Following a tutorial on HF results into Precision is ill-defined and being set to 0.0

Beginners
1

Here’s the code for the simple example.

import evaluate
import numpy as np
import matplotlib.pyplot as plt
from datasets import load_dataset
from transformers import (
    AutoTokenizer,
    AutoConfig,
    AutoModelForSequenceClassification,
    TrainingArguments,
    Trainer,
    DataCollatorWithPadding
)

data = load_dataset("/data/for/binary/text/classification/")
tokenizer = AutoTokenizer.from_pretrained('roberta-base')

def preprocess_data(examples):
    txt_data = []
    for example in examples['values']:
        num2str = ' '.join(str(token) for token in example)
        txt_data.append(num2str)

    examples['text'] = txt_data
    return tokenizer(txt_data, truncation=True)

tokenized_data = data.map(preprocess_data, batched=True)

print(tokenized_data)

DatasetDict({
    train: Dataset({
        features: ['values', 'answer', 'cat', 'text', 'input_ids', 'attention_mask'],
        num_rows: 100000
    })
    validation: Dataset({
        features: ['values', 'answer', 'cat', 'text', 'input_ids', 'attention_mask'],
        num_rows: 14988
    })
    test: Dataset({
        features: ['values', 'answer', 'cat', 'text', 'input_ids', 'attention_mask'],
        num_rows: 15000
    })

tokenized_data = tokenized_data.rename_column("answer", "labels")

data_collator = DataCollatorWithPadding(tokenizer=tokenizer, padding='longest')

config = AutoConfig.from_pretrained('roberta-base')

config.num_labbels = 2

model = AutoModelForSequenceClassification.from_config(config)

metrics = evaluate.combine(['accuracy', 'f1', 'precision', 'recall'] )

def compute_metrics(eval_pred):
    predictions, labels = eval_pred
    predictions = np.argmax(predictions, axis=1)
    return metrics.compute(predictions=predictions, references=labels)

training_args = TrainingArguments(
    output_dir="roberta_checkpoints",
    learning_rate=2e-5,
    per_device_train_batch_size=640,
    per_device_eval_batch_size=640,
    num_train_epochs=2,
    weight_decay=0.01,
    evaluation_strategy="epoch",
    save_strategy="epoch",
    load_best_model_at_end=True,
    push_to_hub=False,
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_data["train"],
    eval_dataset=tokenized_data["validation"],
    tokenizer=tokenizer,
    data_collator=data_collator,
    compute_metrics=compute_metrics,
)

trainer.train()

And here’s the output during training.

The following columns in the training set  don't have a corresponding argument in `RobertaForSequenceClassification.forward` and have been ignored: cat, values, text.
/python3.12/site-packages/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use thePyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning
  warnings.warn(
***** Running training *****
  Num examples = 100000
  Num Epochs = 2
  Instantaneous batch size per device = 640
  Total train batch size (w. parallel, distributed & accumulation) = 640
  Gradient Accumulation steps = 1
  Total optimization steps = 314

Epoch	Training Loss	Validation Loss	Accuracy	F1	    Precision	Recall
  1	       No log	     0.693468	    0.497131  0.000000   0.000000	0.000000
  2	       No log	     0.693163	    0.497131  0.000000   0.000000	0.000000

The following columns in the evaluation set  don't have a corresponding argument in `RobertaForSequenceClassification.forward` and have been ignored: cat, values, text.
***** Running Evaluation *****
  Num examples = 14988
  Batch size = 640
/python3.12/site-packages/sklearn/metrics/_classification.py:1469: **UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 due to no predicted samples. Use `zero_division` parameter to control this behavior.**
  _warn_prf(average, modifier, msg_start, len(result))
Saving model checkpoint to roberta_checkpoints/checkpoint-157
Configuration saved in roberta_checkpoints/checkpoint-157/config.json
Model weights saved in roberta_checkpoints/checkpoint-157/pytorch_model.bin
tokenizer config file saved in roberta_checkpoints/checkpoint-157/tokenizer_config.json
Special tokens file saved in roberta_checkpoints/checkpoint-157/special_tokens_map.json

Finally the trainer outputs the following

TrainOutput(global_step=314, training_loss=0.7030252833275279, metrics={'train_runtime': 188.2461, 'train_samples_per_second': 1062.439, 'train_steps_per_second': 1.668, 'total_flos': 2261110632000000.0, 'train_loss': 0.7030252833275279, 'epoch': 2.0})

Any help understanding what’s causing this issue would be appreciated.