure"], |
| 224 | + train_split={"train": [0, 1, 2]}, |
| 225 | + test_split={"test": [3, 4]}, |
| 226 | +) |
| 227 | +pb.add_input_features([...]) |
| 228 | +pb.add_output_features([...]) |
| 229 | +pb.train_split # direct attribute access |
| 230 | +pb.test_split |
| 231 | +``` |
| 232 | + |
| 233 | +The public surface of `ProblemDefinition` in `1.0.0` is intentionally small: |
| 234 | +`from_path`, `model_validate`, `add_input_features`, `add_output_features`, |
| 235 | +`save_to_file`, and the four model fields (`input_features`, `output_features`, |
| 236 | +`train_split`, `test_split`). The previous `constant_features_identifiers` |
| 237 | +accessors and the `get_*_split_*` / `set_*_split_*` helpers were removed |
| 238 | +together with the in/out identifier accessors; splits are now read and assigned |
| 239 | +directly via the `train_split` / `test_split` attributes, and feature lists are |
| 240 | +normalized (stringified, sorted, deduplicated, non-empty) by pydantic |
| 241 | +validators. The problem name is no longer stored in the model — on disk it is |
| 242 | +the YAML filename stem, in memory it is the dictionary key returned by |
| 243 | +`load_problem_definitions_from_disk`. See the |
| 244 | +[Problem definition concept page](concepts/problem_definition.md) and the |
| 245 | +[`problem_definition` API](api/problem_definition.md). |
| 246 | + |
| 247 | +### Storage / CGNS backend |
| 248 | + |
| 249 | +The constant/variable mechanism used in the CGNS backend reading and writing |
| 250 | +paths was removed. If you relied on that distinction at the storage level, review |
| 251 | +your read/write code against the current |
| 252 | +[backend API](api/storage/backend_api.md) and the |
| 253 | +[CGNS backend API](api/storage/cgns/index.md). The on-disk layout written by |
| 254 | +`save_to_disk` (shared metadata + per-backend payloads) is described in the |
| 255 | +[Disk format concept page](concepts/disk_format.md), and the three backends |
| 256 | +(`hf_datasets`, `cgns`, `zarr`) are compared in the |
| 257 | +[Conversion tutorial](tutorials/storage.md#choosing-a-backend). |
| 258 | + |
| 259 | +### New in v1.0.0 |
| 260 | + |
| 261 | +`v1.0.0` also introduces new building blocks you can adopt: |
| 262 | + |
| 263 | +- **`plaid.infos`** — a dedicated pydantic `Infos` class, now living at the same |
| 264 | + level as `ProblemDefinition` (see [Infos](concepts/infos.md)). |
| 265 | +- **`plaid.viewer`** — an interactive [trame](https://kitware.github.io/trame/) |
| 266 | + application for visual dataset exploration (see [Viewer](concepts/viewer.md)). |
| 267 | +- **`plaid-check`** — a CLI tool that validates the integrity of a local PLAID |
| 268 | + dataset (on-disk layout, `infos.yaml`, splits, sample conversion, invalid |
| 269 | + numeric values, duplicated samples, and optional problem definitions); see |
| 270 | + [Dataset check](concepts/check.md). |
| 271 | + |
| 272 | +--- |
| 273 | + |
| 274 | +## Upgrading from an older 0.1.x release |
| 275 | + |
| 276 | +If you are upgrading from a release earlier than `0.1.15`, first move up to |
| 277 | +`0.1.15` and account for the intermediate breaking changes documented in the |
| 278 | +[`CHANGELOG.md`](https://github.com/PLAID-lib/plaid/blob/main/CHANGELOG.md), in |
| 279 | +particular: |
| 280 | + |
| 281 | +- **0.1.15** — `save_to_disk` API simplified: `generators` replaced by |
| 282 | + `sample_constructor` and `ids`. |
| 283 | +- **0.1.13** — `get_mesh` renamed to `get_tree`; `get_<x>_assignment` renamed to |
| 284 | + `resolve_<x>` (e.g. `get_time_assignment` → `resolve_time`). |
| 285 | +- **0.1.11** — `get_all_mesh_times()` renamed to `get_all_time_values()`; |
| 286 | + `FeatureIdentifier` moved from `plaid.types` to `plaid.containers`; Python 3.10 |
| 287 | + support dropped. |
| 288 | +- **0.1.10** — `Sample` restructured to store globals at time steps (scalars and |
| 289 | + time series unified into CGNS trees). |
| 290 | +- **0.1.8** — `Dataset.from_tabular` → `Dataset.add_features_from_tabular`; |
| 291 | + `Dataset.from_features_identifier` → `Dataset.extract_dataset_from_identifier`; |
| 292 | + `Sample.from_features_identifier` → `Sample.extract_sample_from_identifier`. |
| 293 | + |
| 294 | +Once on `0.1.15`, follow the [Upgrade to v1.0.0 from 0.1.x](#upgrade-to-v100-from-01x) |
| 295 | +section above. |
0 commit comments