~/tora

What is Tora?

Tora is an experiment tracker built for speed and simplicity.

Track metrics, hyperparameters, and experiment metadata with just 2 lines of code.

Core APIs

Global API Functions

NAME Global API - Simple experiment tracking functions

SYNOPSIS python import tora tora.setup(name, workspace_id=None, ...) tora.tlog(name, value, step=None, metadata=None)

DESCRIPTION Global API functions provide a simple interface for single experiment workflows. Uses a global client instance for immediate logging with minimal setup.

FUNCTIONS

setup(name, workspace_id=None, description=None, hyperparams=None, tags=None, api_key=None, server_url=None, max_buffer_len=1) Initialize global experiment session.

   name            Experiment name (required)
   workspace_id    Target workspace ID
   description     Experiment description
   hyperparams     Hyperparameter dictionary
   tags            List of experiment tags
   api_key         Authentication key
   server_url      Custom server URL
   max_buffer_len  Buffer size (default: 1)

   Returns experiment ID string. Prints experiment URL to console.

tlog(name, value, step=None, metadata=None) Log metrics using global experiment.

   name      Metric name (required)
   value     Metric value - int or float (required)
   step      Step number
   metadata  Additional metadata dict (max 10KB)

   Note: Must call setup() first.

flush() Force immediate sending of all buffered metrics.

shutdown() Cleanup global client and flush remaining metrics.

is_initialized() Check if global client is initialized. Returns bool.

get_experiment_id() Get current experiment ID. Returns experiment ID string or None.

Tora Client Class

NAME Tora - Main client for experiment tracking

SYNOPSIS python client = Tora.create_experiment(name, workspace_id=None, ...) client = Tora.load_experiment(experiment_id, ...)

DESCRIPTION The Tora class provides methods for logging metrics, managing experiments, and supports both buffered and immediate metric logging.

CLASS METHODS

Tora.create_experiment(name, workspace_id=None, description=None, hyperparams=None, tags=None, max_buffer_len=25, api_key=None, server_url=None) Create new experiment and return Tora client instance.

   name            Experiment name (required)
   workspace_id    Target workspace ID
   description     Experiment description
   hyperparams     Hyperparameter dictionary
   tags            List of experiment tags
   max_buffer_len  Buffer size (default: 25)
   api_key         Authentication key
   server_url      Custom server URL

Tora.load_experiment(experiment_id, max_buffer_len=25, api_key=None, server_url=None) Load existing experiment and return Tora client instance.

   experiment_id   ID of existing experiment (required)
   max_buffer_len  Buffer size (default: 25)
   api_key         Authentication key
   server_url      Custom server URL

INSTANCE METHODS

log(name, value, step=None, metadata=None) Log single metric with automatic buffering.

   name      Metric name (required)
   value     Metric value - int or float (required)
   step      Step number
   metadata  Additional metadata dict (max 10KB)

log_metrics(metrics, step=None) Log multiple metrics at once.

   metrics   Dictionary of metric names to values (required)
   step      Step number for all metrics

flush() Force immediate sending of all buffered metrics.

shutdown() Flush all metrics and close client. Required for cleanup.

PROPERTIES

experiment_id Get experiment ID string (read-only)

url Get experiment URL string (read-only)

max_buffer_len Get/set maximum buffer length (read-write)

buffer_size Get current number of buffered metrics (read-only)

is_closed Check if client is closed - returns bool (read-only)

CONTEXT MANAGER Supports 'with' statement for automatic cleanup:

   ```python
   with Tora.create_experiment("name") as client:
       client.log("metric", 1.0)
   ```

Workspace Management

NAME create_workspace - Create new workspace for organizing experiments

SYNOPSIS python import tora workspace = tora.create_workspace(name, description=None, api_key=None, server_url=None)

DESCRIPTION Creates a new workspace for organizing experiments. Requires authentication.

FUNCTION

create_workspace(name, description=None, api_key=None, server_url=None) Create new workspace.

   name         Workspace name - max 255 chars (required)
   description  Workspace description - max 1000 chars
   api_key      Authentication key (required)
   server_url   Custom server URL

   Returns workspace data dictionary with id, name, description, etc.

Configuration

ENVIRONMENT VARIABLES

AUTHENTICATION Tora operates in anonymous mode by default, allowing you to track experiments without any authentication. Authentication is only required if you want to:

Associate experiments with your user account
Create and manage workspaces
Access workspace-specific features and collaboration tools

For functions that accept an api_key parameter, you can either:

Set the TORA_API_KEY environment variable (recommended)
Pass the API key directly as a function parameter

When using the environment variable, simply set it in your shell:

export TORA_API_KEY=your_api_key_here

Or in Python:

import os
os.environ['TORA_API_KEY'] = 'your_api_key_here'

Exception Classes

NAME Tora Exceptions - Error handling classes

DESCRIPTION Exception hierarchy for error handling and debugging.

EXCEPTIONS

ToraError Base exception class for all Tora errors

ToraValidationError Input validation errors

ToraNetworkError Network-related errors

ToraAPIError API response errors

ToraAuthenticationError Authentication errors

ToraConfigurationError Configuration errors

ToraExperimentError Experiment-related errors

ToraMetricError Metric logging errors

ToraWorkspaceError Workspace-related errors

ToraTimeoutError Request timeout errors

Pure Speed. Pure Insight.