Coverage for SynRBL/confidence_prediction.py: 29%

34 statements  

« prev     ^ index     » next       coverage.py v7.4.3, created at 2024-04-11 11:17 +0200

1import joblib 

2import pandas as pd 

3import numpy as np 

4import importlib.resources 

5import SynRBL.SynAnalysis 

6 

7from SynRBL.SynAnalysis.analysis_utils import ( 

8 calculate_chemical_properties, 

9 count_boundary_atoms_products_and_calculate_changes, 

10) 

11from SynRBL.SynUtils.common import update_reactants_and_products 

12 

13 

14class ConfidencePredictor: 

15 def __init__( 

16 self, 

17 reaction_col="reaction", 

18 input_reaction_col="input_reaction", 

19 confidence_col="confidence", 

20 solved_by_col="solved_by", 

21 solved_by_method="mcs-based", 

22 mcs_col="mcs", 

23 ): 

24 self.model = joblib.load( 

25 importlib.resources.files(SynRBL.SynAnalysis) 

26 .joinpath("scoring_function.dump") 

27 .open("rb") 

28 ) 

29 self.reaction_col = reaction_col 

30 self.input_reaction_col = input_reaction_col 

31 self.confidence_col = confidence_col 

32 self.solved_by_col = solved_by_col 

33 self.solved_by_method = solved_by_method 

34 self.mcs_col = mcs_col 

35 

36 def predict(self, reactions, stats=None, threshold=0): 

37 reactions = [ 

38 r 

39 for r in reactions 

40 if self.solved_by_col in r.keys() 

41 and r[self.solved_by_col] == self.solved_by_method 

42 ] 

43 conf_success = 0 

44 if len(reactions) > 0: 

45 _reactions = count_boundary_atoms_products_and_calculate_changes( 

46 reactions, self.reaction_col, self.mcs_col 

47 ) 

48 update_reactants_and_products(_reactions, self.input_reaction_col) 

49 _reactions = calculate_chemical_properties(_reactions) 

50 

51 df = pd.DataFrame(_reactions) 

52 

53 X_pred = df[ 

54 [ 

55 "carbon_difference", 

56 "fragment_count", 

57 "total_carbons", 

58 "total_bonds", 

59 "total_rings", 

60 "num_boundary", 

61 "ring_change_merge", 

62 "bond_change_merge", 

63 ] 

64 ] 

65 

66 confidence = np.round(self.model.predict_proba(X_pred)[:, 1], 3) 

67 assert len(reactions) == len(confidence) 

68 for r, c in zip(reactions, confidence): 

69 r[self.confidence_col] = c 

70 if c >= threshold: 

71 conf_success += 1 

72 if stats is not None: 

73 stats["confident_cnt"] = conf_success 

74 return reactions