Replace tqdm with a logger for different modules. - #3
MPMPMPMPMPMPMP wants to merge 27 commits into
Conversation
Sorry, something went wrong.
|
Hey, first thank you very much for the idea and implementation! Three things I spotted:
I will take some time tomorrow to walk through it in detail. |
Sorry, something went wrong.
This is the with all the loggers set to info if you set everything to warning except the optimization, then you see only the steps. |
Sorry, something went wrong.
|
I really like to use the progress bar where I can see the current energy without scrolling through a wall of text. |
Sorry, something went wrong.
JanLuca
left a comment
There was a problem hiding this comment.
Besides the minor comments, I like the PR :) Thank you very much for the work!
Only the optimizer.py file I could not really review since the change of the indention makes the comparison hard
Sorry, something went wrong.
| from __future__ import annotations | ||
|
|
||
| import logging | ||
| from typing import Any | ||
|
|
||
| from varipeps import config as _cfg_mod # uses the global config instance | ||
|
|
||
| _LOGGING_INITIALIZED = False | ||
|
|
||
| def _to_py_log_level(level: Any) -> int: | ||
| # Accept both enum values and raw ints; OFF disables effectively | ||
| try: | ||
| val = int(level) | ||
| except Exception: | ||
| val = logging.INFO | ||
| if val == 0: # OFF | ||
| return logging.CRITICAL + 10 | ||
| return val | ||
|
|
||
| def init_logging(cfg: Any | None = None) -> None: | ||
| """ | ||
| Initialize logging based on the provided config (or global config). | ||
| Safe to call multiple times; replaces handlers to avoid duplicates. | ||
| """ | ||
| global _LOGGING_INITIALIZED | ||
| if cfg is None: | ||
| cfg = _cfg_mod.config | ||
|
|
||
| root = logging.getLogger("varipeps") | ||
| # Remove old handlers to prevent duplicate logs | ||
| for h in list(root.handlers): | ||
| root.removeHandler(h) | ||
|
|
||
| root.setLevel(_to_py_log_level(getattr(cfg, "log_level_global", logging.INFO))) | ||
| root.propagate = False | ||
|
|
||
| # fmt = logging.Formatter( | ||
| # fmt="%(asctime)s %(levelname)s %(name)s: %(message)s", | ||
| # datefmt="%H:%M:%S", | ||
| # ) | ||
|
|
||
| fmt = logging.Formatter( | ||
| fmt="%(asctime)s %(levelname)s %(message)s", | ||
| datefmt="%Y-%m-%d %H:%M:%S", | ||
| ) | ||
|
|
||
| if getattr(cfg, "log_to_console", True): | ||
| sh = logging.StreamHandler() | ||
| sh.setFormatter(fmt) | ||
| root.addHandler(sh) | ||
|
|
||
| if getattr(cfg, "log_to_file", False): | ||
| fh = logging.FileHandler(getattr(cfg, "log_file", "varipeps.log")) | ||
| fh.setFormatter(fmt) | ||
| root.addHandler(fh) | ||
|
|
||
| # Per-module levels | ||
| logging.getLogger("varipeps.optimizer").setLevel( | ||
| _to_py_log_level(getattr(cfg, "log_level_optimizer", logging.INFO)) | ||
| ) | ||
| logging.getLogger("varipeps.ctmrg").setLevel( | ||
| _to_py_log_level(getattr(cfg, "log_level_ctmrg", logging.INFO)) | ||
| ) | ||
| logging.getLogger("varipeps.line_search").setLevel( | ||
| _to_py_log_level(getattr(cfg, "log_level_line_search", logging.INFO)) | ||
| ) | ||
| logging.getLogger("varipeps.expectation").setLevel( | ||
| _to_py_log_level(getattr(cfg, "log_level_expectation", logging.INFO)) | ||
| ) | ||
|
|
||
| _LOGGING_INITIALIZED = True | ||
|
|
||
| def ensure_logging_configured(cfg: Any | None = None) -> None: | ||
| """ | ||
| Initialize logging once on first call; subsequent calls are no-ops. | ||
| """ | ||
| global _LOGGING_INITIALIZED | ||
| if not _LOGGING_INITIALIZED: | ||
| init_logging(cfg) |
There was a problem hiding this comment.
I would move that to the config file since there one could catch changes of the configuration and update the logger instances for the new values. Will prepare a commit to showcase that.
Sorry, something went wrong.
The HDF5 file stores config values like log_level_global as a numeric (e.g., numpy.int64). When loading, that integer is passed into VariPEPS_Config. In setattr, the field type for log_level_global is the Enum LogLevel, so integers must be coerced to LogLevel. Your version’s coercion path doesn’t catch your value, so it falls through and raises: Type mismatch for option 'log_level_global', got '<class numpy.int64>', expected '<enum 'LogLevel'>'. Why it falls through The loader passes a numpy integer (or a 0-d/1-d array) instead of a Python int. The Enum branch in setattr is too strict about the numeric checks, so it doesn’t convert that value into LogLevel.
TLDR;
This pull request introduces a major refactor to the logging and progress reporting system across the codebase, replacing the use of tqdm_loggable and custom print/debug statements with Python's standard logging module. It also adds configurable logging levels and destinations to the configuration, and updates many routines to use structured logging for progress, warnings, and informational output.
If you think something is missing please let me know!
Simple usage (just inlude this in your
It makes the log look the following way (depending on the log levels you choose):
(please don't tell me to remove the emojis I like the eye candy ... )
Key changes include:
Logging System Overhaul:
CTMRG and Optimization Routine Updates:
Configuration and Code Cleanup:
These changes collectively modernize the codebase's output and monitoring capabilities, making it easier for users and developers to control and interpret runtime information.
Logging System Overhaul:
CTMRG and Optimization Routine Updates:
Configuration and Code Cleanup: