Découverte de caractéristiques par autoresearch
Exécute une boucle autoresearch qui propose des questions TypeSafe, convertit du texte libre en caractéristiques numériques et utilise les erreurs du modèle pour améliorer un régresseur CatBoost supervisé.
Les questions TypeSafe transforment du texte libre en caractéristiques numériques pour un modèle CatBoost supervisé ; utilise une boucle autoresearch pour les découvrir.
CatBoost a besoin d’un tableau de nombres, et une note de dégustation n’en est pas un. Ce cookbook construit le tableau à partir de questions sur la note, et aucune n’est écrite à la main. Un LLM propose les questions, TypeSafe y répond pour chaque ligne, et CatBoost s’entraîne sur les réponses. La partie autoresearch, c’est ce qui suit : CatBoost indique quelles questions il a utilisées et quelles lignes il se trompe encore, l’appel de proposition suivant lit ce rapport, et la boucle repart.
À la fin, tu disposes d’une boucle que tu peux pointer vers ton propre texte étiqueté, d’une courbe de l’erreur hors échantillon par tour, et d’un tableau des questions que le modèle final a le plus utilisées.
tasting note
|
v
38 TypeSafe answers
|-- 29 score questions x 2 columns = 58
| expected rubric level + answer uncertainty
`-- 9 noul questions x 1 column = 9
probability true
|
v
67 numeric columns --> CatBoost --> predicted critic score
held-out RMSE: 1.77 pointsUne réponse score devient deux colonnes : le niveau moyen vers lequel pointe la réponse, et son étalement autour de cette moyenne. Une réponse noul est une probabilité, donc une seule colonne.
Les données sont 2 000 critiques de vin : une note de dégustation en entrée, le score du critique sur une échelle de 80 à 100 en sortie. Le RMSE mesure l’erreur de prédiction en points de score de critique, les écarts plus grands comptant davantage, et plus bas c’est mieux. Chaque chiffre du tableau ci-dessous provient des 800 critiques que ni le modèle ni la boucle n’ont jamais vues.
| comment la note devient un score | RMSE |
|---|---|
| prédire le score moyen des lignes d’entraînement | 3.09 |
| le même CatBoost, lisant la note comme des comptages de mots | 2.47 |
| demander le score lui-même à TypeSafe, remis à l’échelle et décalé | 2.15 |
| 18 questions issues d’un seul appel de proposition, sans boucle | 1.87 |
| 38 questions après cinq tours de la boucle | 1.77 |
Les deux dernières lignes sont la boucle. Un appel de proposition, sans rien sur quoi s’appuyer encore, atteint 1.87. Quatre tours de plus à lire ses propres pires prédictions atteignent 1.77. L’essentiel du gain tient à ce premier appel, et ce qu’y ajoutent les quatre tours suivants est mesuré plus bas.
from __future__ import annotations
import json
import os
import random
import textwrap
import urllib.request
from concurrent.futures import ThreadPoolExecutor
from pathlib import Path
from time import perf_counter
from typing import NamedTuple
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
from catboost import CatBoostRegressor
from cooksafe import JsonCache, make_playground_link
from IPython.display import Markdown, display
from typesafe_sdk import Noul, NoulCriteria, Score, TypeSafeClient
matplotlib.use("Agg") # headless render
TYPESAFE_MODEL = "jev-1.12"
FOLDS, REPEATS = 5, 3 # repeats steady the error at this sample size
CATBOOST = dict(
iterations=400,
depth=4,
learning_rate=0.05,
loss_function="RMSE",
verbose=0,
random_seed=0,
thread_count=1,
allow_writing_files=False,
)
client = TypeSafeClient(
# keyless kernels replay the cache
api_key=os.environ.get("TYPESAFE_API_KEY", "cache-only"),
base_url=os.environ.get("TYPESAFE_ENDPOINT"),
timeout=120.0,
)
json_cache = JsonCache(Path("json_cache.json"))
# ----------------------------------------------------------------- the specification
INTENSITY_LEVELS = [
"Not present in this note at all",
"Barely present - mentioned once, in passing",
"Present at a moderate level",
"Present strongly - the note dwells on it",
"Dominant - the note is largely about this",
]
PRESENCE_CRITERIA = NoulCriteria(
true="The note states this or clearly implies it",
false="The note gives no indication of this",
)
# Asking for the score outright: ten quality bands, rescaled onto the 80-100 critic scale.
SCORE_LEVELS = [
"Faulty or unpleasant - the note is mostly criticism",
"Barely acceptable - drinkable, with nothing to recommend it",
"Simple and sound - correct, plain, forgettable",
"Pleasant everyday wine - some appeal, little depth",
"Good - clear varietal character, well made",
"Very good - balanced, with something to say",
"Excellent - complex and structured",
"Outstanding - depth and length, built to age",
"Superb - among the best of its type",
"Profound - the note treats it as exceptional",
]
# Structured output requires every property in `required`, so unused fields come back empty.
PROPOSAL_SCHEMA = {
"type": "object",
"properties": {
"actions": {
"type": "array",
"items": {
"type": "object",
"properties": {
"op": {"type": "string", "enum": ["add", "revise", "drop"]},
"target": {"type": "string"},
"name": {"type": "string"},
"kind": {"type": "string", "enum": ["intensity", "presence"]},
"question": {"type": "string"},
},
"required": ["op", "target", "name", "kind", "question"],
"additionalProperties": False,
},
}
},
"required": ["actions"],
"additionalProperties": False,
}
PROPOSALS = 18 # actions the proposer may return per round
# The one string that knows this is about wine. Point it at your own label and text.
PROPOSER_TASK = f"""You are designing numeric features for a gradient-boosting model that
predicts the score a wine critic gave (an integer from 80 to 100) from the tasting note alone.
The model sees nothing but the features you design.
Return up to {PROPOSALS} actions. Each action is one of:
- {{"op": "add", "target": "", "name": ..., "kind": ..., "question": ...}}
A new feature.
- {{"op": "revise", "target": <name of an existing feature>, "name": ..., "kind": ...,
"question": ...}}
Replace that feature's question with better wording. Use this when a feature measures the
right thing badly: too narrow, too vague, or worded so nearly every note answers the same.
- {{"op": "drop", "target": <name of an existing feature>, "name": "", "kind": "intensity",
"question": ""}}
Remove a feature that is not earning its place.
`kind` is "intensity" for something with a degree, or "presence" for a yes/no fact.
`question` is what gets asked about one tasting note.
An "intensity" question is graded against this fixed five-level rubric, so word it so that the
levels make sense:
{chr(10).join(f" {i}. {level}" for i, level in enumerate(INTENSITY_LEVELS))}
A "presence" question is answered as the probability that it is true of the note.
Good features can be judged from the note's own words, vary from note to note, and carry
information about quality that the other features do not. Reviewers describe structure, fruit,
oak, length, complexity, and drinkability, and they also signal quality through word choice."""
class Split(NamedTuple):
"""The rows, their labels, and which half the loop is allowed to read."""
notes: list[str]
scores: np.ndarray
dev: np.ndarray
test: np.ndarray
# ----------------------------------------------------------------- the data
WINEMAG_CSV = (
"https://huggingface.co/datasets/GroNLP/ik-nlp-22_winemag/resolve/"
"90eb39f35fc64e556fc17f06d4137a4a69ec3297/train.csv"
)
@json_cache
def load_slice(n_dev: int, n_test: int, seed: int) -> dict:
"""Fetch the pinned CSV and take a seeded sample of note + score, one row per note."""
import csv
import io
request = urllib.request.Request(
WINEMAG_CSV, headers={"User-Agent": "typesafe-cookbook/1.0"}
)
with urllib.request.urlopen(request, timeout=300) as response:
text = response.read().decode()
rows, seen = [], set()
for row in csv.DictReader(io.StringIO(text)): # a few notes repeat verbatim
if not row["description"] or not row["points"] or row["description"] in seen:
continue
seen.add(row["description"])
rows.append((row["description"], float(row["points"])))
random.Random(seed).shuffle(rows)
picked = rows[: n_dev + n_test]
return {"notes": [r[0] for r in picked], "points": [r[1] for r in picked]}
def example_rows(split: Split, out_of_fold: np.ndarray | None, n: int) -> list[int]:
"""Select representative dev rows for a proposer round."""
dev = split.dev
if out_of_fold is None:
ranked = dev[np.argsort(split.scores[dev], kind="stable")]
return [int(ranked[round(q * (len(ranked) - 1))]) for q in np.linspace(0, 1, n)]
error = np.abs(split.scores[dev] - out_of_fold)
order = np.argsort(-error, kind="stable")
worst = [int(dev[i]) for i in order[: n // 2]]
best = [int(dev[i]) for i in order[len(order) - (n - n // 2) :]]
return worst + best
def example_block(
rows: list[int],
split: Split,
out_of_fold: np.ndarray | None,
previous: np.ndarray | None = None,
) -> str:
"""Format selected rows for the proposer."""
if out_of_fold is None:
head = "Example notes, with the score each one was given:"
body = [f"- scored {split.scores[r]:.0f}: {split.notes[r]}" for r in rows]
return head + "\n" + "\n".join(body)
head = (
"Dev notes, worst-predicted first. The first half is where your current questions "
"miss by the most and the second half is where they are already right, so what "
"separates the halves is what the questions have not captured."
)
if previous is not None:
head += (
" Each line also carries what the previous round predicted, so you can see which "
"notes your last batch of questions moved."
)
body = []
for r in rows:
line = f"- scored {split.scores[r]:.0f}, predicted {out_of_fold[r]:.1f}"
if previous is not None:
line += f" (last round {previous[r]:.1f})"
body.append(f"{line}: {split.notes[r]}")
return head + "\n" + "\n".join(body)
def load_split(n_dev: int, n_test: int, seed: int = 0) -> Split:
# keyword, because the cache key is the function name plus how each argument was spelled
data = load_slice(n_dev, n_test, seed=seed)
return Split(
notes=data["notes"],
scores=np.array(data["points"]),
dev=np.arange(n_dev),
test=np.arange(n_dev, n_dev + n_test),
)
# ----------------------------------------------------------------- step 1: propose
def proposal_prompt(examples: str, feedback: str, accepted: list[dict]) -> str:
parts = [PROPOSER_TASK, "\n" + examples]
if accepted:
parts.append(
"\nThe features you have now. `add` must not duplicate one of these; `revise` and "
"`drop` refer to one by name:\n"
+ "\n".join(
f"- {f['name']} ({f['kind']}): {f['question']}" for f in accepted
)
)
if feedback:
parts.append("\nHow the model did with those features:\n" + feedback)
return "\n".join(parts)
@json_cache
def propose(model: str, round_index: int, prompt: str) -> dict:
"""One proposal call. Every number in `prompt` is rounded so a replay hits the cache."""
if model.startswith("claude"):
import anthropic
response = anthropic.Anthropic(
api_key=os.environ.get("ANTHROPIC_API_KEY", "cache-only")
).messages.create(
model=model,
max_tokens=16000,
output_config={
"effort": "medium",
"format": {"type": "json_schema", "schema": PROPOSAL_SCHEMA},
},
messages=[{"role": "user", "content": prompt}],
)
body = next(block.text for block in response.content if block.type == "text")
usage = [response.usage.input_tokens or 0, response.usage.output_tokens or 0]
else:
from openai import OpenAI
response = OpenAI(
api_key=os.environ.get("OPENAI_API_KEY", "cache-only")
).chat.completions.create(
model=model,
reasoning_effort="high",
max_completion_tokens=16000,
response_format={"type": "json_object"},
messages=[
{
"role": "user",
"content": prompt
+ "\n\nReply with JSON matching this schema:\n"
+ json.dumps(PROPOSAL_SCHEMA),
}
],
)
body = response.choices[0].message.content
usage = [response.usage.prompt_tokens, response.usage.completion_tokens]
return {"actions": json.loads(body)["actions"][:PROPOSALS], "usage": usage}
def slug(name: str, taken: set[str]) -> str:
"""Names become question ids and column labels, so keep them plain and unique."""
base = (
"".join(c if c.isalnum() else "_" for c in name.lower()).strip("_") or "feature"
)
candidate, n = base, 2
while candidate in taken:
candidate, n = f"{base}_{n}", n + 1
return candidate
def to_candidates(actions: list[dict], accepted: list[dict], round_index: int) -> tuple:
"""Split a round's actions into screenable candidates and a list of names to drop."""
live = {f["name"] for f in accepted}
drops = [a["target"] for a in actions if a["op"] == "drop" and a["target"] in live]
replacing = {
a["target"] for a in actions if a["op"] == "revise" and a["target"] in live
}
# a revision may keep the name it replaces, since that feature is on its way out
taken, candidates = live - replacing, []
for action in actions:
if action["op"] == "drop":
continue
if action["op"] == "revise" and action["target"] not in live:
continue # a revision of something that is not there
name = slug(action["name"], taken)
taken.add(name)
candidates.append(
{
"id": f"{name}@{round_index}", # unique, so earlier rounds keep their columns
"name": name,
"kind": action["kind"],
"question": action["question"],
"replaces": action["target"] if action["op"] == "revise" else "",
}
)
return candidates, drops
# ----------------------------------------------------------------- step 2: answer
def feature_questions(features: list[dict]) -> dict:
questions = {}
for feature in features:
if feature["kind"] == "intensity":
questions[feature["name"]] = Score(
instructions=feature["question"], criteria=INTENSITY_LEVELS
)
else:
questions[feature["name"]] = Noul(
instructions=feature["question"], criteria=PRESENCE_CRITERIA
)
return questions
@json_cache
def answer(model: str, note: str, features_json: str) -> dict:
"""One request per note; every question of the round rides it. Keeps every probability."""
features = json.loads(features_json)
started = perf_counter()
response = client.system_one(
state=note, questions=feature_questions(features), model=model
)
raw = {}
for feature in features:
got = response.answers[feature["name"]]
if feature["kind"] == "intensity":
raw[feature["name"]] = [
got.probabilities.get(i, 0.0) for i in range(len(INTENSITY_LEVELS))
]
else:
raw[feature["name"]] = [got.noul]
return {
"raw": raw,
"seconds": round(perf_counter() - started, 2),
"input_tokens": response.usage.input_tokens or 0,
"output_tokens": response.usage.output_tokens or 0,
}
def featurize(notes: list[str], features: list[dict]) -> dict:
"""Answer one question set for many notes: one request each, eight in flight."""
payload = json.dumps(features, sort_keys=True)
with ThreadPoolExecutor(max_workers=8) as pool:
results = list(
pool.map(lambda note: answer(TYPESAFE_MODEL, note, payload), notes)
)
return {
f["name"]: np.array([r["raw"][f["name"]] for r in results], dtype=float)
for f in features
}
def encode(feature: dict, probabilities: np.ndarray, mode: str) -> list[tuple]:
"""Turn one question's probabilities into named columns."""
name = feature["name"]
if feature["kind"] == "presence":
return [(name, probabilities[:, 0])] # one number is all there is
levels = np.arange(probabilities.shape[1])
mean = probabilities @ levels
if mode == "mean":
return [(name, mean)]
if mode == "mean_spread":
variance = probabilities @ (levels**2) - mean**2
return [(name, mean), (f"{name}_sd", np.sqrt(np.clip(variance, 0, None)))]
return [(f"{name}_p{i}", probabilities[:, i]) for i in levels]
def design(features: list[dict], answers_for: dict, mode: str) -> tuple:
"""Stack every feature's columns into one matrix, plus a label per column."""
columns, labels = [], []
for feature in features:
for label, column in encode(feature, answers_for[feature["id"]], mode):
columns.append(column)
labels.append(label)
return np.column_stack(columns), labels
def plain(features: list[dict]) -> list[dict]:
"""What goes on the wire and into the cache key: no id, no bookkeeping."""
return [
{"name": f["name"], "kind": f["kind"], "question": f["question"]}
for f in features
]
# ----------------------------------------------------------------- step 3: fit
def rmse(y: np.ndarray, p: np.ndarray) -> float:
return float(np.sqrt(np.mean((y - p) ** 2)))
def spearman(a: np.ndarray, b: np.ndarray) -> float:
"""Rank correlation: does the model order the wines the way the critic did?"""
ranks = (
np.argsort(np.argsort(a)).astype(float),
np.argsort(np.argsort(b)).astype(float),
)
return float(np.corrcoef(*ranks)[0, 1])
def folds(y: np.ndarray, k: int, seed: int) -> list[np.ndarray]:
"""Label-stratified k-fold: sort by the label with a seeded tiebreak, then deal off the top."""
rng = np.random.default_rng(seed)
order = np.lexsort((rng.random(len(y)), y))
return [np.sort(order[i::k]) for i in range(k)]
def cross_validate(X: np.ndarray, y: np.ndarray) -> tuple[np.ndarray, float]:
out_of_fold = np.zeros((REPEATS, len(y)))
for repeat in range(REPEATS):
for fold in folds(y, FOLDS, seed=repeat):
train = np.setdiff1d(np.arange(len(y)), fold)
model = CatBoostRegressor(**CATBOOST).fit(X[train], y[train])
out_of_fold[repeat, fold] = model.predict(X[fold])
scores = [rmse(y, out_of_fold[repeat]) for repeat in range(REPEATS)]
return out_of_fold.mean(axis=0), float(np.mean(scores))
def importances(X: np.ndarray, y: np.ndarray) -> np.ndarray:
return CatBoostRegressor(**CATBOOST).fit(X, y).get_feature_importance()
def paired_gain(y: np.ndarray, before: np.ndarray, after: np.ndarray) -> tuple:
"""Bootstrap the paired held-out RMSE change."""
squared = ((y - before) ** 2, (y - after) ** 2)
rng = np.random.default_rng(0)
drawn = []
for _ in range(2000):
rows = rng.integers(0, len(y), len(y))
drawn.append(
np.sqrt(squared[1][rows].mean()) - np.sqrt(squared[0][rows].mean())
)
drawn = np.array(drawn)
return (
rmse(y, after) - rmse(y, before),
float(np.percentile(drawn, 2.5)),
float(np.percentile(drawn, 97.5)),
)
def fit_predict(X: np.ndarray, split: Split) -> np.ndarray:
model = CatBoostRegressor(**CATBOOST).fit(X[split.dev], split.scores[split.dev])
return model.predict(X[split.test])
def fit_predict_text(split: Split) -> np.ndarray:
"""The reference arm: the same model, handed the note instead of the columns."""
from catboost import Pool
raw = np.array([[note] for note in split.notes], dtype=object)
model = CatBoostRegressor(**CATBOOST).fit(
Pool(raw[split.dev], split.scores[split.dev], text_features=[0])
)
return model.predict(Pool(raw[split.test], text_features=[0]))
def evaluate(
features: list[dict], answers_for: dict, split: Split, mode: str
) -> tuple[np.ndarray, float]:
"""Cross-validated error on the dev rows for one candidate question set."""
X, _ = design(features, answers_for, mode)
return cross_validate(X[split.dev], split.scores[split.dev])
def swap_in(accepted: list[dict], feature: dict) -> list[dict] | None:
"""The accepted set with `feature` in place of the one it revises, or None if it is gone."""
at = next(
(i for i, f in enumerate(accepted) if f["name"] == feature["replaces"]), None
)
if at is None:
return None
trial = list(accepted)
trial[at] = {k: feature[k] for k in ("id", "name", "kind", "question")}
return trial
def try_change(
trial: list[dict],
accepted: list[dict],
cv: float,
answers_for: dict,
split: Split,
mode: str,
tolerance: float,
) -> tuple[list[dict], float, str, bool]:
"""Refit with the change and keep it only if the dev error improves. No API calls."""
_, cv_trial = evaluate(trial, answers_for, split, mode)
if cv_trial <= cv + tolerance:
return trial, cv_trial, f"CV {cv:.3f} -> {cv_trial:.3f}", True
return accepted, cv, f"would cost {cv_trial - cv:+.3f}", False
def owner_of(label: str, features: list[dict]) -> dict:
"""Which feature a column label belongs to - encodings suffix the name."""
exact = next((f for f in features if f["name"] == label), None)
if exact:
return exact
return next(f for f in features if label.startswith(f["name"] + "_"))
def importance_per_feature(
features: list[dict], labels: list[str], column_importances: np.ndarray
) -> dict:
"""Sum each question's CatBoost column importances.
Intensity questions can produce multiple model columns. Combining their normalized
importances gives one percentage share per question.
"""
total = {f["name"]: 0.0 for f in features}
for label, column_importance in zip(labels, column_importances):
total[owner_of(label, features)["name"]] += float(column_importance)
return total
def feedback_for(
history: list[float],
accepted: list[dict],
answers_for: dict,
split: Split,
mode: str,
out_of_fold: np.ndarray,
previous: np.ndarray | None,
) -> str:
"""The scoreboard the next proposal call reads. The notes themselves arrive separately,
through `example_block`. Numbers are rounded before they enter the prompt."""
X, labels = design(accepted, answers_for, mode)
dev, scores = split.dev, split.scores
by_name = importance_per_feature(accepted, labels, importances(X[dev], scores[dev]))
lines = ["Cross-validated RMSE in points so far, lower is better:"]
lines += [f" round {i + 1}: {v:.2f}" for i, v in enumerate(history)]
if previous is not None:
now, before = np.abs(scores[dev] - out_of_fold), np.abs(scores[dev] - previous)
better, worse = int((now < before - 0.1).sum()), int((now > before + 0.1).sum())
lines.append(
f"\nAgainst the previous round, {better} of the {len(dev)} dev notes are now "
f"predicted better by more than 0.1 points and {worse} are predicted worse."
)
lines.append(
"\nYour features, with importance as a percentage of the total and the spread of the "
"column across the dev rows. Low importance or low spread means the question is not "
"doing much; revise or drop it."
)
for feature in sorted(accepted, key=lambda f: -by_name.get(f["name"], 0.0)):
column = encode(feature, answers_for[feature["id"]], mode)[0][1]
lines.append(
f" {feature['name']} ({feature['kind']}): "
f"{by_name.get(feature['name'], 0.0):.1f}% importance, "
f"spread {column[dev].std():.2f}"
)
return "\n".join(lines)
# ----------------------------------------------------------------- the loop itself
class Discovery(NamedTuple):
"""Artifacts returned by the discovery loop."""
accepted: list[dict] # the question set it ended with
answers_for: dict # feature id -> (rows x levels) probabilities
snapshots: list[list[dict]] # the set as it stood at the end of each round
history: list[float] # dev CV error after each round
batches: list[tuple] # what each round sent, for the request table
journal: list[tuple] # every action and what became of it
def run_loop(
split: Split,
proposer: str,
rounds: int,
examples: int,
mode: str,
min_spread: float,
tolerance: float,
) -> Discovery:
"""Run the propose, answer, fit, and feedback loop."""
shown = example_rows(split, None, examples) # round 1 has nothing predicted yet
out_of_fold = previous = None
got_from = Discovery([], {}, [], [], [], [])
accepted, answers_for = got_from.accepted, got_from.answers_for
snapshots, history = got_from.snapshots, got_from.history
batches, journal = got_from.batches, got_from.journal
feedback = ""
for round_index in range(1, rounds + 1):
block = example_block(shown, split, out_of_fold, previous)
actions = propose(
proposer, round_index, proposal_prompt(block, feedback, accepted)
)["actions"]
keep, drops = to_candidates(actions, accepted, round_index)
if keep: # one request per row, carrying every question this round proposed
batches.append((round_index, plain(keep)))
answers = featurize(split.notes, plain(keep))
for feature in keep:
answers_for[feature["id"]] = answers[feature["name"]]
for (
feature
) in keep: # an add goes in; importance says later whether it earned it
if feature["replaces"]:
continue
column = encode(feature, answers_for[feature["id"]], mode)[0][1]
flat = float(column[split.dev].std()) < min_spread
journal.append(
(round_index, "flat" if flat else "add", feature["name"], "")
)
if not flat:
accepted.append(
{k: feature[k] for k in ("id", "name", "kind", "question")}
)
_, cv = evaluate(accepted, answers_for, split, mode)
trial_args = (answers_for, split, mode, tolerance)
for feature in [f for f in keep if f["replaces"]]: # every revision is tried
trial = swap_in(accepted, feature)
if trial is None: # it revises something an earlier round already dropped
journal.append(
(round_index, "stale", feature["name"], "target is gone")
)
continue
accepted[:], cv, note, took = try_change(trial, accepted, cv, *trial_args)
what = "revise" if took else "reject"
journal.append(
(
round_index,
what,
feature["name"],
f"was {feature['replaces']}, {note}",
)
)
for name in drops: # and so is every drop
trial = [f for f in accepted if f["name"] != name]
if not trial:
continue
accepted[:], cv, note, took = try_change(trial, accepted, cv, *trial_args)
journal.append((round_index, "drop" if took else "keep", name, note))
previous, (out_of_fold, cv) = (
out_of_fold,
evaluate(accepted, answers_for, split, mode),
)
history.append(cv)
snapshots.append(list(accepted))
feedback = feedback_for(
history, accepted, answers_for, split, mode, out_of_fold, previous
)
# next round reads the rows these questions get most wrong, and as many they get right
shown = example_rows(split, out_of_fold, examples)
report(round_index, keep, drops, journal, accepted, cv)
return got_from
def report(
round_index: int,
keep: list[dict],
drops: list[str],
journal: list[tuple],
accepted: list[dict],
cv: float,
) -> None:
"""One block per round: the counts, the names it added, then everything with a number."""
revised = sum(1 for f in keep if f["replaces"])
print(
f"round {round_index}: {len(keep) - revised} add, {revised} revise, "
f"{len(drops)} drop"
)
this_round = [j for j in journal if j[0] == round_index]
added = [name for _, what, name, _ in this_round if what == "add"]
if added:
print(
textwrap.fill(
", ".join(added),
88,
initial_indent=" added ",
subsequent_indent=" " * 10,
)
)
for _, what, name, note in this_round: # everything carrying a number of its own
if what != "add":
print(f" {what:<7}{name:<34}{note}")
print(f" -> {len(accepted)} features, dev CV RMSE {cv:.3f}\n")
# ----------------------------------------------------------------- asking for the score
@json_cache
def ask_score(model: str, note: str) -> dict:
"""One `Score` over ten quality bands, read as a level and rescaled to 80-100."""
response = client.system_one(
state=note,
questions={
"quality": Score(
instructions=(
"Judging only by what this tasting note says, how good is the wine?"
),
criteria=SCORE_LEVELS,
)
},
model=model,
)
got = response.answers["quality"]
top = len(SCORE_LEVELS) - 1
expected = sum(k * v for k, v in got.probabilities.items())
return {
# level 0 is the bottom of the critic's scale, level 9 the top
"expected": 80.0 + 20.0 * expected / top,
"picked": 80.0 + 20.0 * got.score / top,
"input_tokens": response.usage.input_tokens or 0,
"output_tokens": response.usage.output_tokens or 0,
}
# ----------------------------------------------------------------- charts
SURFACE, INK, INK2, MUTED = "#fcfcfb", "#0b0b0b", "#52514e", "#898781"
GRID, AXIS, BLUE, ORANGE = "#e1e0d9", "#c3c2b7", "#2a78d6", "#eb6834"
def style(ax) -> None:
ax.set_facecolor(SURFACE)
for side in ("top", "right"):
ax.spines[side].set_visible(False)
for side in ("left", "bottom"):
ax.spines[side].set_color(AXIS)
ax.tick_params(colors=MUTED, labelcolor=INK2, labelsize=9)
ax.set_axisbelow(True)
def polarity(feature: dict, answers_for: dict, split: Split) -> float:
"""Rank correlation between a question's answer and the critic score, on the dev rows.
Positive means a higher answer goes with a better review, negative the opposite. It is
what orders the rows of the feature map, so the map reads as a gradient that flips.
"""
column = encode(feature, answers_for[feature["id"]], "mean")[0][1]
return spearman(column[split.dev], split.scores[split.dev])
def reviews_heatmap(
plt,
questions: list[dict],
answers_for: dict,
split: Split,
rows: tuple,
):
"""Compare held-out reviews across the discovered questions, best-signal first.
Rows arrive sorted from the questions that rise with the score to the ones that fall with
it, so a row above the divider shades left to right and a row below it shades right to
left.
"""
def value_of(feature: dict, row: int) -> float:
return float(encode(feature, answers_for[feature["id"]], "mean")[0][1][row])
signs = [polarity(question, answers_for, split) for question in questions]
flip = next((i for i, s in enumerate(signs) if s < 0), len(questions))
raw = np.array(
[[value_of(question, row) for row in rows] for question in questions]
)
normalized = np.array(
[
values / (4 if question["kind"] == "intensity" else 1)
for question, values in zip(questions, raw)
]
)
cmap = matplotlib.colors.LinearSegmentedColormap.from_list(
"typesafe_heat", [SURFACE, "#f7c7ad", ORANGE]
)
fig, ax = plt.subplots(
figsize=(9.5, 1.8 + 0.58 * len(questions)), facecolor=SURFACE
)
image = ax.imshow(normalized, aspect="auto", cmap=cmap, vmin=0, vmax=1)
row_labels = []
for question, sign in zip(questions, signs):
kind = "score" if question["kind"] == "intensity" else "noul"
prefix = f"{sign:+.2f} ({kind}) "
lines = textwrap.wrap(
" ".join(question["question"].split()),
width=52,
max_lines=2,
placeholder="...",
break_long_words=False,
break_on_hyphens=False,
)
row_labels.append(prefix + (f"\n{' ' * len(prefix)}").join(lines))
column_labels = [
f"#{i}\n{split.scores[row]:.0f} points\n{' '.join(split.notes[row].split())[:15]}..."
for i, row in enumerate(rows, 1)
]
ax.set_yticks(np.arange(len(questions)), row_labels)
ax.set_xticks(np.arange(len(rows)), column_labels)
ax.tick_params(
axis="x", top=True, labeltop=True, bottom=False, labelbottom=False, pad=8
)
ax.tick_params(axis="y", labelsize=8.5)
for side in ax.spines.values():
side.set_visible(False)
ax.set_xticks(np.arange(-0.5, len(rows), 1), minor=True)
ax.set_yticks(np.arange(-0.5, len(questions), 1), minor=True)
ax.grid(which="minor", color=SURFACE, linewidth=2)
ax.tick_params(which="minor", bottom=False, left=False)
for i, question in enumerate(questions):
for j, value in enumerate(raw[i]):
label = (
f"{value:.1f}" if question["kind"] == "intensity" else f"{value:.2f}"
)
color = SURFACE if normalized[i, j] > 0.58 else INK2
ax.text(j, i, label, ha="center", va="center", color=color, fontsize=8)
# the line where the questions stop rising with the score and start falling with it
if 0 < flip < len(questions):
ax.axhline(flip - 0.5, color=INK, linewidth=1.2)
ax.annotate(
"a higher answer means a worse review, below this line",
(len(rows) - 0.5, flip - 0.5),
xytext=(-4, 5),
textcoords="offset points",
va="bottom",
ha="right",
color=INK2,
fontsize=8.5,
)
colorbar = fig.colorbar(image, ax=ax, fraction=0.025, pad=0.025)
colorbar.set_ticks([0, 0.5, 1])
colorbar.set_label("normalized answer", color=INK2, fontsize=8.5)
colorbar.ax.tick_params(labelsize=8, colors=INK2)
fig.suptitle(
"Every question, on five held-out reviews from worst to best",
x=0.01,
y=0.995,
ha="left",
color=INK,
fontsize=11,
)
fig.text(
0.01,
0.972,
"sorted by how the answer moves with the score, so each row above the line shades "
"left to right and each row below it shades the other way",
color=MUTED,
fontsize=9,
)
fig.text(
0.01,
0.005,
"Row labels lead with the rank correlation between that question's answer and the "
"critic score. Cell text is each question's native scale: score 0-4, noul 0-1.",
color=MUTED,
fontsize=8.5,
)
return fig
def rounds_chart(
plt, curve: list[tuple], history: list[float], n_test: int, gain: tuple
):
"""Dev error and held-out error per round. The trend is the point, not the gap."""
rounds = list(range(1, len(curve) + 1))
values = [v for _, v in curve]
fig, ax = plt.subplots(figsize=(7, 3.9), facecolor=SURFACE)
style(ax)
ax.grid(axis="y", color=GRID, linewidth=0.8)
# each dev fold trains on four fifths of the rows, so the dev line sits the higher of the two
ax.fill_between(rounds, history, values, color=GRID, alpha=0.75, linewidth=0)
ax.plot(
rounds,
history,
marker="o",
color=BLUE,
linewidth=2,
linestyle="--",
label="dev, cross-validated - what the loop optimises",
)
ax.plot(
rounds,
values,
marker="o",
color=ORANGE,
linewidth=2,
label="held out - what that actually buys",
)
# label each point on the outside of the pair, so neither line crowds its own numbers
for x, dev_value, test_value in zip(rounds, history, values):
for value, other in ((dev_value, test_value), (test_value, dev_value)):
ax.annotate(
f"{value:.2f}",
(x, value),
textcoords="offset points",
xytext=(0, 8 if value >= other else -16),
ha="center",
color=INK2,
fontsize=8.5,
)
ax.set_xticks(
rounds, [f"round {x}\n{n} features" for x, (n, _) in zip(rounds, curve)]
)
ax.set_ylabel("RMSE in points (lower is better)", color=INK2, fontsize=9)
# tight around the two lines: the whole finding lives inside 0.15 of a point
low, high = min(values + history), max(values + history)
ax.set_ylim(low - 0.10, high + 0.05)
difference, low_ci, high_ci = gain
ax.set_title(
f"{len(rounds)} rounds of the loop, scored on {n_test} held-out reviews",
loc="left",
color=INK,
fontsize=11,
pad=20,
)
# the number the chart is really about: is the held-out move bigger than the noise?
ax.text(
0,
1.015,
f"round 1 to round {len(rounds)}, held out: {difference:+.3f} points, "
f"95% CI [{low_ci:+.3f}, {high_ci:+.3f}]",
transform=ax.transAxes,
color=MUTED,
fontsize=9,
)
ax.legend(frameon=False, labelcolor=INK2, fontsize=9, loc="lower left")
return figConfiguration
pip install anthropic openai catboost numpy matplotlib ipython 'cooksafe>=0.2.0,<0.3.0'puis définis TYPESAFE_API_KEY et ANTHROPIC_API_KEY. Chaque appel API est mis en cache
dans json_cache.json, livré avec le cookbook, donc un nouveau rendu rejoue ces chiffres
sans rien appeler. Supprime-le pour réexécuter en direct. Les chiffres proviennent de
TypeSafe jev-1.12 et claude-sonnet-5 le 2026-08-03. propose() a une seconde branche
pour gpt-5.6-luna, qui n’a pas été exécutée.
La première cellule de code est toute l’implémentation : appels API, encodages, métriques, style des graphiques. Elle est là pour que ce fichier tourne tout seul, et le site de docs la replie. Saute-la à la première lecture – la recette commence juste en dessous.
N_DEV, N_TEST = 1200, 800 # the loop reads dev labels only; test is scored once
ROUNDS = 5 # a round answers questions for all 2,000 rows: 2,000 requests
PROPOSER = "claude-sonnet-5" # or "gpt-5.6-luna"; the cache holds the Anthropic run
EXAMPLES = 60 # dev notes the proposer reads per round, half of them its worst misses
MIN_SPREAD = 0.05 # a column this flat cannot separate anything, so it is not kept
CHANGE_TOLERANCE = 0.0 # a revision or drop has to improve dev error, not just not hurt
ENCODING = "mean_spread" # a score answer becomes two columns: its mean and spread
split = load_split(N_DEV, N_TEST, seed=0)
NOTES, SCORES, DEV, TEST = split.notes, split.scores, split.dev, split.test
print(
f"{len(DEV)} dev rows, {len(TEST)} held out; scores run "
f"{SCORES.min():.0f}-{SCORES.max():.0f}, mean {SCORES.mean():.2f}, sd {SCORES.std():.2f}"
)
print(f"\none of the notes:\n{NOTES[0]}")1200 dev rows, 800 held out; scores run 80-98, mean 88.73, sd 3.17
one of the notes:
A Champagne that is very much wine. The structure and the richness are just right for a food wine, showing ripe acidity, flavors of plums and apricots, and balancing these primary fruits with a dense, complex structure that takes in yeast, maturity and a tight apple skin finish.La boucle relit sans cesse les mêmes 1 200 des 2 000 lignes (les lignes de développement), et garde une question quand elle aide à prédire ces 1 200 scores. Évaluer sur les mêmes lignes mesurerait surtout à quel point la boucle s’y est ajustée, donc les 800 autres sont mises de côté et évaluées une seule fois, à la fin.
Deux types de question
Une question proposée est de l’un de deux types, et le type décide quel nombre revient.
intensitydevient unScore, pour tout ce qui se mesure en degrés. Ses cinq niveaux sont imprimés ci-dessous, et la colonne est le niveau moyen, donc une note qui se situe entre « moderate » et « strongly » ressort entre les deux.presencedevient unNoul, pour un fait oui/non comme le fait qu’un défaut soit nommé. La colonne est cette unique probabilité.
La méthode
questions <- {}
repeat for each round:
notes <- round 1 ? 60 dev notes across the score range
: the 30 worst-predicted dev notes + the 30 best,
each with its score, this prediction and the last
actions <- LLM(brief, questions, notes, importance and error so far)
answers[q] <- TypeSafe(note, all new questions of this round) for every row
for each added q: keep it unless its column is flat
for each revised q: refit; keep the change only if dev error drops
for each dropped q: refit; drop it only if dev error drops
out_of_fold <- k-fold CatBoost on the columns # judges, and picks next round's notesAucune question n’est filtrée avant d’avoir reçu sa réponse. Toutes les questions d’un tour partent dans la même requête, donc une question de plus ne coûte aucune requête supplémentaire. Une question qui ne s’applique qu’à une ligne sur dix semblera inutile dans les 60 notes que lit le proposeur, et restera peut-être la colonne la plus utile de l’ensemble.
La validation croisée k-fold consiste à découper les lignes de développement en k parties et à prédire chaque partie avec un modèle entraîné sur les autres. Ces prédictions remplissent trois rôles : elles jugent chaque révision et chaque suppression, elles choisissent les notes que lit le tour suivant, et elles indiquent au proposeur lesquelles de ses questions ont aidé, selon de combien elles ont bougé depuis le tour précédent.
print("every intensity question is graded on these five levels:\n")
for i, level in enumerate(INTENSITY_LEVELS):
print(f" {i}. {level}")
print("\nevery presence question is judged true or false against these:\n")
print(f" true: {PRESENCE_CRITERIA['true']}")
print(f" false: {PRESENCE_CRITERIA['false']}")
print("\nthe brief the proposer works from:\n")
print("\n".join(PROPOSER_TASK.splitlines()[:6]) + "\n ...")every intensity question is graded on these five levels:
0. Not present in this note at all
1. Barely present - mentioned once, in passing
2. Present at a moderate level
3. Present strongly - the note dwells on it
4. Dominant - the note is largely about this
every presence question is judged true or false against these:
true: The note states this or clearly implies it
false: The note gives no indication of this
the brief the proposer works from:
You are designing numeric features for a gradient-boosting model that
predicts the score a wine critic gave (an integer from 80 to 100) from the tasting note alone.
The model sees nothing but the features you design.
Return up to 18 actions. Each action is one of:
...La boucle autoresearch
run_loop exécute les cinq tours et imprime un bloc par tour. Une question ajoutée entre
directement : ses réponses ont déjà été récupérées, et son importance montrera plus tard si
elle valait la peine d’être posée. Une révision ou une suppression retire une colonne que
le modèle utilise déjà, donc chacune est d’abord essayée : réajuster avec le changement, et
ne le garder que si l’erreur de développement baisse. Un réajustement ne coûte aucun appel
API, donc essayer un changement et le rejeter est gratuit.
run = run_loop(
split, PROPOSER, ROUNDS, EXAMPLES, ENCODING, MIN_SPREAD, CHANGE_TOLERANCE
)
accepted, answers_for = run.accepted, run.answers_for
snapshots, history = run.snapshots, run.historyround 1: 18 add, 0 revise, 0 drop
added complexity, fruit_intensity, tannin_structure, acidity_intensity,
oak_intensity, finish_length, balance_harmony, aging_potential,
positive_superlative_language, negative_critical_language,
drinkability_easiness, body_richness, sweetness_level, texture_descriptors,
earthy_savory_notes, flaw_or_defect_mentioned,
single_vineyard_or_prestige_signal, varietal_blend_detail
-> 18 features, dev CV RMSE 1.903
round 2: 5 add, 3 revise, 3 drop
added power_concentration_language, flavor_distinctiveness, generic_fruit_language,
candied_artificial_flavor, rustic_authentic_character
reject oak_dominance was oak_intensity, would cost +0.005
revise negative_critical_language was negative_critical_language, CV 1.897 -> 1.894
revise single_vineyard_or_prestige_signalwas single_vineyard_or_prestige_signal, CV 1.894 -> 1.881
keep finish_length would cost +0.009
keep texture_descriptors would cost +0.001
keep varietal_blend_detail would cost +0.023
-> 23 features, dev CV RMSE 1.881
round 3: 7 add, 2 revise, 1 drop
added elegance_finesse_language, minerality_precision_language,
hedged_qualified_praise, underripe_green_character,
reviewer_overall_verdict_strength, unusual_or_funky_descriptor_valence,
botrytis_or_special_winemaking_signal
revise negative_critical_language was negative_critical_language, CV 1.868 -> 1.864
revise finish_quality was finish_length, CV 1.864 -> 1.861
keep candied_artificial_flavor would cost +0.014
-> 30 features, dev CV RMSE 1.861
round 4: 5 add, 2 revise, 3 drop
added excess_or_imbalance_signal, descriptive_detail_density,
critic_enthusiasm_confidence, savory_food_wine_seriousness,
note_overall_tone_positivity
revise rustic_authentic_character was rustic_authentic_character, CV 1.843 -> 1.838
reject hedged_qualified_praise was hedged_qualified_praise, would cost +0.014
keep botrytis_or_special_winemaking_signalwould cost +0.011
keep candied_artificial_flavor would cost +0.009
keep unusual_or_funky_descriptor_valencewould cost +0.010
-> 35 features, dev CV RMSE 1.838
round 5: 4 add, 2 revise, 8 drop
added structural_seriousness, youthful_tension_signal, surface_prettiness_vs_depth,
price_value_signal
reject unconventional_character_as_virtuewas rustic_authentic_character, would cost +0.010
revise flavor_distinctiveness was flavor_distinctiveness, CV 1.849 -> 1.843
keep candied_artificial_flavor would cost +0.002
keep botrytis_or_special_winemaking_signalwould cost +0.003
keep hedged_qualified_praise would cost +0.006
keep excess_or_imbalance_signal would cost +0.005
drop underripe_green_character CV 1.843 -> 1.840
keep unusual_or_funky_descriptor_valencewould cost +0.002
keep texture_descriptors would cost +0.001
keep generic_fruit_language would cost +0.000
-> 38 features, dev CV RMSE 1.840Le pointer vers tes propres données
PROPOSER_TASK est la seule chaîne qui mentionne le vin, et featurize() accepte n’importe
quelle liste de chaînes. Modifier ce brief change le prompt de proposition, et le prompt
fait partie de la clé de cache, donc l’exécution suivante rappelle l’API à chaque tour.
Le nombre de requêtes croît avec les lignes, pas avec les questions : une requête par ligne et par tour, donc 100 000 lignes font 100 000 requêtes par tour. Une révision compte comme une nouvelle question, donc elle coûte un passage supplémentaire sur chaque ligne. Augmente le pool de workers lentement. Huit suffisent déjà à atteindre une limite de débit sur une clé partagée.
Ce que voient les questions
Cinq critiques hors échantillon, une à chaque quart de la plage de scores, face à quinze des 38 questions : les huit meilleures questions score par importance, plus les sept meilleurs nouls.
Ces quinze lignes sont ensuite triées selon le sens dans lequel la réponse évolue avec le score du critique. Les questions dont la réponse monte avec le score viennent d’abord, celles dont la réponse baisse avec lui viennent après le séparateur. Donc de gauche à droite, de la pire critique à la meilleure, les réponses au-dessus du séparateur devraient grimper et celles en dessous chuter.
X, labels = design(accepted, answers_for, ENCODING)
column_importances = importances(X[DEV], SCORES[DEV])
# an encoding gives a feature more than one column, so add a feature's columns back up
feature_importances = importance_per_feature(accepted, labels, column_importances)
ranked = sorted(accepted, key=lambda f: -feature_importances[f["name"]])
score_questions = [f for f in ranked if f["kind"] == "intensity"][:8]
noul_questions = [f for f in ranked if f["kind"] == "presence"][:7]
# ordered by which way the answer moves with the score, so the map flips halfway down
heatmap_questions = sorted(
score_questions + noul_questions,
key=lambda f: -polarity(f, answers_for, split),
)
ordered_test = TEST[np.argsort(SCORES[TEST], kind="stable")]
positions = np.linspace(0, len(ordered_test) - 1, 5).round().astype(int)
review_rows = tuple(ordered_test[positions])
print("the five held-out heatmap columns:\n")
for i, row in enumerate(review_rows, 1):
excerpt = " ".join(NOTES[row].split())
print(f" {i}. {SCORES[row]:.0f} points: {excerpt[:100]}...")
fig = reviews_heatmap(plt, heatmap_questions, answers_for, split, review_rows)
display(fig)
plt.close(fig)the five held-out heatmap columns:
1. 80 points: Raw cherry and plum aromas are resiny and suggest wet cement. This is shearing and so jacked up with...
2. 86 points: A slight spritz brightens the mouthfeel of this lemony wine. Aromas are a bit musky, but flavors of ...
3. 89 points: This is a European-style Syrah, cofermented with 2% Viognier. It's soft and round, medium in body, a...
4. 91 points: From the producer's dry-farmed estate vineyard, and supported by small amounts of Merlot and Caberne...
5. 97 points: A thoroughly elegant, serious and yet immensely enjoyable wine that stays lively many days after ope...
Le tableau du haut de la page, calculé. Les cinq bras sont évalués une seule fois sur les
mêmes 800 lignes hors échantillon, et les trois premiers sautent la découverte de
caractéristiques. L’un prédit la moyenne des scores de développement et ne lit rien du tout
dans la note. L’un confie la note au même CatBoost via sa gestion text_features, qui la
transforme en comptages de mots. L’un demande le score lui-même à TypeSafe.
Ce troisième est un seul Score par ligne sur dix bandes de qualité, de « faulty or
unpleasant » jusqu’à « profound ». Dix parce que dix niveaux est le maximum qu’une question
Score accepte – onze revient comme une erreur serveur. Le niveau 0 correspond à 80 points
et le niveau 9 à 100. Répartir ainsi les bandes sur l’échelle ne suffit pas en soi, car rien
dans la question ne dit où se situent réellement les scores de cette publication. Chaque
réponse est donc ensuite décalée d’un seul offset, mesuré sur les scores de développement.
Cet offset est imprimé dans le libellé de ligne, et c’est la seule chose que ce raccourci
apprend des scores.
Spearman est une corrélation de rang, où 1.0 placerait les vins hors échantillon exactement dans l’ordre du critique. La ligne des comptages de mots est la gestion de texte propre à CatBoost, pas un pipeline de régression textuelle réglé. Tout cela relève d’un seul jeu de données et d’un seul passage de la boucle.
predicted = fit_predict(X, split)
text_predicted = fit_predict_text(split)
# ask TypeSafe for the score itself, one request per row
with ThreadPoolExecutor(max_workers=8) as pool:
direct = list(pool.map(lambda note: ask_score(TYPESAFE_MODEL, note), NOTES))
asked = np.array([d["expected"] for d in direct])
shift = float(SCORES[DEV].mean() - asked[DEV].mean()) # one number, from the dev labels
# what one proposal call gets you, before any feedback: the set round 1 ended with
first_round, _ = design(snapshots[0], answers_for, ENCODING)
print(f"{'arm':<46}{'RMSE':>7}{'spearman':>10}")
for label, p in (
("predict the mean of the dev rows", np.full(len(TEST), SCORES[DEV].mean())),
("the note as word counts, same CatBoost", text_predicted),
(f"ask for the score itself, shifted {shift:+.2f}", asked[TEST] + shift),
(
f"{len(snapshots[0])} questions from round 1, no loop",
fit_predict(first_round, split),
),
(f"{len(accepted)} questions after all {ROUNDS} rounds", predicted),
):
print(f"{label:<46}{rmse(SCORES[TEST], p):>7.3f}{spearman(SCORES[TEST], p):>10.3f}")arm RMSE spearman
predict the mean of the dev rows 3.088 -0.014
the note as word counts, same CatBoost 2.466 0.605
ask for the score itself, shifted -1.71 2.145 0.761
18 questions from round 1, no loop 1.869 0.778
38 questions after all 5 rounds 1.772 0.799Les tours d’autoresearch ont-ils aidé ?
Les deux lignes tracent l’erreur de l’ensemble de questions à la fin de chaque tour, en partant de la première proposition. La ligne en pointillés est l’erreur de développement validée de façon croisée, le nombre sur lequel se prend chaque décision d’acceptation et de rejet. La ligne pleine évalue le même ensemble de questions sur les lignes hors échantillon, que la boucle ne lit jamais. Chaque point est l’ensemble tel qu’il était à la clôture du tour, donc un tour qui n’a fait que réviser ou supprimer une question déplace quand même les deux lignes. La carte des caractéristiques dit ce que mesurent les questions ; l’erreur dit si les tours après la première proposition ont amélioré les prédictions.
L’axe est resserré : tout ce qu’il contient se joue dans un cinquième de point, et chaque raccourci du tableau ci-dessus se situe bien au-dessus du sommet. La ligne de développement reste au-dessus de la ligne hors échantillon tout du long, et c’est un effet de taille d’entraînement. Chaque pli de développement s’entraîne sur les quatre cinquièmes des lignes de développement, tandis que le nombre hors échantillon vient d’un modèle qui a reçu les 1 200. Les deux lignes bougent ensemble, donc le nombre de développement que la boucle suit reflète le nombre hors échantillon qu’elle ne voit jamais. L’intervalle sous le titre provient d’un rééchantillonnage des lignes hors échantillon, donc il indique si le mouvement du tour 1 au tour 5 est plus grand que le bruit dans 800 lignes.
curve, per_round = [], []
for features in snapshots:
X_round, _ = design(features, answers_for, ENCODING)
per_round.append(fit_predict(X_round, split))
curve.append((len(features), rmse(SCORES[TEST], per_round[-1])))
# the same held-out rows resampled 2,000 times, both arms scored on each resample
gain = paired_gain(SCORES[TEST], per_round[0], per_round[-1])
print(
f"round 1 -> round {ROUNDS} on the held-out rows: {gain[0]:+.3f} points, "
f"95% CI [{gain[1]:+.3f}, {gain[2]:+.3f}]"
)
fig = rounds_chart(plt, curve, history, len(TEST), gain)
display(fig)
plt.close(fig)round 1 -> round 5 on the held-out rows: -0.097 points, 95% CI [-0.147, -0.050]
La ligne hors échantillon baisse plus que celle du développement. Le tour 1 a écrit ses questions sans aucun retour sur lequel s’appuyer, et les quatre tours suivants valent 0,10 point sur les lignes hors échantillon, IC à 95 % [-0.147, -0.050].
Le tour 5 a proposé quatre ajouts, deux reformulations et huit suppressions, et a donné le premier nombre de développement qui ne s’améliorait pas. Il n’y a qu’une quantité limitée de choses à demander sur une note de 245 caractères, et au tour 5 les propositions avaient basculé de l’ajout de questions vers leur suppression.
kinds = {f["name"]: f["kind"] for f in accepted}
print("feature importance share: % of total CatBoost importance across all questions")
print(f"{'feature':<38}{'asked as':<10}{'importance share':>16}")
for name, importance_share in sorted(feature_importances.items(), key=lambda p: -p[1])[
:12
]:
kind = "score" if kinds[name] == "intensity" else "noul"
print(
f"{name[:36]:<38}{kind:<10}{importance_share:>8.1f}% "
f"{'#' * round(importance_share)}"
)
counts = f"{sum(1 for k in kinds.values() if k == 'intensity')} score"
counts += f", {sum(1 for k in kinds.values() if k == 'presence')} noul"
print(f"\nthe {len(accepted)} questions the loop kept: {counts}")
top = max(feature_importances, key=feature_importances.get)
print(
f'the question behind the top row:\n {top}: "{owner_of(top, accepted)["question"]}"'
)feature importance share: % of total CatBoost importance across all questions
feature asked as importance share
note_overall_tone_positivity score 17.4% #################
savory_food_wine_seriousness score 8.7% #########
positive_superlative_language score 8.4% ########
single_vineyard_or_prestige_signal noul 7.2% #######
descriptive_detail_density score 5.7% ######
elegance_finesse_language score 5.0% #####
complexity score 5.0% #####
aging_potential score 5.0% #####
balance_harmony score 2.9% ###
drinkability_easiness score 2.9% ###
critic_enthusiasm_confidence score 2.7% ###
flavor_distinctiveness score 2.6% ###
the 38 questions the loop kept: 29 score, 9 noul
the question behind the top row:
note_overall_tone_positivity: "Setting aside specific descriptors, how positive is the overall emotional tone and word choice of the note taken as a whole (warm, admiring language throughout vs. flat, neutral, or lukewarm phrasing)?"importance share est l’importance de caractéristique de CatBoost, normalisée pour que les
38 questions totalisent 100 %. Ce n’est pas une part de lignes, de questions ni d’exactitude
de prédiction. Une question score possède deux colonnes, une moyenne et un étalement, donc
ses deux importances de colonne sont additionnées avant l’impression du pourcentage.
note_overall_tone_positivity représente 17,4 % du total. La quatrième ligne est un noul :
le fait que la note nomme un seul vignoble ou un autre signal de prestige est un fait
oui/non, donc il a été posé comme tel.
Étapes suivantes
Cette exécution garde la boucle réduite. Extensions directes :
- Filtre un candidat avant de payer pour y répondre. Traite la question proposée elle-même comme l’état et pose des nouls à son sujet : peut-elle recevoir une réponse à partir du texte source, ne signifie-t-elle qu’une seule chose selon ses critères, s’applique-t-elle à la plupart des lignes, variera-t-elle d’une ligne à l’autre. N’envoie que les questions qui passent les quatre avec assez de confiance.
- Élague les caractéristiques corrélées. Mesure la corrélation entre les colonnes encodées sur les lignes de développement, regroupe les quasi-doublons, et garde la question la plus claire ou la plus importante de chaque groupe.
- Ajoute des baselines simples. Compare le TF-IDF, les comptages de caractères et d’autres caractéristiques structurelles seuls, puis ajoute-les aux colonnes découvertes pour mesurer ce que chacune apporte.
- Mélange les familles de proposeurs. Génère des lots candidats avec Anthropic, OpenAI, Google Gemini et des modèles open source, puis fusionne-les et déduplique-les avant qu’aucun n’atteigne TypeSafe. Des familles différentes devraient élargir la recherche plus que des appels répétés à un seul proposeur.
- Compare les modèles et méthodes prédictifs. Essaie la régression linéaire ou elastic-net, un régresseur à vecteurs de support, des forêts aléatoires, et une recalibration là où la sortie en aval est probabiliste. Vérifie si les caractéristiques découvertes aident en dehors de CatBoost.
- Ajoute une baseline d’embeddings. Un embedding transforme une note en quelques centaines
de nombres sans question attachée :
sentence-transformers/all-MiniLM-L6-v2tourne en local,text-embedding-3-smalld’OpenAI est un appel hébergé. Ajoutes-en un aux colonnes découvertes et mesure s’il porte quelque chose qu’elles n’ont pas. - Aligne la validation sur le déploiement. Utilise des découpages chronologiques pour prédire le futur, des découpages groupés quand des lignes liées doivent rester ensemble, et garde un jeu de test final intact, touché ni par la découverte de caractéristiques ni par la sélection de modèle.
- Arrête-toi sur un plateau. Termine la boucle quand le RMSE validé de façon croisée cesse de s’améliorer pendant un nombre de tours fixé, ou quand elle atteint un budget de questions ou de requêtes.
- Lance une recherche plus longue en mode Goal d’un agent. Donne-lui une métrique, un budget et une règle d’arrêt explicites, puis laisse-le proposer, évaluer et affiner davantage de tours.
- Vérifie la stabilité. Répète la découverte sur plusieurs graines ou tranches de données et garde les questions qui restent utiles, plutôt que celles dont l’importance repose sur un seul découpage.
Ouvre-le dans le playground
Ce lien de partage contient une note de dégustation plus toutes les questions avec lesquelles la boucle a fini.
playground_link = make_playground_link(
NOTES[0], feature_questions(accepted), models=[TYPESAFE_MODEL]
)
display(
Markdown(
f"🔗 [Open the note + questions in the TypeSafe playground]({playground_link})"
)
)