def es_predict_sum(count_list, virtual_info, double_or_triple, alpha): # temp_list = [] es_predict_result = {} for flavor, info in virtual_info.items(): temp_list1 = [] temp_list2 = [] temp_list3 = [] for i in range(len(count_list)): temp_list1.append(count_list[i][flavor - 1]) temp_a = Tool.mid([count_list[j][flavor - 1] for j in range(i - 2, i + 2) if 0 <= j < len(count_list)]) temp_list2.append( math.log( int(temp_a) + 1)) temp_list3.append(count_list[i][flavor - 1]) temp_list1 = temp_list1[::-1] temp_list2 = temp_list2[::-1] es_list3 = Tool.line_diff(temp_list3[::-1]) if double_or_triple == 2: a1, b1 = compute_double(alpha, temp_list1) result1 = max(int(a1[-1] + b1[-1] * 1), 0) a2, b2 = compute_double(alpha, temp_list2) result2 = max(math.exp(a2[-1] + b2[-1] * 1) - 1, 0) a3, b3 = compute_double(alpha, es_list3) aaa = int(a3[-1] + b3[-1] * 1) result3 = max(temp_list3[0] + aaa, 0) es_predict_result[flavor] = max(0.33 * result1 + 0.33 * result2 + 0.33 * result3, 0) elif double_or_triple == 3: a1, b1, c1 = compute_triple(alpha, temp_list1) result1 = max(int(a1[-1] + b1[-1] * 1 + c1[-1] * (1 ** 2)), 0) a2, b2, c2 = compute_triple(alpha, temp_list2) result2 = max(math.exp(a2[-1] + b2[-1] * 1 + c2[-1] * (1 ** 2)) - 1, 0) a3, b3, c3 = compute_triple(alpha, es_list3) bbb = int(a3[-1] + b3[-1] * 1 + c3[-1] * (1 ** 2)) result3 = max(temp_list3[0] + bbb, 0) es_predict_result[flavor] = max(0.33 * result1 + 0.33 * result2 + 0.33 * result3, 0) return es_predict_result
def es_predict_diff(count_list, virtual_info, double_or_triple, alpha): # temp_list = [] es_predict_result = {} for flavor, info in virtual_info.items(): temp_list = [] for i in range(len(count_list)): temp_list.append(count_list[i][flavor - 1]) es_list = Tool.line_diff(temp_list[::-1]) if double_or_triple == 2: a, b = compute_double(alpha, es_list) aaa = int(a[-1] + b[-1] * 1) es_predict_result[flavor] = max(temp_list[0] + aaa, 0) elif double_or_triple == 3: a, b, c = compute_triple(alpha, es_list) bbb = int(a[-1] + b[-1] * 1 + c[-1] * (1 ** 2)) es_predict_result[flavor] = max(temp_list[0] + bbb, 0) return es_predict_result
def predict(lg_count_list, lr_count_list, rf_count_list, es_count_list, virtual_info, lg_round, mix_rf, mix_lr, mix_es, rf_day_gap, lr_day_gap, es=3, alpha=0.5, seed=1000, floor=0.0, rf_diff=0): print '-----start predict-----' lg_predict_result = {} rf_predict_result = {} lr_predict_result = {} es_predict_result = {} for flavor, info in virtual_info.items(): # ----------- 拉格朗日 ---------- lg_window = [] for i in range(lg_round): lg_window.append( # Tool.mid([lg_count_list[j][flavor - 1] for j in range(i - 7, i + 7) if 0 <= j < lg_round])) Tool.mid([ lg_count_list[j][flavor - 1] for j in range(i - 3, i + 3) if 0 <= j < lg_round ])) # window_list = [Tool.mean(lg_window[15:29]), Tool.mean(lg_window[0:29]), Tool.mean(lg_window[0:14])] window_list = [ Tool.mean(lg_window[4:7]), Tool.mean(lg_window[0:7]), Tool.mean(lg_window[0:3]) ] lg_predict_result[flavor] = max( int(Tool.LG(3, list(range(0, 3)), window_list)), 0) # ----------- 随机森林 ---------- rf_predict_list = [] temp_list = [] for i in range(len(rf_count_list)): # temp_list.append(Tool.mid([rf_count_list[j][flavor - 1] for j in range(i - 4, i + 4) if 0 <= j < len(rf_count_list)])) temp_list.append(rf_count_list[i][flavor - 1]) test = temp_list[:rf_day_gap - 1][::-1] diff_list = Tool.line_diff(temp_list) if rf_diff == 0: for j in range(len(rf_count_list) - rf_day_gap): rf_predict_list.append(temp_list[j:j + rf_day_gap][::-1]) my_labels = [i for i in range(rf_day_gap)] rf_result = rf_predict(rf_predict_list, my_labels, test, seed) rf_predict_result[flavor] = rf_result else: for j in range(len(rf_count_list) - rf_day_gap): rf_predict_list.append(diff_list[j:j + rf_day_gap][::-1]) my_labels = [i for i in range(rf_day_gap)] rf_result = rf_predict(rf_predict_list, my_labels, test, seed) rf_predict_result[flavor] = rf_result + temp_list[-1] # ----------- 线性回归 ---------- lr_window = [] lr_data = [] for i in range(20 - lr_day_gap + 1): for j in range(1, lr_day_gap): # lr_window.append(lr_count_list[i + j][int(flavor) - 1]) lr_window.append( Tool.mid([ lr_count_list[i + jj][int(flavor) - 1] for jj in range(j - 3, j + 3) if 0 <= jj < lr_day_gap ])) # lr_window.append(Tool.mid([lr_count_list[i+jj][int(flavor) - 1] for jj in range(j - 3, j + 3) if 0 <= jj < lr_day_gap])) lr_window.append(lr_count_list[i][int(flavor) - 1]) lr_data.append(lr_window) lr_window = [] # print lr_data w = Tool.zeros(lr_day_gap) w = linear_regression(w, lr_data, 0.02, 700) # 0.03 500 x = [lr_data[0][lr_day_gap - 1]] + lr_data[0][:lr_day_gap - 2] x = x + [u * u for u in x] max_x = max(x) min_x = min(x) if max_x - min_x > 0: x = [1] + [(i - (sum(x) / len(x))) / (max_x - min_x) for i in x] else: x = [1] + x # lr_predict_result[flavor] = max(int(LR.w_mul_x(w, x)), 0) lr_predict_result[flavor] = max(w_mul_x(w, x), 0) # ---------- 指数平滑 ------------- es_predict_result = es_predict(es_count_list, virtual_info, es, alpha) print 'LG预测结果:{}'.format(lg_predict_result) print 'LR预测结果:{}'.format(lr_predict_result) print 'RF预测结果:{}'.format(rf_predict_result) print 'ES预测结果:{}'.format(es_predict_result) predict_result = lg_predict_result.copy() for key, value in predict_result.items(): # predict_result[key] = int(MIX_NUM*lg_predict_result[key] + (1-MIX_NUM)*lr_predict_result[key]) # 向下取整 predict_result[key] = int( mix_lr * lr_predict_result[key] + mix_rf * rf_predict_result[key] + mix_es * es_predict_result[key] + (1 - mix_lr - mix_rf - mix_es) * lg_predict_result[key] + floor) # 四舍五入 # predict_result[key] = int( # mix_lr * lr_predict_result[key] + mix_rf * rf_predict_result[key] + mix_es * es_predict_result[key] + # (1 - mix_lr - mix_rf - mix_es) * lg_predict_result[key]) # 地板除 print '最终预测结果:{}'.format(predict_result) print '-----end predict-----' return predict_result