Releases

0.9.3

  1. The new version supports training policies using the SAC algorithm;

  2. The new version supports custom node loss functions;

  3. The new version supports custom network nodes;

  4. The new version supports introducing expert constraints;

  5. The new version optimizes the control strategy for multiple control nodes;

  6. The new version supports configuring multiple discriminators;

  7. The new version fixes some bugs and optimizes training stability;

  8. The documentation has been updated in the new version.

0.9.2

  1. Updated RNN Backbone, improved training performance when using LSTM and GRU as Backbone;

  2. Optimized asynchronous decision flowchart loading, compatible with more loading mechanisms;

  3. Optimized time step nodes, empty node mechanisms, and supports more rich visual content;

  4. Fixed some bugs, improved training stability;

  5. Updated documentation content, providing more explanations for added functionality, and improved the organization of tutorial examples.

0.9.1

  1. Support loading training completion environment model as initialization model;

  2. Support the use of multi head networks;

  3. Support freezing model parameters of some nodes in training environment;

  4. Support multiple matchers configuration;

  5. Add response curve drawing function;

  6. Support the use of random, grid, zoom and Bayes algorithms;

  7. New function cases are added to the document;

  8. New task examples (IB) have been added to the document;

  9. Fixed some errors;

  10. Some new test cases have been added.

0.9.0

  1. Add causal discovery algorithm to assist in generating decision flow graph;

  2. Support for training multi-environment adaptive policy;

  3. Support for training environments and training policy using different decision flow graphs;

  4. Support defining nodes without data binding;

  5. Support for calling individual node models for inference through virtual environment models;

  6. The optimization policy can explain the style of the tree;

  7. Fixed some bugs.

0.8.0

  1. Upgrade dependency package version;

  2. Support the action for multi-steps in env;

  3. Delete nohup cmd in pyarmor register;

  4. Support plot image in dataset_val;

  5. Support append step node;

  6. Support miss some data point;

  7. Fixed some bugs.

0.7.3

  1. Rename venv_algo（revive_p, revive_t）;

  2. Decision diagram supporting training stateless transfer;

  3. Update revive_p algorithm;

  4. Append gym_env example script in examples;

  5. Fixed some bugs.

0.7.2

  1. Support automatic generation of multi-time frame data nodes.

  2. Support license verification by accesskey.

  3. Fixed some other bugs.

0.7.1

  1. Add refrigerator task example.

  2. Update revive algorithm, training is more stable and better;

  3. Update model rollout mechanism.

  4. Fixed some other bugs.

0.7.0

  1. Support Online Identity Verification;

  2. Support for training multi agent policy;

  3. Update revive algorithm, training is more stable and better;

  4. Fixed some other bugs.

0.6.0

  1. Update function node import mechanism.

  2. Update value test for the expert functions.

  3. Update ray version==1.8.

  4. Add license register script.

  5. Fix some bug.