def plot_perf():
import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
# SOME ERROR WERE HERE CORRECTED 27/04/2014 think its good
#INPUT_vbm = "/home/ed203246/mega/data/2015_logistic_nestv/adni/MCIc-CTL/MCIc-CTL_cs.csv"
INPUT = os.path.join(WD, "MCIc-CTL_csi_modselectcv.csv")
y_col = 'recall_mean'
x_col = 'tv'
y_col = 'auc'
a = 0.01
#color_map = {0.:'#D40000', 0.01: 'black', 0.1:'#F0a513', 0.5:'#2CA02C', 0.9:'#87AADE', 1.:'#214478'}
color_map = {0.:'#D40000', 0.01:'#F0a513', 0.1:'#2CA02C', 0.5:'#87AADE', .9:'#214478', 1.: 'black'}
# reds dark => brigth, green blues: brigth => dark
input_filename = INPUT
#input_filename = INPUTS[data_type]["filename"]
outut_filename = input_filename.replace(".csv", "_%s.pdf" % y_col)
#print outut_filename
# Filter data
data = pd.read_csv(input_filename)
#data.l1l2_ratio = data.l1l2_ratio.round(5)
# avoid poor rounding
data.l1l2_ratio = np.asarray(data.l1l2_ratio).round(3)
data.tv = np.asarray(data.tv).round(5)
data.a = np.asarray(data.a).round(5)
data = data[data.k == -1]
data = data[data.l1l2_ratio.isin([0, 0.01, 0.1, 0.5, 0.9, 1.])]
data = data[(data.tv >= 0.1) | (data.tv == 0)]
def close(vec, val, tol=1e-4):
return np.abs(vec - val) < tol
assert np.sum(data.l1l2_ratio == 0.01) == np.sum(close(data.l1l2_ratio, 0.01))
#data = data[data.a <= 1]
# for each a, l1l2_ratio, append the last point tv==1
last = list()
for a_ in np.unique(data.a):
full_tv = data[(data.a == a_) & (data.tv == 1)]
for l1l2_ratio in np.unique(data.l1l2_ratio):
new = full_tv.copy()
new.l1l2_ratio = l1l2_ratio
last.append(new)
#
last = pd.concat(last)
data = pd.concat([data, last])
data.drop_duplicates(inplace=True)
#
from brainomics.plot_utilities import plot_lines
figures = plot_lines(df=data,
x_col=x_col, y_col=y_col, colorby_col='l1l2_ratio',
splitby_col='a', color_map=color_map)
pdf = PdfPages(outut_filename)
for fig in figures:
print fig, figures[fig]
pdf.savefig(figures[fig]); plt.clf()
pdf.close()
for global_pen in global_pens for tv_ratio in tv_ratios
for l1_ratio in l1_ratios])
df = pd.DataFrame(index, columns=['global_pen', 'tv_ratio', 'l1_ratio'])
def metric_fun(r):
return (r['tv_ratio'] + r['l1_ratio']) * r['global_pen']
df['metric'] = df.apply(metric_fun, axis=1)
# Test
handles = plot_utilities.plot_lines(df,
x_col='tv_ratio',
y_col='metric',
splitby_col='global_pen',
colorby_col='l1_ratio',
use_suptitle=False,
use_subplots=True)
plt.show()
# Test with index
df.set_index(['global_pen', 'tv_ratio', 'l1_ratio'], inplace=True)
handles = plot_utilities.plot_lines(df,
x_col=1,
y_col='metric',
splitby_col=0,
colorby_col=2,
use_suptitle=False,
use_subplots=True)
print "Found", n_indiv, "persons"
###############################################################################
# Plot metrics #
###############################################################################
L1_RATIOS = [0.0, 0.5, 1.0]
l1_ratio_filter = lambda v: v in L1_RATIOS
df_noindex = df.reset_index()
struct_pca_df = df_noindex.loc[(df_noindex.model == 'struct_pca')
& (df_noindex.l1_ratio.apply(l1_ratio_filter))]
for metric in METRICS:
handles = plot_utilities.plot_lines(struct_pca_df,
x_col="tv_ratio",
y_col=metric,
splitby_col="global_pen",
colorby_col="l1_ratio")
for val, handle in handles.items():
filename = OUTPUT_CURVE_FILE_FORMAT.format(metric=metric,
global_pen=val)
handle.savefig(filename)
###############################################################################
# Components #
###############################################################################
# Load components
components = np.zeros((len(COND), np.prod(IM_SHAPE), N_COMP))
for j, (params, name) in enumerate(COND):
key = '_'.join([str(param) for param in params])
# Subsample df
struct_pca_df = df[df.model == EXAMPLE_MODEL]
# GroupBy SNR
snr_groups = struct_pca_df.groupby('snr')
# Summary per SNR value (pivot the table to have better display)
for metric, metric_name in zip(METRICS, METRICS_NAME):
summary = pd.DataFrame(snr_groups[metric].describe()).unstack(1)[0]
filename = TAB_FILE_FORMAT.format(metric=metric)
summary.to_latex(open(filename, 'w'))
# Plot some metrics for struct_pca for each SNR value
for snr_val, snr_group in snr_groups:
for metric, metric_name in zip(METRICS, METRICS_NAME):
handles = plot_utilities.plot_lines(snr_group,
x_col='tv_ratio',
y_col=metric,
splitby_col='global_pen',
colorby_col='l1_ratio',
use_suptitle=False)
for val, handle in handles.items():
# Tune the figure
ax = handle.get_axes()[0]
ax.set_xlabel("TV ratio")
ax.set_ylabel(metric_name)
l = ax.get_legend()
l.set_title("$\ell_1$ ratio")
s = r'$ \alpha $ =' + str(val)
handle.suptitle(r'$ \alpha $ =' + str(val))
filename = CURVE_FILE_FORMAT.format(metric=metric,
snr=snr_val,
global_pen=val)
handle.savefig(filename)
# Subsample it & add a column based on name
summary = df.loc[PARAMS][METRICS]
name_serie = pd.Series([item[1] for item in COND], name='Name', index=PARAMS)
summary['name'] = name_serie
# Write in a CSV
summary.to_csv(OUTPUT_RESULTS_FILE)
# Subsample df
struct_pca_df = df.xs(EXAMPLE_MODEL)
# Plot some metrics for struct_pca
for metric in METRICS:
handles = plot_utilities.plot_lines(struct_pca_df,
x_col=1,
y_col=metric,
splitby_col=0,
colorby_col=2)
for val, handle in handles.items():
filename = CURVE_FILE_FORMAT.format(metric=metric, global_pen=val)
handle.savefig(filename)
#####################################################################
# Loading as images #
# Projection to csv #
# eventually flip Projection and Loading such that they positivetly #
# correlate with clinical severity #
#####################################################################
# Open mask
mask_ima = nib.load(INPUT_MASK)
