This model repository contains artifacts from an AMIS commodity relevance classifier training run. It includes the Transformer model, any configured TF-IDF or sentence-embedding baselines, prediction files, and the training report.

Dataset: faodl/amis-agri-rice
Dataset subset: ``
Dataset revision: main
Text column: chunk_text
Label column: label
Transformer: FacebookAI/xlm-roberta-base
Generated at: 2026-06-08T17:42:20.320378+00:00

Dataset Summary

Split	Rows	Label 0	Label 1	Unique groups	Mean text length
train	4823	3873	950	2128	711.6
validation	1071	846	225	456	702.2
test	954	772	182	457	694.3

Threshold Comparison on Validation Split

Validation metrics document threshold selection and tuning behavior; test metrics remain the primary estimate of out-of-sample performance.

Model	Threshold	Accuracy	Precision	Recall	F1	ROC AUC	Average precision
logistic_tfidf	0.500	0.948	0.882	0.867	0.874	0.958	0.926
logistic_tfidf	0.517	0.951	0.899	0.867	0.882	0.958	0.926
xgboost_tfidf	0.500	0.957	0.891	0.907	0.899	0.975	0.947
xgboost_tfidf	0.522	0.957	0.891	0.907	0.899	0.975	0.947
embedding-logistic_sentence_embeddings	0.500	0.945	0.827	0.933	0.877	0.978	0.910
embedding-logistic_sentence_embeddings	0.617	0.954	0.879	0.907	0.893	0.978	0.910
embedding-svm_sentence_embeddings	0.500	0.952	0.914	0.853	0.883	0.977	0.908
embedding-svm_sentence_embeddings	0.496	0.953	0.915	0.858	0.885	0.977	0.908
embedding-lightgbm_sentence_embeddings	0.500	0.947	0.900	0.840	0.869	0.979	0.922
embedding-lightgbm_sentence_embeddings	0.037	0.953	0.866	0.920	0.892	0.979	0.922
transformer	0.500	0.960	0.889	0.924	0.906	0.977	0.918
transformer	0.966	0.962	0.904	0.916	0.909	0.977	0.918

Threshold Comparison on Test Split

Model	Threshold	Accuracy	Precision	Recall	F1	ROC AUC	Average precision
logistic_tfidf	0.500	0.942	0.829	0.879	0.853	0.963	0.893
logistic_tfidf	0.517	0.942	0.832	0.874	0.853	0.963	0.893
xgboost_tfidf	0.500	0.950	0.872	0.863	0.867	0.980	0.936
xgboost_tfidf	0.522	0.951	0.881	0.857	0.869	0.980	0.936
embedding-logistic_sentence_embeddings	0.500	0.954	0.832	0.951	0.887	0.982	0.930
embedding-logistic_sentence_embeddings	0.617	0.954	0.845	0.929	0.885	0.982	0.930
embedding-svm_sentence_embeddings	0.500	0.950	0.872	0.863	0.867	0.983	0.938
embedding-svm_sentence_embeddings	0.496	0.950	0.872	0.863	0.867	0.983	0.938
embedding-lightgbm_sentence_embeddings	0.500	0.953	0.878	0.874	0.876	0.980	0.935
embedding-lightgbm_sentence_embeddings	0.037	0.948	0.814	0.940	0.872	0.980	0.935
transformer	0.500	0.962	0.876	0.934	0.904	0.990	0.972
transformer	0.966	0.961	0.901	0.896	0.898	0.990	0.972

Confusion Matrices on Test Split

Rows are true labels and columns are predicted labels.

logistic_tfidf at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	739	33
RELEVANT	22	160

logistic_tfidf at threshold 0.517

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	740	32
RELEVANT	23	159

xgboost_tfidf at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	749	23
RELEVANT	25	157

xgboost_tfidf at threshold 0.522

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	751	21
RELEVANT	26	156

embedding-logistic_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	737	35
RELEVANT	9	173

embedding-logistic_sentence_embeddings at threshold 0.617

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	741	31
RELEVANT	13	169

embedding-svm_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	749	23
RELEVANT	25	157

embedding-svm_sentence_embeddings at threshold 0.496

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	749	23
RELEVANT	25	157

embedding-lightgbm_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	750	22
RELEVANT	23	159

embedding-lightgbm_sentence_embeddings at threshold 0.037

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	733	39
RELEVANT	11	171

transformer at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	748	24
RELEVANT	12	170

transformer at threshold 0.966

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	754	18
RELEVANT	19	163

Validation-Tuned Thresholds

logistic_tfidf: threshold 0.517 (validation F1 0.882); test F1 change vs 0.5: -0.001.
xgboost_tfidf: threshold 0.522 (validation F1 0.899); test F1 change vs 0.5: +0.002.
embedding-logistic_sentence_embeddings: threshold 0.617 (validation F1 0.893); test F1 change vs 0.5: -0.002.
embedding-svm_sentence_embeddings: threshold 0.496 (validation F1 0.885); test F1 change vs 0.5: +0.000.
embedding-lightgbm_sentence_embeddings: threshold 0.037 (validation F1 0.892); test F1 change vs 0.5: -0.004.
transformer: threshold 0.966 (validation F1 0.909); test F1 change vs 0.5: -0.006.

Artifacts

logistic_tfidf: /content/agri-rice-classifier/baselines/logistic
xgboost_tfidf: /content/agri-rice-classifier/baselines/xgboost
embedding-logistic_sentence_embeddings: /content/agri-rice-classifier/baselines/embedding-logistic
embedding-svm_sentence_embeddings: /content/agri-rice-classifier/baselines/embedding-svm
embedding-lightgbm_sentence_embeddings: /content/agri-rice-classifier/baselines/embedding-lightgbm
transformer: /content/agri-rice-classifier/transformer

Inference

Install the runtime dependencies:

pip install transformers torch huggingface_hub pandas joblib scikit-learn xgboost lightgbm

Transformer

python

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

MODEL_ID = "faodl/agri-rice-classifier"

texts = [
    "Rice export prices increased after new procurement rules were announced.",
    "The finance ministry released its monthly fuel tax bulletin.",
]

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, subfolder="transformer")
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID, subfolder="transformer")
threshold = float(getattr(model.config, "threshold", 0.5))

encoded = tokenizer(
    texts,
    truncation=True,
    padding=True,
    max_length=256,
    return_tensors="pt",
)

with torch.no_grad():
    logits = model(**encoded).logits
    probabilities = torch.softmax(logits, dim=-1)[:, 1].tolist()

for text, probability in zip(texts, probabilities):
    label = model.config.id2label[int(probability >= threshold)]
    print({"text": text, "probability_positive": probability, "label": label})

TF-IDF Baselines

Available baseline names in this run: "logistic", "xgboost".

python

import json
import joblib
from huggingface_hub import hf_hub_download

MODEL_ID = "faodl/agri-rice-classifier"
BASELINE = "logistic"

texts = [
    "Maize production forecasts were revised after delayed rains.",
    "The central bank published new exchange rate statistics.",
]

model_path = hf_hub_download(
    repo_id=MODEL_ID,
    repo_type="model",
    filename=f"baselines/{BASELINE}/{BASELINE}_tfidf.joblib",
)
report_path = hf_hub_download(
    repo_id=MODEL_ID,
    repo_type="model",
    filename="report.json",
)

pipeline = joblib.load(model_path)
with open(report_path, encoding="utf-8") as handle:
    report = json.load(handle)

threshold = next(
    result["validation_best_threshold"]["threshold"]
    for result in report["results"]
    if result["model_type"] == f"{BASELINE}_tfidf"
)

probabilities = pipeline.predict_proba(texts)[:, 1]
for text, probability in zip(texts, probabilities):
    label = "RELEVANT" if probability >= threshold else "NOT_RELEVANT"
    print({"text": text, "probability_positive": float(probability), "label": label})

Sentence-Embedding Baselines

Available embedding baseline names in this run: "embedding-logistic", "embedding-svm", "embedding-lightgbm".

python

import joblib
import torch
from huggingface_hub import hf_hub_download
from transformers import AutoModel, AutoTokenizer

MODEL_ID = "faodl/agri-rice-classifier"
BASELINE = "embedding-logistic"

texts = [
    "Wheat export inspections rose as demand from importers increased.",
    "The sports ministry announced a new stadium renovation plan.",
]

model_path = hf_hub_download(
    repo_id=MODEL_ID,
    repo_type="model",
    filename=f"baselines/{BASELINE}/{BASELINE}.joblib",
)
artifact = joblib.load(model_path)
tokenizer = AutoTokenizer.from_pretrained(artifact["embedding_model_name"])
encoder = AutoModel.from_pretrained(artifact["embedding_model_name"])
encoder.eval()

encoded_batches = []
batch_size = artifact.get("embedding_batch_size", 64)
for start in range(0, len(texts), batch_size):
    batch_texts = texts[start : start + batch_size]
    inputs = tokenizer(
        batch_texts,
        padding=True,
        truncation=True,
        max_length=artifact.get("embedding_max_length", 256),
        return_tensors="pt",
    )
    with torch.no_grad():
        outputs = encoder(**inputs)
    token_embeddings = outputs.last_hidden_state
    attention_mask = inputs["attention_mask"].unsqueeze(-1).to(token_embeddings.dtype)
    embeddings = (token_embeddings * attention_mask).sum(dim=1)
    embeddings = embeddings / attention_mask.sum(dim=1).clamp(min=1e-9)
    if artifact.get("normalize_embeddings", True):
        embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=1)
    encoded_batches.append(embeddings)
embeddings = torch.cat(encoded_batches).numpy()
probabilities = artifact["classifier"].predict_proba(embeddings)[:, 1]
threshold = artifact["validation_best_threshold"]["threshold"]

for text, probability in zip(texts, probabilities):
    label = "RELEVANT" if probability >= threshold else "NOT_RELEVANT"
    print({"text": text, "probability_positive": float(probability), "label": label})

Files

REPORT.md: Markdown report for this training run.
report.json: Machine-readable report containing metrics and thresholds.
transformer/: Fine-tuned Transformer artifacts, when Transformer training is enabled.
baselines/: TF-IDF and sentence-embedding baseline artifacts, when baseline training is enabled.
*/validation_predictions.csv and */test_predictions.csv: Split-level predictions.