faodl/agri-maize_corn-classifier

AMIS Commodity Classifier

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-maize_corn
• Dataset subset: ``
• Dataset revision: main
• Text column: chunk_text
• Label column: label
• Transformer: FacebookAI/xlm-roberta-base
• Generated at: 2026-06-05T20:32:51.106165+00:00

Dataset Summary

Split	Rows	Label 0	Label 1	Unique groups	Mean text length
train	4724	3822	902	2226	702.9
validation	1060	843	217	477	708.3
test	1054	819	235	478	711.9

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	0.896	0.700	0.862	0.773	0.929	0.841
logistic_tfidf	0.586	0.915	0.777	0.820	0.798	0.929	0.841
xgboost_tfidf	0.500	0.957	0.913	0.871	0.892	0.967	0.914
xgboost_tfidf	0.379	0.958	0.902	0.894	0.898	0.967	0.914
embedding-logistic_sentence_embeddings	0.500	0.881	0.649	0.912	0.759	0.959	0.867
embedding-logistic_sentence_embeddings	0.744	0.923	0.808	0.816	0.812	0.959	0.867
embedding-svm_sentence_embeddings	0.500	0.913	0.849	0.700	0.768	0.955	0.858
embedding-svm_sentence_embeddings	0.401	0.914	0.789	0.793	0.791	0.955	0.858
embedding-lightgbm_sentence_embeddings	0.500	0.916	0.791	0.802	0.796	0.963	0.878
embedding-lightgbm_sentence_embeddings	0.145	0.916	0.746	0.894	0.813	0.963	0.878
transformer	0.500	0.958	0.913	0.876	0.894	0.973	0.943
transformer	0.328	0.959	0.907	0.894	0.900	0.973	0.943

Threshold Comparison on Test Split

Model	Threshold	Accuracy	Precision	Recall	F1	ROC AUC	Average precision
logistic_tfidf	0.500	0.910	0.787	0.817	0.802	0.953	0.863
logistic_tfidf	0.586	0.915	0.857	0.740	0.795	0.953	0.863
xgboost_tfidf	0.500	0.942	0.914	0.817	0.863	0.968	0.920
xgboost_tfidf	0.379	0.948	0.895	0.868	0.881	0.968	0.920
embedding-logistic_sentence_embeddings	0.500	0.884	0.704	0.830	0.762	0.936	0.844
embedding-logistic_sentence_embeddings	0.744	0.896	0.831	0.668	0.741	0.936	0.844
embedding-svm_sentence_embeddings	0.500	0.894	0.892	0.596	0.714	0.932	0.842
embedding-svm_sentence_embeddings	0.401	0.902	0.851	0.681	0.757	0.932	0.842
embedding-lightgbm_sentence_embeddings	0.500	0.907	0.870	0.685	0.767	0.953	0.873
embedding-lightgbm_sentence_embeddings	0.145	0.901	0.784	0.770	0.777	0.953	0.873
transformer	0.500	0.935	0.892	0.804	0.846	0.953	0.890
transformer	0.328	0.935	0.881	0.817	0.848	0.953	0.890

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	767	52
RELEVANT	43	192

logistic_tfidf at threshold 0.586

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	790	29
RELEVANT	61	174

xgboost_tfidf at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	801	18
RELEVANT	43	192

xgboost_tfidf at threshold 0.379

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	795	24
RELEVANT	31	204

embedding-logistic_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	737	82
RELEVANT	40	195

embedding-logistic_sentence_embeddings at threshold 0.744

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	787	32
RELEVANT	78	157

embedding-svm_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	802	17
RELEVANT	95	140

embedding-svm_sentence_embeddings at threshold 0.401

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	791	28
RELEVANT	75	160

embedding-lightgbm_sentence_embeddings at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	795	24
RELEVANT	74	161

embedding-lightgbm_sentence_embeddings at threshold 0.145

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	769	50
RELEVANT	54	181

transformer at threshold 0.500

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	796	23
RELEVANT	46	189

transformer at threshold 0.328

True / Predicted	NOT_RELEVANT	RELEVANT
NOT_RELEVANT	793	26
RELEVANT	43	192

Validation-Tuned Thresholds

• logistic_tfidf: threshold 0.586 (validation F1 0.798); test F1 change vs 0.5: -0.007.
• xgboost_tfidf: threshold 0.379 (validation F1 0.898); test F1 change vs 0.5: +0.018.
• embedding-logistic_sentence_embeddings: threshold 0.744 (validation F1 0.812); test F1 change vs 0.5: -0.021.
• embedding-svm_sentence_embeddings: threshold 0.401 (validation F1 0.791); test F1 change vs 0.5: +0.042.
• embedding-lightgbm_sentence_embeddings: threshold 0.145 (validation F1 0.813); test F1 change vs 0.5: +0.010.
• transformer: threshold 0.328 (validation F1 0.900); test F1 change vs 0.5: +0.002.

Artifacts

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

Inference

Install the runtime dependencies:

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

Transformer

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

MODEL_ID = "faodl/agri-maize_corn-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".

import json
import joblib
from huggingface_hub import hf_hub_download

MODEL_ID = "faodl/agri-maize_corn-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".

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

MODEL_ID = "faodl/agri-maize_corn-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.