testing_dewakss.py

import os
import numpy as np
import scipy as sp
import pandas as pd
import scanpy as sc
import scprocessing.pipeline as scpipe
import scprocessing.preprocessing as scpp
import scprocessing.preprocessing.Svensson2019 as Svensson2019
import scprocessing.plotting.anndata as scpl
import dewakss.decomposition as dede
import dewakss.denoise as dewakss

datadir = '../data/master/Paul2015'

adata = sc.read(os.path.join(datadir, "loaded_data.h5ad"))

adata.obs['n_counts'] = adata.X.sum(axis=1).A1
adata.obs['n_genes'] = adata.X.astype(bool).sum(axis=1).A1

sc.pp.filter_cells(adata, min_counts=1000)
sc.pp.filter_genes(adata, min_cells=10)
adata.layers['counts'] = adata.X.copy()
# sc.pp.normalize_per_cell(adata, counts_per_cell_after=np.median(adata.obs['n_counts']))
sc.pp.normalize_per_cell(adata)
dede.ftt(adata)
adata.raw = adata

# ======================================================================
decay = [1]
set_diags = [0]
modes = ['connectivities', 'distances']
denoisetypes = ['mean']
# denoisetypes = ['median']
# symmetrize = [True, False]
symmetrize = [True, False]
hyperp = {}
n = 0
iterations = 1
# neigbours = [3, 5, 10, 20, 50, 75, 100, 125, 150]  # , 200]
neigbours = [20, 50, 75, 100, 125, 150, 200]
# neigbours = [100, 200]
# npcss = [dpca.optimal_, 5, 10, 20, 50, 100, 200, 0]
npcss = [5, 10, 15, 20, 25, 50, 100, 200]
# npcss = [13, 5, 10, 20, 50, 100]
# npcss = [200, 0]

sc.pp.pca(adata, n_comps=max(npcss), random_state=0)
hyperp = []
for denoiset in denoisetypes:
    for N in neigbours:
        for pcs in npcss:
            adata = adata.copy()
            sc.pp.neighbors(adata, n_neighbors=N, n_pcs=pcs)
            for d in decay:
                for m in modes:
                    for s in symmetrize:
                        for I in set_diags:
                            print('decay', 'mode', 'sym', 'diag', 'N', 'nPC')
                            print(d, m, s, I, N, pcs, denoiset)

                            dewaxer = dewakss.DEWAKSS(adata, iterations=iterations, init_diag=I, set_diag=(I if I == 0 else None), run2best=False, denoise_type=denoiset, decay=d, mode=m, symmetrize=s, verbose=False, max_dense=1.0)

                            dewaxer.fit(adata)

                            performance = pd.DataFrame(dewaxer.prediction_).T
                            performance.index.name = "iteration"
                            performance.columns = ['MSE', "R2"]
                            performance = performance.reset_index()
                            performance['decay'] = d
                            performance['mode'] = m
                            performance["symmetrize"] = s
                            performance["diag"] = I
                            performance['neighbors'] = N
                            performance['pcs'] = pcs
                            performance['denoisetype'] = denoiset
                            hyperp.append(performance)


performance_data = pd.concat(hyperp)
performance_data = performance_data.reset_index(drop=True)