v0.4.3

Changes

PyPi Release

• ./examples directory is now included in the release
  • added MANIFEST.in

Command Line Interface (CLI)

• Fixed: sapsan get_examples command

v0.4.0

Changes

General

• Loaders for Train and Evaluate now have the same format
• The functions above have an identical interface for both PyTorch and Sklearn

Estimators

• Fixed: Model Saving & Loading

  • loaded models can continue to be trained
  • upon their initialization when loading the model, you can redefine its previous config, such as n_epoch, lr
  • optimizer dict state is correctly saved and loaded
  • optimizer state is moved to cpu or gpu depending on a setup (catalyst doesn't do it on its own, which caused issues when evaluating a loaded model)
  • Added dummy estimators for loading (otherwise all estimators have load and save)
    • load_estimator() for torch
    • load_sklearn_estimator() for sklearn

• Reworked how the models are initialized

  • upon calling the estimator, ex: estimator = CNN3d(loaders=loaders)
    • before: when training the model, upon estimator.train
  • model initialization requires to provide loaders now

• All self vars in ModelConfig() get recorded in tracking by default

• Added options in ModelConfig()

  • lr and min_lr - learning rate parameters are no longer hard-coded
  • device - sets a specific device to run the models on, either cpu or cuda

• Added sklearn_backend and torch_backend to be used by all estimators

  • sklearn-based estimators have a structure close to torch-based
  • pytorch_estimator -> torch_backend
  • cleared up variable name conventions throughout

Evaluation

• Evaluate and Data loader accept data without target
  • useful when there is no ground truth to compare to
  • will still output pdf, cdf, and spatial, without comparison metrics
  • Evaluate.run() now output a dict of "pred_cube" and "target_cube" (if the latter is provided)
• PDF and CDF plots are now combined under a single figure
  • recorded as 'pdf_cdf.png' in MLflow
• Fixed: definition of n_output_channel in Evaluate()

Command Line Interface (CLI)

• Added new option: sapsan create --ddp option copies torch_backend.py

  • gives ability to customize Catalyst Runner
  • adjust DDP settings based on the linked Catalyst DDP tutorial in the Wiki
  • will be useful when running on HPC
  • refer to Parallel GPU Training on the Wiki for more details

• Fixed: CLI click initialization

Graphical User Interface (GUI)

• Up to date with Streamlit 0.87.0
• PDF and CDF plots are now showed as well
• Fixed: data loading issue in regards to train_fraction

MLflow

• MLflow: evaluate runs will be nested under the recent train run
  • significantly aids organization
• Added estimator.model.forward() to be recorded by MLflow (if torch is used)

Plotting

• Plotting routines return Axes object
• All parameters are changed for the Axes instead of plt which allows individual tweaking after return
• figsize and ax arguments added to most plotting routines
  • useful if you create a figure and subplots outside of the plotting routines
• Universal plotting params expanded and were made easily accessible through plot_params()

Other

• Edited the examples, tests, and estimator template to reflect model initialization changes
• Requirements Updated:
  • streamlit >= 0.87.0
  • plotly >= 5.2.0
  • tornado >= 6.1.0
  • notebook >= 6.4.3 (fixes security vulnerabilities)
• Added a few data_loader warnings
• Cleaned up debug prints throughout the code
• Expanded code comments

v0.3.0

Changes

Command Line Interface (CLI)

• New & Changed Commands
  • sapsan create --name {name} or sapsan create -n {name} - creates a custom project template tree
  • sapsan test - runs pytest to make sure Sapsan is working correctly
  • sapsan get_examples - copies Sapsan's examples into the current working directory for easy access
  • sapsan --version - to check the installed version
• Updated CLI options and --help

Testing

• moved tests to Sapsan's root folder, so they are always accessible when installed via pip
• notebook tests don't create a separate folder, but test on the existing example notebooks
• sapsan test allows for the user to run pytest tests to check if everything was installed correctly

Custom Estimators

• To get started, run: sapsan create -n {name}
  • where {name} should be replaced with your custom project name
• Significantly simplified the template structure, making it much easier to navigate and get started
  • cleaned up by removing all unnecessary templates, leaving a few scripts that allow to customize the estimator (i.e. ML network), Jupyter Notebook interface, and Dockerfile to easily share your project
  • pre-filled all templates with the custom project name

Examples

• sapsan get_examples: copies examples into ./sapsan_examples
  • makes the example jupyter notebooks with various ML algorithms easily accessible

Other

• sapsan now has __version__ attribute
• Updated the following sections in Documentation to reflect the CLI and template changes
  • Installation
  • Getting Started
  • Custom Estimators
  • Custom Docker
• Minor comment updates to example notebooks

v0.2.11

Changes

Testing

• notebook testing errors trigger pytest
• estimator loading testing is the same for both PyTorch and Scikit-learn

MLflow

• for evaluation, some model training pars and metrics are also recorded to make it easier to see with what model the prediction was done

PICAE Estimator

• removed redundant torch.device pass for PICAE estimator

Estimators

• simplified ModelConfig() in both included estimators and custom template
  • ModelConfig now only has __init__
  • moved load() to backend, removed to_dict()
  • added load_config() under core Estimator class in models.py to enforce it to be included in custom estimators (most cases won't be affected, as it is part of the backend pytorch_estimator.py)
  • changes reflected in the estimator template + wiki
  • Save config now saves all input parameters instead of only the ones tracked by MLflow
  • Loading is now consistent between PyTorch and scikit-learn based models

Other

• minor formatting fixes in examples
• minor bug fixes

v0.2.10

Changes

There are no new features in this update, but numerous quality of life improvements and bug fixes.

Data Loader

• fixed multi-checkpoint loading for hdf5 module
  • they are added as new batches to the loaded data np.array
• fixed splitting by batches
  • the rounding error when computing the fraction to split into batches has been corrected
  • default batch_num=1, hence batch_size = input_size if it is not specified (no need to specify the batching in the loader)
• fixed input data size processing, when sampling is not specified

Backend

• FakeBackend is now the default if backend is not specified

Sampling

• no longer need to pass original input shape
  • it is determined from the data itself
• added a warning on dealing with irregular shapes
  • the function will try to match the requested shape, but if it can't - a warning is issued

Tensor Calculation

• fixed Tensor calculation indices
• added "only_x_components" flag
  • at times full tensor is too taxing on memory

Requirements

• corrected installation requirements under python 3.8.10+

v0.2.9

Changes

Parallel GPU

• Parallel GPU training is done via Distributed Data Parallel (DDP), through Catalyst
  • Besides locally, capable of running on HPC
• Fixes in train and evaluation methods to account for DDP
• Added a Parallel GPU tutorial on Wiki

Physics

• Gradient Model - full 3D tensor is now calculated, fixes to the algorithm
• Dynamic Smagorinsky Model - fully operational, outputs full 3D tensor
• Added tensor() function to calculate a stress tensor (see API entry)
• assertion checks in various functions to make the errors more intelligible

Docker

• Minor quality of life simplifications
• Added a Docker tutorial on Wiki

MLflow

• Updated compatibility with MLflow 18.0
• No need to restart kernel when re-running the model in jupyter notebooks
• Fixed compatibility with Catalist 21.7 logging
• Added MLflow tutorial on Wiki

GUI

• Fixed compatibility with Catalist 21.7 - correct logging and plotting

Other

• Expanded Custom Estimator tutorial on Wiki
• Added Community Guidelines to add new ML models and contribute to Sapsan
• Set up a jupyter notebook example on Google Colab to play around with Sapsan
• Minor bug fixes

v0.2.8

Changes

• requirement updates
• simplified hierarchy
• compatibility with Catalyst 21.5+
• removed automatic PyTorch install: user will need to do it manually

v0.2.7

Changes

• fixed PICAE in pypi

v0.2.6

Changes

• GUI fixes - brought up to date with the latest backend
• minor bug fixes

v0.2.5

Changes

• Physics Informed CAE method

  • reworked PICAE now adheres Sapsan's interface (sapsan/lib/estimators/picae.py)
  • added PICAE example with random data

• Data loaders

  • loaders are now consistent between pytorch and sklearn
  • train() takes in loaders, instead of inputs & targets separately. Those can be either Pytorch.Dataloader or a list of inputs & targets (i.e loaders = [x,y]). This is done to accommodate Pytorch-based and sklearn-based models, which require the input of different formats.
    • data can be loaded as a numpy array by calling load_numpy(), instead of load
    • loaded numpy data can be converted to torch dataloader via convert_to_torch([x, y])
    • alternatively, both of the steps can be combined by just calling load()
  • cleaned up data_functions; added new methods: flatten, split_by_batch, get_loader_shape
  • corrected split into train and valid datasets + enhanced with train_test_split function from sklearn
  • added new params to data_loader: train_fraction, shuffle
  • support for irregular input data shapes

• Examples

  • added PICAE example with random data
  • cleaned up examples further

• Estimators

  • cleaned up CNN3d estimator, deleted legacy functions
  • further generalized Pytorch estimator to be used as a backend for any PyTorch-based models

• Templates

  • reflect the changes from above - streamlined

• Tests

  • added PICAE related tests on push

• Misc

  • further improvement to backend handling of data transformations
  • minor bug fixes