def experiment_2b(W_init, H_init, results_folder, ranks, no_iterations):
for r in ranks:
W_init_r = W_init[:, :r]
H_init_r = H_init[:r, :]
W, H, divergence_by_it = NMF_divergence(data_matrix, W_init_r, H_init_r, n, m, r, no_iterations, 1,
report_progress=True,
save_progress_to=results_folder)
save_results(results_folder, W, H, unique_name='r={}_final'.format(r), additional_data_name_to_array_dict={
'divergence_record' : divergence_by_it}, row_names_list=genes, column_names_list=patients)
def experiment_2c(ranks, no_iterations, rpkm_matrix_path, ensg_to_go_path,
results_folder):
genes_with_GO_annotations = list(
set(pd.read_csv(ensg_to_go_path, index_col=0)['ENSEMBL']))
rpkm_matrix = pd.read_csv(rpkm_matrix_path,
index_col=0).loc[genes_with_GO_annotations]
patients = list(rpkm_matrix.columns)
n = len(genes_with_GO_annotations)
m = len(patients)
rpkm_matrix = np.array(rpkm_matrix)
W_init, H_init = NNDSVDar_initialization(rpkm_matrix, n, m, max(ranks))
for r in ranks:
W_init_r = W_init[:, :r]
H_init_r = H_init[:r, :]
W, H, divergence_by_it = NMF_divergence(rpkm_matrix,
W_init_r,
H_init_r,
n,
m,
r,
no_iterations,
100,
report_progress=True)
save_results(results_folder,
W,
H,
unique_name='r={}_final'.format(r),
additional_data_name_to_array_dict={
'divergence_record': divergence_by_it
},
row_names_list=genes_with_GO_annotations,
column_names_list=patients)
def experiment_2d(ranks, no_iterations, rpkm_matrix_path,
genes_with_GO_annotations, results_folder, repeat_no):
rpkm_matrix = pd.read_csv(rpkm_matrix_path,
index_col=0).loc[genes_with_GO_annotations]
patients = list(rpkm_matrix.columns)
n = len(genes_with_GO_annotations)
m = len(patients)
rpkm_matrix = np.array(rpkm_matrix)
random_state = np.random.RandomState(repeat_no)
for r in ranks:
print('\n\nRANK {}, REPEAT {}\n\n'.format(r, repeat_no))
W_init = random_state.uniform(size=(n, r))
H_init = random_state.uniform(size=(r, m))
W, H, divergence_by_it = NMF_divergence(rpkm_matrix,
W_init,
H_init,
n,
m,
r,
no_iterations,
100,
report_progress=True)
save_results(results_folder,
W,
H,
unique_name='r={}_repeat={}'.format(r, repeat_no),
additional_data_name_to_array_dict={
'divergence_record': divergence_by_it
},
row_names_list=genes_with_GO_annotations,
column_names_list=patients)
np.save(initializations_path + 'W_init_' + code, W_init)
np.save(initializations_path + 'H_init_' + code, H_init)
# REAL NMF
W, H, divergence_by_it = NMF_divergence(V_15000,
W_init.copy(),
H_init.copy(),
n,
m,
r,
iterations,
divergence_calc_frequency,
report_progress=True)
save_results(W,
H, {'divergence_record': divergence_by_it},
row_names_list=genes,
column_names_list=patients,
main_results_folder=results_path_main)
# PERMUTATION NMF
W, H, divergence_by_it = NMF_divergence(V_15000_columns_permuted,
W_init,
H_init,
n,
m,
r,
iterations,
divergence_calc_frequency,
report_progress=True)
save_results(W,
H, {'divergence_record': divergence_by_it},
data_path = "C:/Users/hanne/Documents/PROJECT/Project Data/CM_experiment_2_data.csv"
main_results_folder = "C:/Users/hanne/Documents/PROJECT/Project Data/Experiment_2a_results/"
data_matrix = pd.read_csv(data_path, index_col=0)
genes = list(data_matrix.index)
patients = list(data_matrix.columns)
n = len(genes)
m = len(patients)
ranks = [2, 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 100, 150]
data_matrix = np.array(data_matrix)
for r in ranks:
random = np.random.RandomState(42)
W_init = random.uniform(0, 1, (n, r))
H_init = random.uniform(0, 1, (r, m))
W, H, divergence_by_it = NMF_divergence(data_matrix, W_init, H_init, n, m, r, 5000, 1, report_progress=True,
save_progress_to=main_results_folder)
save_results(main_results_folder, W, H, unique_name='r={}_final'.format(r), additional_data_name_to_array_dict={
'divergence_record' : divergence_by_it}, row_names_list=genes, column_names_list=patients)
r = ranks[idx]
repeat = best_repeats[idx]
# Loading best initialisations by divergence after 100 iterations as initialisation for longer run
code = 'rep={}_rank={}'.format(repeat, r)
print(code)
W_init = np.load(initializations_path + 'W_init_' + code + '.npy')
H_init = np.load(initializations_path + 'H_init_' + code + '.npy')
# REAL NMF
W, H, divergence_by_it = NMF_divergence(V_15000,
W_init.copy(),
H_init.copy(),
n,
m,
r,
iterations,
divergence_calc_frequency,
report_progress=True)
save_results(results_path_main,
W,
H,
unique_name='rank={}'.format(r),
additional_data_name_to_array_dict={
'divergence_record': divergence_by_it
},
row_names_list=genes,
column_names_list=patients)