schema.py

Go to the documentation of this file.

# ===========================================
# schema.py
# ===========================================
 

 
 
import polars as pl
 
 
"""!Canonical schema used throughout the pipeline.
 
Each key represents a column name, and the value is a tuple:
(dtype: pl.DataType, nullable: bool). This schema is used to:
- Cast raw CSV data safely
- Generate ClickHouse-compatible SQL
- Validate data consistency during preprocessing
 
@note Nullable fields typically represent optional CPU-level metrics (e.g., L2/L3).
@note Timestamps are expected to be in millisecond resolution.
"""
SCHEMA = {
    # Non-nullable fields
    "Timestamp": (pl.Datetime("ms"), False),
    "BatchID": (pl.Utf8(), False),
    "Method": (pl.Utf8(), False),
    "Trials": (pl.Int64(), False),
    "Cycles": (pl.Int64(), False),
    "Instructions": (pl.Int64(), False),
    "IPC": (pl.Float64(), False),
    "Wall Time (s)": (pl.Float64(), False),
    "Wall Time (ns)": (pl.Int64(), False),
    "Cache Loads": (pl.Int64(), False),
    "Cache Misses": (pl.Int64(), False),
    "Cache Miss %": (pl.Float64(), False),
    "L1 Loads": (pl.Int64(), False),
    "L1 Misses": (pl.Int64(), False),
    "L1 Miss %": (pl.Float64(), False),
 
    # Nullable fields
    "L2 Loads": (pl.Int64(), True),
    "L2 Misses": (pl.Int64(), True),
    "L2 Miss %": (pl.Float64(), True),
    "L3 Loads": (pl.Int64(), True),
    "L3 Misses": (pl.Int64(), True),
    "L3 Miss %": (pl.Float64(), True),
 
    "TLB Loads": (pl.Int64(), False),
    "TLB Misses": (pl.Int64(), False),
    "TLB Miss %": (pl.Float64(), False),
    "Branch Instructions": (pl.Int64(), False),
    "Branch Misses": (pl.Int64(), False),
    "Branch Miss %": (pl.Float64(), False),
    "Misses/Trial": (pl.Float64(), False),
    "Cycles/Trial": (pl.Float64(), False),
}