GRDC - CCDM/CIC genomic prediction report

<strong> S1Data-simulated.tar.gz </strong>- These are simulated data evaluating prediction accuracy for different traits, number of markers, numbers of samples on different testing scenarios. <br> <br> Other files are example results of the SelectML methods. These are the combined results of the optimise and predict scripts. The compressed folders are named by the simulated dataset that they correspond to and the model. <br> The trait is first (e.g. A1 means an additive trait with 1 causal marker), N1000 means 1000 samples (in the training population), M1000 means 1000 markers sampled, "_CAUSAL" means that the causal loci were included in the sampled markers (note that this means for M1000 all markers sampled did have a genuine effect), and the final section before ".tar.gz" indicates the model used (e.g. sgd, xgb, BGLR). <br> Inside each of these compressed folders are the following files. From the `selectml optimise` command: <br> <strong>regression_*_best.json </strong>- the best performing combination of hyper parameters for these data and model type. <strong>regression_*_results.tsv </strong>- The optuna running logs showing sampled parameters and average mean squared error (of cross validated samples) of models from that parameter set. <strong>regression_*_full_results.tsv</strong> - Like _results.tsv but includes other statistics relevant to the task, such as pearsons correlation. <br> And from `selectml predict`: <strong>regression_*_model.pkl </strong>- a stored version of the trained model given the best parameters from optimise, trained from the complete train dataset. <strong>regression_*_predictions.tsv </strong>- predicted results for all training datasets. <strong>regression_*_stats.tsv </strong>- summary statistics (e.g. MSE, pearsons correlation) for the model in different test populations.<strong>S1Data-simulated.tar.gz</strong> - 该压缩包内含用于评估不同性状、标记数量及样本量在各类测试场景下预测精度的模拟数据集。 其余文件为SelectML方法的示例结果，系优化脚本与预测脚本的整合输出产物。此类压缩文件夹以其对应的模拟数据集与所用模型进行命名。 命名规则如下：性状名称置于最前（例如A1代表包含1个因果标记的加性性状），N1000代表训练群体含1000个样本，M1000代表采样得到1000个标记，"_CAUSAL"表示因果位点已包含在采样标记中（需注意，这意味着在M1000的场景下，所有采样标记均具有真实效应）；".tar.gz"后缀前的最后一段标识代表所用模型（例如sgd、xgb、BGLR）。 每个压缩文件夹内均包含以下文件： 从`selectml optimise`命令生成的文件： <strong>regression_*_best.json</strong>：针对当前数据集与模型类型的最优超参数（hyper parameters）组合文件。 <strong>regression_*_results.tsv</strong>：Optuna运行日志，记录了采样参数与该参数集下模型交叉验证样本的平均均方误差（Mean Squared Error, MSE）。 <strong>regression_*_full_results.tsv</strong>：与_results.tsv格式类似，但包含更多与任务相关的统计指标，例如皮尔逊相关系数（Pearson Correlation Coefficient）。 从`selectml predict`命令生成的文件： <strong>regression_*_model.pkl</strong>：基于优化步骤得到的最优参数，使用完整训练数据集训练得到的已保存模型文件。 <strong>regression_*_predictions.tsv</strong>：所有训练数据集的预测结果文件。 <strong>regression_*_stats.tsv</strong>：不同测试群体下模型的汇总统计指标（例如均方误差、皮尔逊相关系数）。