o Uݢg@sdZddlmZddlmZddlZddlZddl m m Z er/ddl m Z mZddlmZdd d Zdd dZdddZ ddddZdS)zGFunctions to compute various statistical operations on sparse matrices.) annotations) TYPE_CHECKINGN) csc_matrix csr_matrix) CountMatrixmatrixrreturncsc_matrix | csr_matrixcCsB|}tdt|}||}|jtj}t |||S)Ng?) Zget_counts_per_bcmaxnpmedianmcopyastypefloat64 sparsefuncsinplace_column_scale)rZ counts_per_bcZmedian_counts_per_bcZscaling_factorsr rh/oak/stanford/groups/akundaje/marinovg/programs/cellranger-9.0.1/lib/python/cellranger/analysis/stats.pynormalize_by_umis  rc Cs|}t|jdtd|}|jtj}t ||t t t |j dddk}|D]}d|d|f<q5|S)zPerform feature normalization.raxisgV瞯<)Zget_numbcs_per_featurer logbcs_dimr rrrrinplace_row_scalewheresqueezeasarraysum)rZnumbcs_per_featureZscaling_factors_rowr ZzeroedZbc_ixrrrnormalize_by_idf!s " r tuple[np.ndarray, np.ndarray]cCs*tj|dd\}}t|gt|gfS)zICalculate mean and variance of each column, in a sparsity-preserving way.rr)rmean_variance_axisr array)rmuvarrrrsummarize_columns:sr&mat_mean np.ndarraymat_varnbinsintc Cs||t|}t|tddd|}t||}t|}t|dkr+|Stj j ||d|d\}}}||d}t ||} tj j || d|d\} }}| |d} ||| } | S)zCalculates the normalized dispersion. The dispersion is calculated for each feature and then normalized to see how its dispersion compares to samples that had a similar mean value. rdrr ) statisticbins) r square percentilearangeappendr uniquelenscipystatsbinned_statisticabs) r(r*r+Zmat_disp quantilesZ disp_meds_Z disp_binsZ disp_meds_arrZ disp_abs_devZ disp_madsZ disp_mads_arrZ disp_normrrrget_normalized_dispersion@s"       r<)rrrr )rr rr!)r')r(r)r*r)r+r,rr))__doc__ __future__rtypingrnumpyr scipy.statsr6Zsklearn.utils.sparsefuncsutilsr scipy.sparserrcellranger.matrixrrr r&r<rrrrs