openscilab / pymilo Goto Github PK

Description

We can already convert sklearn models to onnx, even add extra metadata if we want, investigate the model intrinsics via https://netron.app/ (locally or on the browser). How is the functionality of pymilo different than ONNX?

Description

add Ensemble Methods support

• AdaBoostClassifier
• AdaBoostRegressor
• BaggingClassifier
• BaggingRegressor
• ExtraTreesClassifier
• ExtraTreesRegressor
• GradientBoostingClassifier
• GradientBoostingRegressor
• IsolationForest
• RandomForestClassifier
• RandomForestRegressor
• RandomTreesEmbedding
• StackingClassifier
• StackingRegressor
• VotingClassifier
• VotingRegressor
• HistGradientBoostingRegressor
• HistGradientBoostingClassifier

Description

add full support for SVM models:

• LinearSVC
• LinearSVR
• NuSVC
• NuSVR
• OneClassSVM
• SVC
• SVR

Description

Refactor the code to enhance the performance and efficiency (Speed up slower parts)

Description

Add Neural Networks support

• Bernoulli Restricted Boltzmann Machine (RBM)
• Multi-layer Perceptron classifier
• Multi-layer Perceptron regressor

Description

Support WebSocket protocol to allow the exported models to be used in model streaming

Description

• Support the Client-Server model to provide the possibility of ML model abstraction and remote access to the models
• PyMilo ML model wrapper to wrap around ML models to abstract model function calls.
• Add fallback function to PyMilo ML model wrapper to delegate function calls to the remote server to get executed in the main model

Description

Enhance the coverage and accuracy of test procedures

Description

• Support model versioning to enable PyMilo to support different model versions

Description

add Neighbors support

• KNeighborsClassifier
• KNeighborsRegressor
• NearestNeighbors
• NearestCentroid
• RadiusNeighborsClassifier
• RadiusNeighborsRegressor
• LocalOutlierFactor

Describe the feature you want to add

Ability to give specific url to PyMilo Import instance and it will then download the given model during the instantiation part.

Describe your proposed solution

No response

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Describe the feature you want to add

numpy.ndarray with dtype of X should get deserialized with exact type even if it is a custom non-serializable numpy-defined data type like intc.

Describe your proposed solution

No response

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Describe the feature you want to add

Support serialization and deserialization of all Scikit-learn preprocessing transformers

• Binarizer
• FunctionTransformer
• KBinsDiscretizer
• KernelCenterer
• LabelBinarizer
• LabelEncoder
• MultiLabelBinarizer
• MaxAbsScaler
• Normalizer
• OneHotEncoder
• OrdinalEncoder
• PolynomialFeatures
• PowerTransformer
• QuantileTransformer
• RobustScaler
• SplineTransformer
• StandardScaler
• TargetEncoder

Describe your proposed solution

No response

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Description

add Decision Trees support

• Decision Tree Classifier
• Decision Tree Regressor
• Extra Tree Classifier
• Extra Tree Regressor

Description

add Clustering support

• K-Means
• Bisecting K-Means
• Mini Batch Kmeans
• Affinity propagation
• Mean-shift
• Spectral clustering
• Spectral Biclustering
• Spectral Coclustering
• Ward hierarchical clustering
• • Agglomerative clustering
• • Feature Agglomeration
• DBSCAN
• HDBSCAN
• OPTICS
• Gaussian mixture
• BayesianGaussianMixture
• BIRCH

Description

Provide a JavaScript/TypeScript SDK to use ML models through PyMilo in a web browser

there is a new feature in GPT 3.5 to fine tune gpt with your own data. this can be integrated into pymilo to export the fine tuned gpt with the given data
https://openai.com/blog/gpt-3-5-turbo-fine-tuning-and-api-updates

Description

There should be constraints that force serialization of the "root_" before the "dummy_leaf_" field in Birch model.

Describe the feature you want to add

Add ability to call specific functions after PyMilo Import

Describe your proposed solution

No response

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Description

• Ability to compress the exported file
• Set the compression method in the Export-Import function constructor parameters

Description

Ability to upload the exported model to IPFS and opensea.

Description

add full support Naive Bayes Models:

• GaussianNB
• MultinomialNB
• ComplementNB
• BernoulliNB
• CategoricalNB

Describe the feature you want to add

Serialization for numpy.intc data structure should be added(it is used in scikit svm models)

Describe your proposed solution

No response

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Description

Add more detailed descriptions to the current document and provide a few detailed tutorial pages.

Description

Support serialization and deserialization of PyTorch models

Description

Ability to Import & Export multiple Pymilo-transported ML models.

Description

Support splitting (and merging) the exported (and imported) model files to enhance memory usage efficiency.

Description

• Refactor the code to enhance simplicity and readability (Pythonic way)

Description

Support serialization and deserialization of Scikit-learn Pipeline models

Description

fix loss function serialization issue in SGDClassifier in scikit1.4.0

Description

please sync the to_json() function of the PyMilo export with the exact content of the exported file.