experiment.py 文件源码

def calc_scores(self, label):
        """
        Combine the scores and standard errors within each condition.
        """
        if self.check_store("/main/{}/scores".format(label)):
            return

        logging.info("Calculating per-condition scores ({})".format(label),
                     extra={'oname': self.name})

        # set up new data frame
        shared_index = self.store.select("/main/{}/scores_shared"
                                         "".format(label),
                                         "columns='index'").index
        columns = pd.MultiIndex.from_product([sorted(self.child_names()),
                                              sorted(["score", "SE",
                                                      "epsilon"])],
                                             names=["condition", "value"])
        data = pd.DataFrame(np.nan, index=shared_index, columns=columns)
        del shared_index
        del columns

        # set up local variables
        idx = pd.IndexSlice

        score_df = self.store.select("/main/{}/scores_shared".format(label))
        if self.scoring_method == "simple":
            # special case for simple ratios that have no SE
            # calculates the average score
            for cnd in score_df.columns.levels[0]:
                data.loc[:, idx[cnd, 'score']] = \
                    score_df.loc[:, idx[cnd, :, 'score']].mean(axis=1)
        else:
            for cnd in score_df.columns.levels[0]:
                y = np.array(score_df.loc[:, idx[cnd, :, 'score']].values).T
                sigma2i = \
                    np.array(score_df.loc[:, idx[cnd, :, 'SE']].values ** 2).T

                # single replicate of the condition
                if y.shape[0] == 1:
                    data.loc[:, idx[cnd, 'score']] = y.ravel()
                    data.loc[:, idx[cnd, 'SE']] = np.sqrt(sigma2i).ravel()
                    data.loc[:, idx[cnd, 'epsilon']] = 0.
                # multiple replicates
                else:
                    betaML, var_betaML, eps = rml_estimator(y, sigma2i)
                    data.loc[:, idx[cnd, 'score']] = betaML
                    data.loc[:, idx[cnd, 'SE']] = np.sqrt(var_betaML)
                    data.loc[:, idx[cnd, 'epsilon']] = eps

                # special case for normalized wild type variant
                if self.logr_method == "wt" and WILD_TYPE_VARIANT in \
                        data.index:
                    data.loc[WILD_TYPE_VARIANT, idx[:, 'SE']] = 0.
                    data.loc[WILD_TYPE_VARIANT, idx[:, 'score']] = 0.
                    data.loc[WILD_TYPE_VARIANT, idx[:, 'epsilon']] = 0.

        # store the data
        self.store.put("/main/{}/scores".format(label), data, format="table")