tessl/pypi-tensorboardx

TensorBoardX lets you watch Tensors Flow without Tensorflow

Overview

Eval results

Files

Record Writer

Name: tessl/pypi-tensorboardx
Author: tessl

Low-level TensorFlow record format writer for direct event file management. Provides basic binary record writing capabilities with support for local filesystem, Amazon S3, and Google Cloud Storage backends.

Capabilities

Initialization

Creates a RecordWriter instance for writing binary protocol buffer records to various storage backends.

class RecordWriter:
    def __init__(self, path: str):
        """
        Creates a RecordWriter for writing TensorFlow record format files.

        Parameters:
        - path: File path for writing records. Supports local paths, S3 ("s3://bucket/path"), 
                and Google Cloud Storage ("gs://bucket/path") URLs.
        """

Record Writing

Write binary data records with CRC32C checksums for data integrity verification.

def write(self, data: bytes):
    """
    Write a binary data record with length header and CRC32C checksums.

    Parameters:
    - data: Binary data to write as a record
    """

File Management

Control file flushing and closing for data persistence and resource cleanup.

def flush(self):
    """
    Flush any buffered data to the underlying storage.
    """

def close(self):
    """
    Close the record writer and release resources.
    """

Cloud Storage Support

RecordWriter supports multiple storage backends through a plugin system:

Amazon S3 Support

Requires pip install boto3 for S3 functionality.

from tensorboardX import RecordWriter

# Write to S3 bucket
writer = RecordWriter("s3://my-bucket/tensorboard/events.out.tfevents")
writer.write(event_data)
writer.close()

Google Cloud Storage Support

Requires pip install google-cloud-storage for GCS functionality.

from tensorboardX import RecordWriter

# Write to GCS bucket
writer = RecordWriter("gs://my-bucket/tensorboard/events.out.tfevents")
writer.write(event_data)
writer.close()

Usage Examples

Basic Local File Writing

from tensorboardX import RecordWriter

# Create writer for local file
writer = RecordWriter("./logs/events.out.tfevents")

# Write binary protocol buffer data
event_data = b"serialized_protocol_buffer_data"
writer.write(event_data)

# Ensure data is written to disk
writer.flush()
writer.close()

Advanced Usage with Context Manager

from tensorboardX import RecordWriter

# Automatic resource cleanup
class RecordWriterContext:
    def __init__(self, path):
        self.writer = RecordWriter(path)
    
    def __enter__(self):
        return self.writer
    
    def __exit__(self, exc_type, exc_val, exc_tb):
        self.writer.close()

# Usage
with RecordWriterContext("./logs/events.out.tfevents") as writer:
    writer.write(event_data)
    # Automatically closed on exit

Utility Functions

Name Validation

Utility functions for TensorFlow-compatible name generation and validation.

def make_valid_tf_name(name: str) -> str:
    """
    Convert arbitrary strings to valid TensorFlow operation names.

    Parameters:
    - name: Input name string

    Returns:
    - Valid TensorFlow operation name string
    """

Storage Backend Registration

Functions for registering custom storage backends.

def register_writer_factory(prefix: str, factory):
    """
    Register a custom writer factory for specific URL prefixes.

    Parameters:
    - prefix: URL prefix (e.g., "s3", "gs")
    - factory: Factory object with open() and directory_check() methods
    
    Raises:
    - ValueError: If prefix contains ':'
    """

def directory_check(path: str):
    """
    Initialize directory structure for the given path.

    Parameters:
    - path: Directory path to initialize
    """

def open_file(path: str):
    """
    Open a writer for the specified path using registered factories.

    Parameters:
    - path: File path, supports custom prefixes

    Returns:
    - File writer object
    """

Install with Tessl CLI