v2: BBOmix evaluation tables for conditional proposals

README.md

@@ -13,15 +13,21 @@ viewer: false
 
 # AUTOENCODIX initial-config portfolios
 
-This repository holds the lookup table that
+This repository holds the data that
 [`autoencodix.propose_initial_config()`](https://github.com/jan-forest/autoencodix_package) uses to suggest
 a strong starting hyperparameter configuration for an AUTOENCODIX model. You don't have to tune
 from the generic schema defaults.
 
-The portfolios were computed **offline** by fitting Syne Tune's `ZeroShotTransfer` scheduler to the
-**BBOmix** benchmark: about 105,000 AUTOENCODIX training runs across several architectures, two
-datasets and many downstream tasks. At call time no optimizer runs. The package just looks up the
-JSON and validates the result with Pydantic.
+Everything here comes from the **BBOmix** benchmark: about 105,000 AUTOENCODIX training runs
+across several architectures, two datasets and many downstream tasks. There are two kinds of files:
+
+- **`initial_configs.json`**: ranked portfolios computed offline by fitting Syne Tune's
+  `ZeroShotTransfer` scheduler to BBOmix. A default call just looks up this JSON and validates the
+  result with Pydantic; no optimizer runs.
+- **`evaluations/*.parquet` + `search_space.json`** (from `v2`): the BBOmix evaluations themselves,
+  in compact form. They are used when you fix some hyperparameters (`fixed={...}`). The package
+  then fits a random-forest surrogate on them and proposes the remaining hyperparameters *given*
+  your fixed values.
 
 ## Usage
 
@@ -35,10 +41,13 @@ config = acx.propose_initial_config("varix")
 config = acx.propose_initial_config("varix", revision="v1")
 
 # list the available versions
-acx.tuning.list_portfolio_versions()   # e.g. ["v1"]
+acx.tuning.list_portfolio_versions()   # e.g. ["v1", "v2"]
+
+# fix some hyperparameters, propose the rest given them (needs v2 or later)
+config = acx.propose_initial_config("varix", fixed={"latent_dim": 16, "batch_size": 64})
 ```
 
-The package downloads `initial_configs.json` from here on first use and caches it locally with
+The package downloads the files it needs from here on first use and caches them locally with
 `huggingface_hub`. After that it also works offline (`HF_HUB_OFFLINE=1`). You can also use a
 portfolio of your own:
 
@@ -47,8 +56,23 @@ acx.propose_initial_config("varix", repo_id="<user>/<repo>")
 acx.propose_initial_config("varix", artifact_path="path/to/initial_configs.json")  # local file, no Hub
 ```
 
+With `artifact_path` and `fixed`, the evaluation files are expected in this repository's layout
+next to the JSON (`search_space.json`, `evaluations/<architecture>.parquet`).
+
 See the *Propose Initial Config* tutorial in the AUTOENCODIX docs for all options (`objective`,
-`budget_epochs`, `dataset`, `top_k`, overrides).
+`budget_epochs`, `dataset`, `top_k`, `fixed`, overrides).
+
+### `fixed` vs. overrides
+
+Keyword overrides (`propose_initial_config("varix", latent_dim=16)`) only overwrite that field of
+the static proposal. The other values stay the ones tuned together with a different `latent_dim`.
+`fixed={"latent_dim": 16}` re-proposes the other hyperparameters for `latent_dim=16`.
+
+In a leave-one-task-out check on the `v2` data, we held out each task, fixed one of `latent_dim`,
+`n_layers`, `batch_size` or `k_filter` at several values, and compared proposals on the held-out
+task among configs with that value. Conditional proposals reached the **87.7th percentile** on
+average, compared with the 72.4th for the static top config + override. Conditional was better in
+81% of 546 cases, and it was better on average for every architecture and both objectives.
 
 ## Versions
 
@@ -59,17 +83,22 @@ that tag.
 | Tag  | Description |
 |------|-------------|
 | `v1` | Initial BBOmix zero-shot portfolio: vanillix, varix, ontix, disentanglix; datasets `tcga`, `schc` |
+| `v2` | Same portfolio as `v1`, plus `evaluations/*.parquet` and `search_space.json` for conditional proposals (`fixed=`) |
 
 ### Publishing a new version (maintainers)
 
+The evaluation files are produced by `benchmarking/tuning/export_evaluations.py` in the
+AUTOENCODIX repository, from the local Syne Tune blackbox repository.
+
 ```bash
-hf upload autoencodix/initial-configs initial_configs.json initial_configs.json \
-   --repo-type dataset --commit-message "Portfolio v2: <what changed>"
-hf repo tag create autoencodix/initial-configs v2 --repo-type dataset -m "<what changed>"
+python benchmarking/tuning/export_evaluations.py --out <dir>      # evaluations/ + search_space.json
+hf upload autoencodix/initial-configs <dir> . --repo-type dataset \
+   --commit-message "Portfolio v3: <what changed>"
+hf repo tag create autoencodix/initial-configs v3 --repo-type dataset -m "<what changed>"
 ```
 
-Then add a row to the table above. The new file must keep the layout described below. The
-package checks dataset names and budgets against whatever the file contains, so you can add new
+Then add a row to the table above. The files must keep the layouts described below. The
+package checks dataset names and budgets against whatever the files contain, so you can add new
 datasets or budgets without a package release.
 
 ## File layout
@@ -115,6 +144,39 @@ configs.
   key holds configs tuned for training that many epochs (`epochs` and `checkpoint_interval` are
   set to the budget).
 
+### `search_space.json`
+
+```text
+{
+  "constants": {"epochs": 300, "checkpoint_interval": 300, "loss_reduction": "sum"},  # held constant in BBOmix
+  "budgets": [10, 25, 50, 100, 150, 200, 300],    # reconstruction budgets in the evaluation tables
+  "architectures": {
+    "<architecture>": {
+      "<hp>": {"choices": [...], "low": .., "high": ..},              # grid-sampled (integer) hps
+      "<hp>": {"scale": "log" | "linear", "low": .., "high": ..},    # continuous hps
+      ...
+    }
+  }
+}
+```
+
+`low`/`high` are the range actually covered by the evaluations, which can be narrower than the
+declared BBOmix search space. Only these hyperparameters can be fixed. Fixed values outside the
+range trigger a warning, because the proposal is then extrapolated.
+
+### `evaluations/<architecture>.parquet`
+
+One row per (task, config). All tasks of a dataset share the same 1000 sampled configs.
+
+| Column | Meaning |
+|--------|---------|
+| `dataset` | BBOmix dataset, e.g. `tcga`, `schc` |
+| `task` | BBOmix task (data modality combination, plus ontology for ontix) |
+| `config_id` | Index of the config within its dataset |
+| `<hp>` columns | Hyperparameter values, as in `search_space.json` |
+| `downstream` | `metric_avg_ml_task_performance` at the final epoch, mean over 3 seeds (higher is better) |
+| `reconstruction_<b>` | `metric_valid_recon_loss` at epoch `b`, mean over 3 seeds (lower is better) |
+
 ## License
 
 Apache-2.0, the same license as the AUTOENCODIX package.

evaluations/disentanglix.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:095bbd4fc8da7dac0f3c73dcadd44b86ff3462b3b177036df520dbada46259f9
+size 395506

evaluations/ontix.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b3bf11dc873b77da184d675ec2ede5f08d87c9cb2d589f0d76add02db87d191
+size 706454

evaluations/vanillix.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e23d565cb77e52c5789b96a2a187b6d5c3d48c2143a6aab7a315c3bbce1e0df
+size 363896

evaluations/varix.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f258380ba3f1497a2b55a7f02d73a1bc1bd5e37826367006baa14f1d49e62be
+size 374823

search_space.json

@@ -0,0 +1,311 @@
+{
+  "constants": {
+    "epochs": 300,
+    "checkpoint_interval": 300,
+    "loss_reduction": "sum"
+  },
+  "budgets": [
+    10,
+    25,
+    50,
+    100,
+    150,
+    200,
+    300
+  ],
+  "architectures": {
+    "vanillix": {
+      "k_filter": {
+        "choices": [
+          128,
+          256,
+          512,
+          1024,
+          2048,
+          4096
+        ],
+        "low": 128,
+        "high": 4096
+      },
+      "n_layers": {
+        "choices": [
+          2,
+          3,
+          4
+        ],
+        "low": 2,
+        "high": 4
+      },
+      "enc_factor": {
+        "choices": [
+          1,
+          2,
+          3,
+          4
+        ],
+        "low": 1,
+        "high": 4
+      },
+      "batch_size": {
+        "choices": [
+          32,
+          64,
+          128,
+          256
+        ],
+        "low": 32,
+        "high": 256
+      },
+      "learning_rate": {
+        "scale": "log",
+        "low": 1.0207919389761473e-05,
+        "high": 0.09854177738894777
+      },
+      "drop_p": {
+        "scale": "linear",
+        "low": 0.00031825130718524667,
+        "high": 0.8995006217338251
+      },
+      "weight_decay": {
+        "scale": "log",
+        "low": 1.0077964816880098e-05,
+        "high": 0.0999578174192585
+      },
+      "latent_dim": {
+        "choices": [
+          2,
+          4,
+          8,
+          16,
+          32,
+          64
+        ],
+        "low": 2,
+        "high": 64
+      }
+    },
+    "varix": {
+      "k_filter": {
+        "choices": [
+          128,
+          256,
+          512,
+          1024,
+          2048,
+          4096
+        ],
+        "low": 128,
+        "high": 4096
+      },
+      "n_layers": {
+        "choices": [
+          2,
+          3,
+          4
+        ],
+        "low": 2,
+        "high": 4
+      },
+      "enc_factor": {
+        "choices": [
+          1,
+          2,
+          3,
+          4
+        ],
+        "low": 1,
+        "high": 4
+      },
+      "batch_size": {
+        "choices": [
+          32,
+          64,
+          128,
+          256
+        ],
+        "low": 32,
+        "high": 256
+      },
+      "learning_rate": {
+        "scale": "log",
+        "low": 1.0052950994489815e-05,
+        "high": 0.09875932708929694
+      },
+      "drop_p": {
+        "scale": "linear",
+        "low": 0.002751654311746721,
+        "high": 0.8995006217338251
+      },
+      "weight_decay": {
+        "scale": "log",
+        "low": 1.000328971683056e-05,
+        "high": 0.09837910977086362
+      },
+      "beta": {
+        "scale": "log",
+        "low": 0.0001015579593600663,
+        "high": 0.9995781741925843
+      },
+      "latent_dim": {
+        "choices": [
+          2,
+          4,
+          8,
+          16,
+          32,
+          64
+        ],
+        "low": 2,
+        "high": 64
+      }
+    },
+    "ontix": {
+      "k_filter": {
+        "choices": [
+          128,
+          256,
+          512,
+          1024,
+          2048,
+          4096
+        ],
+        "low": 128,
+        "high": 4096
+      },
+      "n_layers": {
+        "choices": [
+          2,
+          3,
+          4
+        ],
+        "low": 2,
+        "high": 4
+      },
+      "enc_factor": {
+        "choices": [
+          1,
+          2,
+          3,
+          4
+        ],
+        "low": 1,
+        "high": 4
+      },
+      "batch_size": {
+        "choices": [
+          32,
+          64,
+          128,
+          256
+        ],
+        "low": 32,
+        "high": 256
+      },
+      "learning_rate": {
+        "scale": "log",
+        "low": 1.000328971683056e-05,
+        "high": 0.09921147863447119
+      },
+      "drop_p": {
+        "scale": "linear",
+        "low": 0.001259965328730206,
+        "high": 0.8994950018997666
+      },
+      "weight_decay": {
+        "scale": "log",
+        "low": 1.0160512115112301e-05,
+        "high": 0.09997846955826961
+      },
+      "beta": {
+        "scale": "log",
+        "low": 0.00010146454456689534,
+        "high": 0.9684601729867341
+      }
+    },
+    "disentanglix": {
+      "k_filter": {
+        "choices": [
+          128,
+          256,
+          512,
+          1024,
+          2048,
+          4096
+        ],
+        "low": 128,
+        "high": 4096
+      },
+      "n_layers": {
+        "choices": [
+          2,
+          3,
+          4
+        ],
+        "low": 2,
+        "high": 4
+      },
+      "enc_factor": {
+        "choices": [
+          1,
+          2,
+          3,
+          4
+        ],
+        "low": 1,
+        "high": 4
+      },
+      "batch_size": {
+        "choices": [
+          32,
+          64,
+          128,
+          256
+        ],
+        "low": 32,
+        "high": 256
+      },
+      "learning_rate": {
+        "scale": "log",
+        "low": 1.0298900014766878e-05,
+        "high": 0.09875932708929694
+      },
+      "drop_p": {
+        "scale": "linear",
+        "low": 0.0003089737316932961,
+        "high": 0.899499219679429
+      },
+      "weight_decay": {
+        "scale": "log",
+        "low": 1.0209532496205165e-05,
+        "high": 0.09615619068896737
+      },
+      "latent_dim": {
+        "choices": [
+          2,
+          4,
+          8,
+          16,
+          32,
+          64
+        ],
+        "low": 2,
+        "high": 64
+      },
+      "beta_mi": {
+        "scale": "log",
+        "low": 0.00010060475401698013,
+        "high": 0.9977803145650433
+      },
+      "beta_tc": {
+        "scale": "log",
+        "low": 0.010472424642301324,
+        "high": 985.2670858275624
+      },
+      "beta_dimKL": {
+        "scale": "log",
+        "low": 0.00010032622015055023,
+        "high": 0.9973010044095937
+      }
+    }
+  }
+}