Closed zhangliang-04 closed 5 months ago
Hi, Zhangliang Most of the functions used are functions provided by numpy, including max, min, substract, etc. The specific list is as follows: func_list = [ 'getitem', 'select', 'numpy.array', 'numpy.ndarray', 'numpy.argmax', 'numpy.argmin', 'numpy.max', 'numpy.min', 'numpy.sum', 'numpy.add', 'numpy.subtract', 'numpy.multiply', 'numpy.divide', 'numpy.<', 'numpy.<=', 'numpy.>', 'numpy.>=', 'numpy.==', 'numpy.!=', 'numpy.mean', 'numpy.median', 'numpy.std', 'numpy.var', 'numpy.abs', 'numpy.sqrt', 'numpy.square', 'numpy.log', 'numpy.exp', 'numpy.power', 'numpy.sort', 'numpy.delete', 'numpy.all', 'numpy.any', 'numpy.diff', 'numpy.corrcoef', 'numpy.cov', 'np.array', 'np.ndarray', 'np.argmax', 'np.argmin', 'np.max', 'np.min', 'np.sum', 'np.add', 'np.subtract', 'np.multiply', 'np.divide', 'np.<', 'np.<=', 'np.>', 'np.>=', 'np.==', 'np.!=', 'np.mean', 'np.median', 'np.std', 'np.var', 'np.abs', 'np.sqrt', 'np.square', 'np.log', 'np.exp', 'np.power', 'np.sort', 'np.delete', 'np.all', 'np.any', 'np.diff', 'np.corrcoef', 'np.cov', 'max', 'min', 'sum', 'len', 'str', 'int', 'float', 'abs', 'round', '<', '<=', '>', '>=', '==', '!=' ]
During testing, we first use ChartAst to generate executable json, and then execute this json through the backend and call the corresponding function to get the calculated answer. In this way, the model can avoid most calculation problems; at the same time, compared to Directly train the answer end-to-end. The training data required by this method will be much smaller than the end-to-end case.
Thank you for the great work! After reading the paper carefully, I have the following questions: