o .wi @sddlZddlZddlmZddlmZddlmZddlmZdededed ed ed ed ed ededede de eeeeeeeeffddZ d ed ed ed edededefddZ dededefddZ dS)N)Tensor) _check_data_shape_to_num_outputs)rank_zero_warn)_check_same_shapepredstargetmean_xmean_y max_abs_dev_x max_abs_dev_yvar_xvar_ycorr_xy num_prior num_outputsreturnc Cst||t||| |jd} |d} |d} | |}| |}| | }||| |}||| |}| dkrU| |}| |}|||}|||}|||}nA|| }|| }|dd}|dd}||d}| | |}|||d|}|||d|}|||||}t|tj||ddd}t|tj||ddd}||||||||fS)aUpdate and returns variables required to compute Pearson Correlation Coefficient. Check for same shape of input tensors. Args: preds: estimated scores target: ground truth scores mean_x: current mean estimate of x tensor mean_y: current mean estimate of y tensor max_abs_dev_x: current maximum absolute value of x tensor max_abs_dev_y: current maximum absolute value of y tensor var_x: current variance estimate of x tensor var_y: current variance estimate of y tensor corr_xy: current covariance estimate between x and y tensor num_prior: current number of observed observations num_outputs: Number of outputs in multioutput setting r)dim) rrshapemeansumtorchmaximummaxabs)rrrr r r r r rrrnum_obs batch_mean_x batch_mean_ydelta_xdelta_yn_totalmx_newmy_newdelta2_xdelta2_ypreds_centeredtarget_centered batch_var_x batch_var_y batch_cov_xy correctionr,g/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/torchmetrics/functional/regression/pearson.py_pearson_corrcoef_updates8       ""r.nbc Cs||d}||d}||d}|jtjkr(|jtdkr(|}|}|t|d}|t|d}|||}tt|jj }||k sl||k slt | slt | slt | rvt d|jdt ||k||kBt |Bt |Bt |B}tj|td|j|jd}|} | r|| || || |j|| <t|dd }|S) aCompute the final pearson correlation based on accumulated statistics. Args: max_abs_dev_x: maximum absolute value of x tensor max_abs_dev_y: maximum absolute value of y tensor var_x: variance estimate of x tensor var_y: variance estimate of y tensor corr_xy: covariance estimate between x and y tensor nb: number of observations rcpuzThe variance of predictions or target is close to zero. This can cause instability in Pearson correlationcoefficient, leading to wrong results. Consider re-scaling the input if possible or computing using alarger dtype (currently using z.). Setting the correlation coefficient to nan.nan)devicedtypegg?)r4rfloat16r3bfloat16powmathsqrtfinfoepsanyisfiniter UserWarning full_likefloatsqueezetoclamp) r r r r rr/bound zero_var_maskcorrcoef valid_maskr,r,r-_pearson_corrcoef_compute[sF      2&rHc Cs|jdkr |jdnd}tj||j|jd}|||}}}|||}}} ||} } t||||| | |||| |jdkrMdn|jdd \} } } } }}}} t| | |||| S)aCompute pearson correlation coefficient. Args: preds: estimated scores target: ground truth scores Example (single output regression): >>> from torchmetrics.functional.regression import pearson_corrcoef >>> target = torch.tensor([3, -0.5, 2, 7]) >>> preds = torch.tensor([2.5, 0.0, 2, 8]) >>> pearson_corrcoef(preds, target) tensor(0.9849) Example (multi output regression): >>> from torchmetrics.functional.regression import pearson_corrcoef >>> target = torch.tensor([[3, -0.5], [2, 7]]) >>> preds = torch.tensor([[2.5, 0.0], [2, 8]]) >>> pearson_corrcoef(preds, target) tensor([1., 1.]) rr)r4r3) rrrr r r r r rrr) ndimrrzerosr4r3cloner.rH) rrd_temprr r r rr/r r _r,r,r-pearson_corrcoefs& rP)r8rr(torchmetrics.functional.regression.utilsrtorchmetrics.utilitiesrtorchmetrics.utilities.checksrinttupler.rHrPr,r,r,r-s^         C :