brando/random-all-ascii-dataset

Random ASCII Dataset

This dataset contains random sequences of ASCII characters, with "train," "validation," and "test" splits, designed to simulate text-like structures using all printable ASCII characters. Each sequence consists of pseudo-randomly generated "words" of various lengths, separated by spaces to mimic natural language text.

Dataset Details

• Splits: Train, Validation, and Test
• Number of sequences:
  • Train: 5000 sequences
  • Validation: 5000 sequences
  • Test: 5000 sequences
• Sequence length: 512 characters per sequence
• Character pool: All printable ASCII characters, including letters, digits, punctuation, and whitespace.

Sample Usage

To load this dataset in Python, you can use the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("brando/random-ascii-dataset")

# Access the train, validation, and test splits
train_data = dataset["train"]
val_data = dataset["validation"]
test_data = dataset["test"]

# Print a sample
print(train_data[0]["text"])

Example Data

Below are examples of random sequences generated in this dataset:

#Example 1:
"!Q4$^V3w L@#12 Vd&$%4B+ (k#yFw! [7*9z"

#Example 2:
"T^&3xR f$xH&ty ^23M* qW@# Lm5&"

#Example 3:
"b7$W %&6Zn!!R xT&8N z#G m93T +%^0"

License

This dataset is released under the Apache License 2.0. You are free to use, modify, and distribute this dataset under the terms of the Apache License.

Citation

@misc{miranda2021ultimateutils,
    title={Ultimate Utils - the Ultimate Utils Library for Machine Learning and Artificial Intelligence},
    author={Brando Miranda},
    year={2021},
    url={https://github.com/brando90/ultimate-utils},
    note={Available at: \url{https://www.ideals.illinois.edu/handle/2142/112797}},
    abstract={Ultimate Utils is a comprehensive library providing utility functions and tools to facilitate efficient machine learning and AI research, including efficient tensor manipulations and gradient handling with methods such as `detach()` for creating gradient-free tensors.}
}

Code that generate it

#ref: https://chatgpt.com/c/671ff56a-563c-8001-afd5-94632fe63d67 
import os
import random
import string
from huggingface_hub import login
from datasets import Dataset, DatasetDict

# Function to load the Hugging Face API token from a file
def load_token(file_path: str) -> str:
    """Load API token from a specified file path."""
    with open(os.path.expanduser(file_path)) as f:
        return f.read().strip()

# Function to log in to Hugging Face using a token
def login_to_huggingface(token: str) -> None:
    """Authenticate with Hugging Face Hub."""
    login(token=token)
    print("Login successful")

# Function to generate a random word of a given length
def generate_random_word(length: int, character_pool: str) -> str:
    """Generate a random word of specified length from a character pool."""
    return "".join(random.choice(character_pool) for _ in range(length))

# Function to generate a single random sentence with "words" of random lengths
def generate_random_sentence(sequence_length: int, character_pool: str) -> str:
    """Generate a random sentence of approximately sequence_length characters."""
    words = [
        generate_random_word(random.randint(3, 10), character_pool)
        for _ in range(sequence_length // 10)  # Estimate number of words to fit length
    ]
    sentence = " ".join(words)
    # print(f"Generated sentence length: {len(sentence)}\a")  # Print length and sound alert
    return sentence

# Function to create a dataset of random "sentences"
def create_random_text_dataset(num_sequences: int, sequence_length: int, character_pool: str) -> Dataset:
    """Create a dataset with random text sequences."""
    data = {
        "text": [generate_random_sentence(sequence_length, character_pool) for _ in range(num_sequences)]
    }
    return Dataset.from_dict(data)

# Main function to generate, inspect, and upload dataset with train, validation, and test splits
def main() -> None:
    # Step 1: Load token and log in
    key_file_path: str = "/lfs/skampere1/0/brando9/keys/brandos_hf_token.txt"
    token: str = load_token(key_file_path)
    login_to_huggingface(token)

    # Step 2: Dataset parameters
    num_sequences_train: int = 5000
    num_sequences_val: int = 5000
    num_sequences_test: int = 5000
    sequence_length: int = 512
    character_pool: str = string.printable  # All ASCII characters (letters, digits, punctuation, whitespace)

    # Step 3: Create datasets for each split
    train_dataset = create_random_text_dataset(num_sequences_train, sequence_length, character_pool)
    val_dataset = create_random_text_dataset(num_sequences_val, sequence_length, character_pool)
    test_dataset = create_random_text_dataset(num_sequences_test, sequence_length, character_pool)

    # Step 4: Combine into a DatasetDict with train, validation, and test splits
    dataset_dict = DatasetDict({
        "train": train_dataset,
        "validation": val_dataset,
        "test": test_dataset
    })

    # Step 5: Print a sample of the train dataset for verification
    print("Sample of train dataset:", train_dataset[:5])

    # Step 6: Push the dataset to Hugging Face Hub
    dataset_name: str = "brando/random-ascii-dataset"
    dataset_dict.push_to_hub(dataset_name)
    print(f"Dataset uploaded to https://huggingface.co/datasets/{dataset_name}")

# Run the main function
if __name__ == "__main__":
    main()