Skip to main content

​
Overview

The Hospital Data Analysis Platform uses a centralized configuration system based on the SystemConfig dataclass. This approach provides type safety, default values, and clear documentation of all configurable parameters.

​
SystemConfig Dataclass

The SystemConfig dataclass is defined in config.py:6 and serves as the single source of truth for all system parameters. 
from dataclasses import dataclass, field
from pathlib import Path

@dataclass
class SystemConfig:
    random_seed: int = 42
    test_size: float = 0.25
    data_dir: Path = Path(__file__).resolve().parent.parent / "test"
    output_dir: Path = Path(__file__).resolve().parent / "artifacts"
    stream_chunk_size: int = 16
    stream_interval_ms: int = 10
    hardware_memory_limit_mb: int = 256
    hardware_compute_budget: int = 10_000
    benchmark_runs: int = 5
    confidence_level: float = 0.95
    feature_columns: list[str] = field(
        default_factory=lambda: ["age", "height", "weight", "bmi", "children", "months"]
    )
    target_risk: str = "diagnosis"
    target_outcome: str = "blood_test"
    experiment_memory_limits_mb: list[int] = field(default_factory=lambda: [64, 128, 256])
    experiment_compute_budgets: list[int] = field(default_factory=lambda: [2_000, 5_000, 10_000])
    experiment_stream_speeds_ms: list[int] = field(default_factory=lambda: [5, 10, 20])

​
Configuration Parameters

​
Reproducibility

random_seed (int, default: 42) Global random seed for reproducible experiments. This seed is propagated to Python’s random, NumPy’s random number generator, and the PYTHONHASHSEED environment variable.

​
Data Splitting

test_size (float, default: 0.25) Fraction of the dataset reserved for testing. The remaining data is used for training.

​
File Paths

data_dir (Path, default: ../test) Directory containing input CSV datasets. Defaults to the test directory relative to the project root. output_dir (Path, default: artifacts/) Directory for storing outputs including trained models, benchmark results, and diagnostic reports. The directory is automatically created if it doesn’t exist.

​
Streaming Parameters

stream_chunk_size (int, default: 16) Number of records processed per streaming batch. Smaller values reduce memory footprint but increase I/O overhead. stream_interval_ms (int, default: 10) Delay in milliseconds between streaming chunks. Used to simulate real-time data ingestion.

​
Hardware Constraints

hardware_memory_limit_mb (int, default: 256) Memory budget in megabytes for resource-constrained execution. Used for deployment sizing and constraint validation. hardware_compute_budget (int, default: 10_000) Computational budget representing the number of operations or inference cycles allowed. Used for performance profiling.

​
Benchmarking

benchmark_runs (int, default: 5) Number of repeated benchmark iterations for statistical confidence. Higher values improve estimate reliability but increase runtime. confidence_level (float, default: 0.95) Confidence level for statistical intervals (e.g., 0.95 = 95% confidence). Supported values: 0.90, 0.95, 0.99.

​
Features and Targets

feature_columns (list[str], default: ["age", "height", "weight", "bmi", "children", "months"]) List of column names used as input features for model training. target_risk (str, default: "diagnosis") Column name for the primary risk classification target. target_outcome (str, default: "blood_test") Column name for the secondary outcome prediction target.

​
Experiment Ranges

experiment_memory_limits_mb (list[int], default: [64, 128, 256]) Memory limits to test during resource constraint experiments. experiment_compute_budgets (list[int], default: [2_000, 5_000, 10_000]) Compute budgets to test during performance experiments. experiment_stream_speeds_ms (list[int], default: [5, 10, 20]) Streaming intervals to test during latency experiments.

​
Usage Example

​
Using Default Configuration

from config import CONFIG

print(f"Random seed: {CONFIG.random_seed}")
print(f"Output directory: {CONFIG.output_dir}")
print(f"Feature columns: {CONFIG.feature_columns}")

​
Creating Custom Configuration

from config import SystemConfig
from pathlib import Path

custom_config = SystemConfig(
    random_seed=123,
    test_size=0.3,
    output_dir=Path("/tmp/my_experiment"),
    benchmark_runs=10,
    confidence_level=0.99
)

custom_config.output_dir.mkdir(parents=True, exist_ok=True)

​
Modifying Global Configuration

from config import CONFIG

# Adjust for low-memory environment
CONFIG.hardware_memory_limit_mb = 128
CONFIG.stream_chunk_size = 8

# Use more benchmark runs for critical evaluation
CONFIG.benchmark_runs = 20

​
Best Practices

  1. Immutable Defaults: Avoid modifying the global CONFIG object across modules. Instead, pass configuration objects as function parameters.
  2. Path Validation: Always ensure output_dir exists before writing artifacts: 
    CONFIG.output_dir.mkdir(parents=True, exist_ok=True)
  3. Seed Propagation: Use the reproducibility utilities to properly set the random seed: 
    from utils.reproducibility import set_global_seed
set_global_seed(CONFIG.random_seed)
  4. Environment-Specific Overrides: Use environment variables or configuration files for deployment-specific settings rather than hardcoding values.

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us