| import torch | |
| import numpy as np | |
| import pandas as pd | |
| class PredictionAnalyzer: | |
| def __init__(self, model, df_ready): | |
| self.model = model | |
| self.df_ready = df_ready | |
| def predictions_loop(self): | |
| pred_total = [] | |
| for filter_cat in list(self.df_ready.columns): | |
| df_sample = self.df_ready[self.df_ready[filter_cat]==1] | |
| lb_sample = len(df_sample)/len(self.df_ready) | |
| X = df_sample.values | |
| X_test_tensor = torch.tensor(X, dtype=torch.float32) | |
| inputs = X_test_tensor | |
| outputs = self.model(inputs) | |
| outputs_show = outputs.detach().numpy().flatten() | |
| outputs_show[outputs_show > 0.55] = 1.0 | |
| outputs_show[outputs_show < 0.45] = 0.0 | |
| filtered_arr = outputs_show[(outputs_show == 0.0) | (outputs_show == 1.0)] | |
| unique, counts = np.unique(filtered_arr, return_counts=True) | |
| if len(counts) > 1: | |
| sum_cat = counts[0]+counts[1] | |
| pred_total.append([filter_cat, round(counts[0]/sum_cat*100, 2), round(counts[1]/sum_cat*100, 2), round(lb_sample*100, 2)]) | |
| else: | |
| pred_total.append([filter_cat, 0, 0, round(lb_sample*100, 2)]) | |
| pred_df_tot = pd.DataFrame(pred_total, columns=['Categoria', 'Pred==0', 'Pred==1', 'linea_base']) | |
| pred_df_tot_sort = pred_df_tot.sort_values(by='Pred==1', ascending=False, ignore_index=True) | |
| return pred_df_tot_sort |