TFT PyTorch Forecasting - Stallion 튜토리얼 #2

TFT PyTorch Forecasting - Stallion 튜토리얼 #2 #

#2025-05-29

#version check

예제 코드에 맞는 패키지 버전

  • CUDA: 11.7
  • PyTorch: 1.13.1+cu117
  • PyTorch Lightning: 1.9.0
  • PyTorch Forecasting: 0.10.3

PyTorch Forecasting 0.10.3 선택 이유: 최신 버전은 아래 코드랑 호환 안됨

  1. Tuner().lr_find() -> 학습률 탐색, lightning>=2.x에서는 내부 콜백 구조 변경됨
  2. trainer.checkpoint_callback.best_model_path -> 베스트 모델 로드, trainer.checkpoint_callback 속성 제거됨
  3. optimizer="ranger" -> Ranger 옵티마이저 지정, 제거됨
  4. plot_prediction, plot_interpretation, plot_dependency -> 시각화 함수, 제거되거나 구조 변경됨
  5. optimize_hyperparameters -> Optuna 기반 튜닝, deprecated 또는 작동 오류

여기에 맞게 설치해주기.

conda create -n tftspace python=3.9 -y

pip install pip==23.3.1 #pip다운그레이드

#pip uninstall torch -y
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1+cu117 -f https://download.pytorch.org/whl/torch_stable.html
pip install torchmetrics==0.10.3  # torch<2.0 호환
pip install pytorch-lightning==1.6.5
pip install pytorch-forecasting==0.10.3

pip install ranger-adabelief #optimizer="ranger" 수행에 필요함

python 스크립트 상에서 확인하기

pip install ipykernel
python -m ipykernel install --user --name tftspace --display-name "tftspace" #커널등록

#!pip install "numpy<2.0"

import torch
import torchvision
import torchaudio
import torchmetrics
import pytorch_lightning
import pytorch_forecasting

print("torch:", torch.__version__)
print("torchvision:", torchvision.__version__)
print("torchaudio:", torchaudio.__version__)
print("torchmetrics:", torchmetrics.__version__)
print("pytorch_lightning:", pytorch_lightning.__version__)
print("pytorch_forecasting:", pytorch_forecasting.__version__)
)

torch: 1.13.1+cu117
torchvision: 0.14.1+cu117
torchaudio: 0.13.1+cu117
torchmetrics: 0.10.3
pytorch_lightning: 1.6.5
pytorch_forecasting: 0.10.3

제대로 설치됨!!

#load package

import warnings

warnings.filterwarnings("ignore")  # avoid printing out absolute paths

import copy
from pathlib import Path
import warnings

#import lightning.pytorch as pl
#from lightning.pytorch.callbacks import EarlyStopping, LearningRateMonitor
#from lightning.pytorch.loggers import TensorBoardLogger
import pytorch_lightning as pl
from pytorch_lightning.callbacks import EarlyStopping, LearningRateMonitor
from pytorch_lightning.loggers import TensorBoardLogger

import numpy as np
import pandas as pd
import torch

from pytorch_forecasting import Baseline, TemporalFusionTransformer, TimeSeriesDataSet
from pytorch_forecasting.data import GroupNormalizer
from pytorch_forecasting.metrics import MAE, SMAPE, PoissonLoss, QuantileLoss
from pytorch_forecasting.models.temporal_fusion_transformer.tuning import (
    optimize_hyperparameters,
)

lightning import 할때

  • 주석 처리된게 원래 스크립트이고 pytorch-lightning>=2.0에서 동작한다고 함
  • 스크립트 중에 pytorch-lightning<2.0에서만 동작하는 함수가 있어서 >=2.0로는 설치할수없음
  • 그래서 수정함.
from pytorch_forecasting.data.examples import get_stallion_data

data = get_stallion_data()

근데 버전을 낮춰서그런지 get_stallion_data()가 안먹어서 그냥 원래대로 버전을 맞추고 코드를 수정하는쪽으로 해야댈거같다.