Releases: jules-leguy/BBOMol
Releases · jules-leguy/BBOMol
BBOMol v0.1.1
New features
- Adding the possibility to select the number of threads used for each DFT calculation.
- Adding the possibility to select the DFT method and the memory allocated for each DFT calculation.
- Adding the option to select the property to be plotted by ** best so far ** plots.
- Now extracting the objective function computation times from EvoMol result files.
- Adding the possibility to add a point in ** best so far ** plots to represent the last value that has been obtained in all runs in the x axis.
- The probability of improvement (POI) merit function can now be used.
- Adding an option to set a threshold on the training set size of the surrogate. If more samples are in the dataset, a subset of the threshold size is drawn with uniform probability.
- Now supporting a random molecular descriptor.
Behaviour changes
- Updating legends of grid molecules drawing so that integer values are written without decimal values.
- Now printing the number of times the target was reached when displaying an ERT array.
Others
- Updating installation procedure and instructions.
- Updating version name.
Bug fixes
- Updating the code so that it is compatible with recent scikit-learn versions.
- Updating the code so that it is compatible with recent pandas versions.
- Fixing bug so that it is possible to plot properties that are represented in several columns in EvoMol output results.
- Updating postprocessing code so that it is possible to load EvoMol output files that do not contain objective computation time information.
- Fixing bug so that BBOMol can be used with latest EvoMol versions (recording computation time of evaluations).
- Fixing bug so that the calculation of the EI for the complete population works even if there are duplicate solutions.
- Fixing bug that occurs when comparing SMILES from init population by recomputing canonical SMILES before comparison
- Fixing bug so that the draw_best_solutions function can save images to disk even when used from a jupyter notebook.
BBOMol v0.1.0
New features
- Adding the possibility to specify the path to a file that contains the initial SMILES list (design of experiments).
- Adding the possibility to set the
"io_parameters"
attribute of the EvoMol internal optimization. - Adding the possibility to plot the distribution of a property depending on ranges of calls to the objective function.
- Adding the possibility to draw the best solutions according to a property for each run of each experiment.
Behaviour changes
- The choice of whether retrieving data from initial step for experiments plotting is now also applied on pure EvoMol experiments.
- All subscores are now retrieved for EvoMol experiments.
Descriptors computation updates
- The
descriptor
module is now moved to a dedicated and disctinct repositiory : ChemDesc. The documentation of BBOMol is updated in consequence. Before moving, the following changes were made to the module : - Possibility to give a dictionary that maps shingles SMILES to a unique ID that is used as index in the output descriptor.
- Adding a progress bar display during the computation of descriptors.
- Updating the parallel descriptors computation procedure so that no copies are performed to save memory.
- New options for molecular mechanics computation. Now possible to choose the program that performs the optimization and the force field. Options are (OpenBabel, MMFF94), (RDKit, MMFF94) and (RDKit, UFF).