PacktPublishing / Modern-Time-Series-Forecasting-with-Python

Modern Time Series Forecasting with Python, published by Packt
MIT License
451 stars 227 forks source link

Setting up environment: 404 pytorch #28

Open marbel89 opened 1 year ago

marbel89 commented 1 year ago

When setting up the environment with mamba, it fails to download at pytorch.

RuntimeError: Multi-download failed. Reason: Transfer finalized, status: 404 [https://conda.anaconda.org/pytorch/pytorch-2.0.1-py3.9_cpu_0.tar.bz2] 955 bytes

manujosephv commented 1 year ago

The env.yml is more of a convenience, and I realize now that it is too brittle and doesn't age well :)

My suggestion would be to create the environment on your own, just pinning versions of key packages. I have tried to comment packages which aren't essential below

dependencies:
  - python==3.9
  - plotly
  - py-xgboost
  - scikit-learn>=0.21.2
  - pandas
  - catboost
  - lightgbm
  - category_encoders
  - pmdarima>=1.8.0
  - xlrd # Only used for reading excel. Can be ignored
  - lxml # Only used for reading excel. Can be ignored
  - openpyxl # Only used for reading excel. Can be ignored
  - python-kaleido==0.1.0 # Only used for saving plots. Can be ignored
  - numpy
  - scipy
  - tqdm
  - matplotlib
  - humanize
  - optuna
  - pytorch-forecasting
  - numba
  - holidays
  - seaborn
  - statsmodels
  - statsforecast==0.6.0 # Don't think we are using this anywhere. You can try excluding it
  - u8darts<=0.19.0
  - pytorch::pytorch # PyTorch my suggestion would be to follow instructions from pytorch.org to install correctly
  - pytorch::torchvision
  - pytorch::torchaudio
  - pytorch-lightning
  #- pytorch::cudatoolkit==10.2
  - pip
  - pip:
    - kaggle>=1.5.2 # This is only to download datasets. You can ignore if dataset is present in the correct manner already
    - missingno>=0.5.0
    - window_ops
    - pymannkendall
    - tsfel
    - rich
    - git+https://github.com/manujosephv/pytorch_tabular.git@modern_ts_freeze #specifying a commit that is working because of pandas depencency
    - omegaconf==2.1.2
    - tensor-sensor[torch]