results['h1'] = barcodes[1]
# H2. Need to subsample points for computational tractability if
# number of points is large (can go higher but very slow)
if len(rates) > 1500:
idx = np.random.choice(np.arange(len(rates)), 1500, replace=False)
H2_rates = rates[idx]
else:
H2_rates = rates
barcodes = tda(H2_rates, maxdim=2, coeff=2)['dgms']
results['h2'] = barcodes[2]
# save
gff.save_pickle_file(
results, save_dir + '%s_%s%s_ph_barcodes.p' % (session, state,
('_thresholded' * thrsh)))
# If plotting from a saved file, uncomment this and replace with appropriate file.
# results = gff.load_pickle_file(gen_params['results_dir'] + '2019_03_22_tda/Mouse28-140313_Wake_ph_barcodes.p')
if plot_barcode:
col_list = ['r', 'g', 'm', 'c']
h0, h1, h2 = results['h0'], results['h1'], results['h2']
# replace the infinity bar (-1) in H0 by a really large number
h0[~np.isfinite(h0)] = 100
# Plot the longest barcodes only
plot_prcnt = [99, 98, 90] # order is h0, h1, h2
to_plot = []
for curr_h, cutoff in zip([h0, h1, h2], plot_prcnt):
bar_lens = curr_h[:, 1] - curr_h[:, 0]
# Some commonly used paths
base_dir = '/Users/rchaudhuri/data/2019_03_hd/'
# Location where Peyrache et al data is unzipped
param_dict['raw_data_dir'] = base_dir + 'raw_data/'
# Contains the Peyrache et al data accumulated into single files.
# These files are created by read_in_data/preprocess_raw_data.py. Note that
# preprocess_raw_data will make this directory if it doesn't exist.
param_dict['processed_data_dir'] = base_dir + 'processed/'
# Spike counts in bins. Currently not used.
# param_dict['spike_counts_dir'] = base_dir + 'analyses/2016_06_spike_matrix_new/'
# Firing rates estimated by convolution with Gaussian kernel. Again, generated
# by read_in_data/preprocess_raw_data.py and directory created if needed.
param_dict['kernel_rates_dir'] = base_dir + 'analyses/2019_03_kernel_rates/'
# Analyses and results will be saved under this directory (and loaded from here too).
param_dict['results_dir'] = base_dir + 'analyses/'
# Colors for the plots.
param_dict['cols'] = {
'REM': (0.392, 0.549, 0.0784),
'SWS': (0.824, 0.627, 0.0392),
'Wake': (0.0118, 0.235, 0.392),
'measured': (0.3, 0.3, 0.3),
'fit': (0.490, 0.961, 0.961)
}
save_pickle_file(param_dict, 'general_params.p')
session = 'Mouse28-140313'
make_processed_files = True
make_rates = True
if make_processed_files:
data_path = gen_params['raw_data_dir'] + session + '/'
params = {
'session': 'Mouse28-140313',
'data_path': data_path,
'eeg_sampling_rate': 1250.,
'spike_sampling_interval': 1.0 / (20e3)
}
data = drf.gather_session_spike_info(params)
save_dir = gff.return_dir(gen_params['processed_data_dir'])
gff.save_pickle_file(data, save_dir + '%s.p' % session)
if make_rates:
print 'Getting kernel rates'
t0 = time.time()
sigma = 0.1
params = {'dt': 0.05, 'method': 'gaussian', 'sigma': sigma}
inp_data = gff.load_pickle_file(gen_params['processed_data_dir'] +
'%s.p' % session)
rates = rf.get_rates_and_angles_by_interval(inp_data,
params,
smooth_type='kernel',
just_wake=True)
save_dir = gff.return_dir(gen_params['kernel_rates_dir'] +
'%0.0fms_sigma/' % (sigma * 1000))
gff.save_pickle_file(rates, save_dir + '%s.p' % session)
print('Counts for each ', len(counts1), len(counts2))
nSamples = min(len(counts1), len(counts2), desired_nSamples)
print('nSamples = ', nSamples)
sel1 = counts1[:nSamples]
sel2 = counts2[:nSamples]
concat_counts = np.concatenate((sel1, sel2), 0)
proj = run_dim_red(concat_counts, params=dim_red_params, method=method)
to_save = {
'seed': sd,
state: proj[:nSamples].copy(),
state2: proj[nSamples:].copy()
}
fname = '%s_%s_kern_%dms_sigma_%dms_binsep_%s_%s_embed_%s_%ddims_%dneighbors_%d.p' % (
session, area, sigma * 1000, dt_kernel * 1000, state, state2, method,
target_dim, n_neighbors, sd)
gff.save_pickle_file(to_save, dir_to_save + fname)
print('Time ', time.time() - t0)
if to_plot:
fig = plt.figure(figsize=(8, 8))
if target_dim == 2:
ax = fig.add_subplot(111)
ax.scatter(to_save[state][:, 0],
to_save[state][:, 1],
s=10,
alpha=0.4,
color=cols[state])
if condition == 'joint':
ax.scatter(to_save[state2][:, 0],
to_save[state2][:, 1],
s=10,
else:
results[k] = [[
mse, curr_fit_result['fit_err'],
np.array(curr_fit_result['final_knots'])
]]
print 'Time ', time.time() - tic
to_save = {
'fit_results': results,
'session': session,
'area': area,
'state': state,
'embed_file': embed_fname
}
gff.save_pickle_file(
to_save, dir_to_save + '%s_%s_dim%d_trainfrac%.2f_decode_errors_sd%d.p' %
(session, state, fit_dim, train_frac, sd))
rmse_to_plot = np.sqrt([x[0] for x in results[k]])
fig = plt.figure(figsize=(8, 8))
ax = fig.add_subplot(111)
if nTests < 20:
ax.scatter(np.ones_like(rmse_to_plot), rmse_to_plot)
ax.set_xticks([1])
# ax.set_xticklabels(['Samples'])
else:
vp = ax.violinplot([rmse_to_plot],
positions=[1],
points=100,
widths=0.75,
c=(0.5, 0.5, 0.5),
s=5)
ax.plot(fit_results['loop_init_knots'][:, i0],
fit_results['loop_init_knots'][:, i1],
fit_results['loop_init_knots'][:, i2],
c='b',
lw=2)
ax.plot(fit_results['loop_final_knots'][:, i0],
fit_results['loop_final_knots'][:, i1],
fit_results['loop_final_knots'][:, i2],
c='r',
lw=2)
plt.show()
dec_angle, mse = mff.decode_from_passed_fit(data_to_decode,
fit_results['tt'][:-1],
fit_results['curve'][:-1],
ref_angles[test_idx])
to_save = {
'fit_results': fit_results,
'session': session,
'area': area,
'state': state,
'embed_file': fname if not run_dim_red_here else None
}
gff.save_pickle_file(
to_save,
dir_to_save + '%s_%s_dim%d_trainfrac%.2f_interactive_fits_sd%d.p' %
(session, state, fit_dim, train_frac, sd))
