safe_evaluation

safe_evaluation - is an implementation of polish notation algorithm for mathematical and dataframe operations

how to use

1. Installation

   pip install safe-evaluation
   

2. Tutorial

   • import evaluation class

   from safe_evaluation import Evaluator
   

   • initialize evaluator

   evaluator = Evaluator()
   

   • send string that you want to evaluate

   result = evaluator.solve(command="2 + 2")
   

   • method will return result

   print(result)
   

3. Examples of usage

   • evaluator.solve(command="2 + 2")  # 4
     

   • evaluator.solve(command="lambda x: x * 2")  # lambda x: x * 2
     

   • evaluator.solve(command="2 * x - y", local={"x": 2, "y": 3})  # 1
     

   • import pandas as pd
     
     df = pd.DataFrame(data={'col1': [1, 2], 'col2': [3, 4]})
     evaluator.solve(command="${col1} + ${col2}", df=df)  # 0   4
                                                             1   6
                                                             dtype: int64
     

   • evaluator.solve(command="list(map(lambda x, y: x * 2 + y, range(5), range(3, 8)))")  # [3,6,9,12,15]
     

   • evaluator.solve(command="list(map(lambda x: x + y, [0,1,2,3,4]))", local={"y": 10})  # [10,11,12,13,14]
     

   • import pandas as pd
     
     df = pd.DataFrame(data={'col1': [1, 2, 4, 7], 'col2': [3, 4, 8, 9]})
     evaluator.solve(command="np.mean(${col1}) + np.mean(${col2})", df=df)  # 9.5
     

   • import pandas as pd
     
     df = pd.DataFrame(data={'col1': [1, 2, 4, 7], 'col2': [3, 4, 8, 9]})
     evaluator.solve(command="${col1}.apply(lambda v: v ** 2 > 0).all()", df=df)  # True
     

   • import pandas as pd
     
     df = pd.DataFrame(data={'col1': [1, 2, 4, 7], 'col2': [3, 4, 8, 9]})
     evaluator.solve(command="${col1}.apply(lambda v: 1 if v < 3 else 2)", df=df)  # 0    1
                                                                                      1    1
                                                                                      2    2
                                                                                      3    2
                                                                                      Name: col1, dtype: int64
     

   • from datetime import datetime
     
     import pandas as pd
     
     dates = [datetime(year=2022, month=11, day=11 + i) for i in range(7)]
     df = pd.DataFrame(data={'dates': dates})
     evaluator.solve(command="${dates}.dt.dayofweek", df=df)  # 0    4
                                                                 1    5
                                                                 2    6
                                                                 3    0
                                                                 4    1
                                                                 5    2
                                                                 6    3
                                                                 Name: dates, dtype: int64
     

   • range = evaluator.solve(command="pd.date_range(start='2021-02-05', end='2021-03-05', freq='1D')"")
     len(range)  # 29
     

   • df = pd.DataFrame(data={'col1': [1, 2], 'col2': [3, 4]})
     sorted_df = evaluator.solve(command="${__df}.sort_values('col2', ascending=False)", df=df)  #    col1  col2
                                                                                                    0     2     4
                                                                                                    1     1     3
     

4. Supported operations

   • in is not supported yet
   • comparation (<=, <, > , >=, !=, ==)
   • unary (+, -)
   • boolean (~, &, |, ^)
   • binary (+, -, /, *, //, %, **)

5. Supported functions

   • map, filter, list, range
   • bool, int, float, complex, str
   • numpy module functions (np.mean, etc.)
   • pandas module functions (pd.date_range, etc.)
   • anonymous functions

6. Supported access to data

   • ${col_name} is same as df['col_name']
   • ${__df} is same as df

7. It is possible to use your own class for preprocessing or calculating

   • from safe_evaluation import Evaluator, BasePreprocessor
     from safe_evaluation.constants import TypeOfCommand
     
     class MyPreprocessor(BasePreprocessor):
         def __init__(self, evaluator):
             self.evaluator = evaluator
     
         def prepare(self, command, df, local):
             return [(TypeOfCommand.VARIABLE, 'v'), '**', (TypeOfCommand.VALUE, 2), '>', (TypeOfCommand.VALUE, 5)]
     
     
     evaluator = Evaluator(preprocessor=MyPreprocessor)
     
     expression = 'v ** 2 > 5'
     
     output = evaluator.solve(expression, local={'v': 2})    #    False
     

8. It is possible to use your own settings

   • from safe_evaluation import Evaluator, Settings
     
     evaluator = Evaluator()
     settings = Settings(allowed_funcs=['filter', 'list'])
     evaluator.change_settings(settings)
     
     output = evaluator.solve("list(filter(lambda x: x < 0, [-1,0,1]))")   #    [-1]
     

   • from safe_evaluation import Evaluator, Settings
     
     evaluator = Evaluator()
     settings = Settings(allowed_funcs=['list'])
     evaluator.change_settings(settings)
     
     output = evaluator.solve("list(filter(lambda x: x < 0, [-1,0,1]))")   #    Exception: Unsupported function filter