Is there any way to generate new data?

Question

I have these kinds of data. Original data is from 1st column to 5th column.

Is there any way to create new data by randomly combining columns without specific criteria?

For example, I made a new data by making formula between columns. I know there is some way to make formula on my own with python. But I want to make about more than 100 columns. Is there any automatic way to do this? (Numpy, pandas or things like that)

Answer 1

You could use some builtin modules like random and operator which contains a set of mathematical functions. I used this approach to generate some random columns with a random combination of the original columns. Notice that the number of original columns used for every operation is also random, but you could set this into a specific number if you prefer. The [ ] in the name of the columns are just for visualization purpose as in some cases the names of the columns can appear too close and be hard to read. In this code the mathematical operations just use the original columns, if you would like to use also the new created columns you should delete the original_columns variable and take the df.columns inside the loop. Also consider that with this code all the operations happen from left to right with pairs of two numbers, that means that if you see A + B * C the order will be:

1. A + B

2. (A+B) * C

   NUM_OF_COLUMNS = 3
   
   data = np.random.rand(10, 5)
   df = pd.DataFrame(data, columns=['A', 'B', 'C', 'D', 'E'])
   
   operations = [operator.add, operator.sub, operator.mul, operator.truediv]
   operations_dict = {'add':'+','sub':'-','mul':'*','truediv':'/'}
   original_columns = df.columns
   
   for _ in range(NUM_OF_COLUMNS):
   
       num_operators = random.randint(2,5)
       selected_columns = random.sample(list(original_columns), num_operators)
       selected_operations = [random.choice(operations) for _ in range(num_operators-1)]
   
       new_column = df[selected_columns[0]] 
       column_name = f"[{selected_columns[0]}"
       for i in range(1, len(selected_columns)):
           new_column = selected_operations[i-1](new_column, df[selected_columns[i]])
           column_name += f" {operations_dict[selected_operations[i-1].__name__]} {selected_columns[i]}"
       column_name += "]"
       df[column_name] = new_column
   
   print(df)
   

Output:

        A         B         C         D         E    [B / A * C]  [B / C * D + E]   [C - B]
0  0.133616  0.518619  0.681410  0.903823  0.971338     2.644830         1.659235  0.162791
1  0.968481  0.363165  0.274024  0.009211  0.821189     0.102755         0.833397 -0.089140
2  0.267062  0.197164  0.187881  0.862648  0.151512     0.138707         1.056784 -0.009283
3  0.460873  0.841071  0.150538  0.469626  0.424029     0.274725         3.047882 -0.690534
4  0.172947  0.602186  0.004098  0.937253  0.093663     0.014267       137.835107 -0.598088
5  0.015732  0.502398  0.380856  0.593829  0.725551    12.162291         1.508887 -0.121542
6  0.606476  0.854250  0.235426  0.306123  0.205574     0.331609         1.316350 -0.618825
7  0.220014  0.415449  0.418816  0.788880  0.001986     0.790843         0.784525  0.003366
8  0.479869  0.480209  0.587922  0.318358  0.116142     0.588339         0.376173  0.107713
9  0.927947  0.261066  0.225387  0.910492  0.268781     0.063410         1.323409 -0.035680