README.md

---
library_name: transformers
pipeline_tag: text-classification
base_model: FacebookAI/xlm-roberta-base
tags:
- text-classification
- binary-classification
- amis
- agriculture
language: multilingual
---

# 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-utilization`
- Dataset subset: ``
- Dataset revision: `ada4a04088a98f8f64bc7485c57d4c7f422c2151`
- Text column: `chunk_text`
- Label column: `label`
- Transformer: `FacebookAI/xlm-roberta-base`
- Generated at: `2026-05-27T10:50:45.867038+00:00`

## Dataset Summary

| Split | Rows | Label 0 | Label 1 | Unique groups | Mean text length |
| --- | ---: | ---: | ---: | ---: | ---: |
| train | 4877 | 4347 | 530 | 2513 | 696.6 |
| validation | 978 | 899 | 79 | 538 | 690.6 |
| test | 1016 | 904 | 112 | 539 | 690.7 |

## 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 | 0.912 | 0.465 | 0.582 | 0.517 | 0.872 | 0.594 |
| logistic_tfidf | 0.608 | 0.942 | 0.696 | 0.494 | 0.578 | 0.872 | 0.594 |
| xgboost_tfidf | 0.500 | 0.945 | 0.931 | 0.342 | 0.500 | 0.823 | 0.588 |
| xgboost_tfidf | 0.177 | 0.934 | 0.592 | 0.570 | 0.581 | 0.823 | 0.588 |
| embedding-logistic_sentence_embeddings | 0.500 | 0.912 | 0.476 | 0.861 | 0.613 | 0.953 | 0.762 |
| embedding-logistic_sentence_embeddings | 0.722 | 0.957 | 0.703 | 0.810 | 0.753 | 0.953 | 0.762 |
| embedding-svm_sentence_embeddings | 0.500 | 0.955 | 0.807 | 0.582 | 0.676 | 0.952 | 0.754 |
| embedding-svm_sentence_embeddings | 0.310 | 0.957 | 0.713 | 0.785 | 0.747 | 0.952 | 0.754 |
| embedding-lightgbm_sentence_embeddings | 0.500 | 0.954 | 0.750 | 0.646 | 0.694 | 0.948 | 0.782 |
| embedding-lightgbm_sentence_embeddings | 0.042 | 0.952 | 0.670 | 0.797 | 0.728 | 0.948 | 0.782 |
| transformer | 0.500 | 0.970 | 0.798 | 0.848 | 0.822 | 0.966 | 0.854 |
| transformer | 0.471 | 0.971 | 0.800 | 0.861 | 0.829 | 0.966 | 0.854 |

## Threshold Comparison on Test Split

| Model | Threshold | Accuracy | Precision | Recall | F1 | ROC AUC | Average precision |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| logistic_tfidf | 0.500 | 0.926 | 0.691 | 0.598 | 0.641 | 0.899 | 0.726 |
| logistic_tfidf | 0.608 | 0.930 | 0.902 | 0.411 | 0.564 | 0.899 | 0.726 |
| xgboost_tfidf | 0.500 | 0.924 | 1.000 | 0.312 | 0.476 | 0.892 | 0.692 |
| xgboost_tfidf | 0.177 | 0.918 | 0.663 | 0.527 | 0.587 | 0.892 | 0.692 |
| embedding-logistic_sentence_embeddings | 0.500 | 0.891 | 0.503 | 0.884 | 0.641 | 0.955 | 0.710 |
| embedding-logistic_sentence_embeddings | 0.722 | 0.935 | 0.689 | 0.750 | 0.718 | 0.955 | 0.710 |
| embedding-svm_sentence_embeddings | 0.500 | 0.930 | 0.741 | 0.562 | 0.640 | 0.956 | 0.704 |
| embedding-svm_sentence_embeddings | 0.310 | 0.934 | 0.686 | 0.741 | 0.712 | 0.956 | 0.704 |
| embedding-lightgbm_sentence_embeddings | 0.500 | 0.937 | 0.740 | 0.661 | 0.698 | 0.960 | 0.791 |
| embedding-lightgbm_sentence_embeddings | 0.042 | 0.929 | 0.639 | 0.821 | 0.719 | 0.960 | 0.791 |
| transformer | 0.500 | 0.951 | 0.777 | 0.777 | 0.777 | 0.968 | 0.817 |
| transformer | 0.471 | 0.950 | 0.770 | 0.777 | 0.773 | 0.968 | 0.817 |

## 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 | 874 | 30 |
| RELEVANT | 45 | 67 |

### logistic_tfidf at threshold 0.608

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 899 | 5 |
| RELEVANT | 66 | 46 |

### xgboost_tfidf at threshold 0.500

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 904 | 0 |
| RELEVANT | 77 | 35 |

### xgboost_tfidf at threshold 0.177

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 874 | 30 |
| RELEVANT | 53 | 59 |

### embedding-logistic_sentence_embeddings at threshold 0.500

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 806 | 98 |
| RELEVANT | 13 | 99 |

### embedding-logistic_sentence_embeddings at threshold 0.722

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 866 | 38 |
| RELEVANT | 28 | 84 |

### embedding-svm_sentence_embeddings at threshold 0.500

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 882 | 22 |
| RELEVANT | 49 | 63 |

### embedding-svm_sentence_embeddings at threshold 0.310

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 866 | 38 |
| RELEVANT | 29 | 83 |

### embedding-lightgbm_sentence_embeddings at threshold 0.500

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 878 | 26 |
| RELEVANT | 38 | 74 |

### embedding-lightgbm_sentence_embeddings at threshold 0.042

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 852 | 52 |
| RELEVANT | 20 | 92 |

### transformer at threshold 0.500

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 879 | 25 |
| RELEVANT | 25 | 87 |

### transformer at threshold 0.471

| True / Predicted | NOT_RELEVANT | RELEVANT |
| --- | ---: | ---: |
| NOT_RELEVANT | 878 | 26 |
| RELEVANT | 25 | 87 |


## Validation-Tuned Thresholds

- `logistic_tfidf`: threshold `0.608` (validation F1 `0.578`); test F1 change vs 0.5: `-0.077`.
- `xgboost_tfidf`: threshold `0.177` (validation F1 `0.581`); test F1 change vs 0.5: `+0.111`.
- `embedding-logistic_sentence_embeddings`: threshold `0.722` (validation F1 `0.753`); test F1 change vs 0.5: `+0.077`.
- `embedding-svm_sentence_embeddings`: threshold `0.310` (validation F1 `0.747`); test F1 change vs 0.5: `+0.073`.
- `embedding-lightgbm_sentence_embeddings`: threshold `0.042` (validation F1 `0.728`); test F1 change vs 0.5: `+0.021`.
- `transformer`: threshold `0.471` (validation F1 `0.829`); test F1 change vs 0.5: `-0.003`.

## Artifacts

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

## Inference

Install the runtime dependencies:

```bash
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-utilization-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-utilization-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-utilization-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.