print __doc__

if msg is not None:

print msg

# return a Bunch

print argv

if len(argv) not in (1, 2, 3):

usage('invalid number of arguments')

pcl = ParseCommandLine(argv)

arg = Bunch(

base_name='chart-05',

data=pcl.has_arg('--data'),

test=pcl.has_arg('--test'),

)

if len(argv) == 3:

both = arg.data and arg.test

if not both:

usage('there is an extra invocation option')

random_seed = 123

random.seed(random_seed)

dir_working = Path().dir_working()

debug = False

reduced_file_name = ('test-' if arg.test else '') + 'data.pickle'

# assure output directory exists

dir_path = dir_working + arg.base_name + '/'

if not os.path.exists(dir_path):

os.makedirs(dir_path)

return Bunch(

arg=arg,

debug=debug,

path_in_ege=dir_working + 'valgbr/*.pickle',

path_reduction=dir_path + reduced_file_name,

path_chart_base=dir_path,

random_seed=random_seed,

test=arg.test,

'write one txt file for each n_months_back'

def make_subplot(test_period, n_months_back, loss_metric):

'mutate the default axes'

alphas = sorted(set(df.alpha))

l1_ratios = sorted(set(df.l1_ratio))

n_ticks = len(alphas) * len(l1_ratios)

line_index = -1

for units_X in ('natural', 'log'):

for units_y in ('natural', 'log'):

line_index += 1

x_label = []

y = np.empty((n_ticks,), dtype=float)

tick_index = -1

for alpha in alphas:

for l1_ratio in l1_ratios:

tick_index += 1

mask = (

(df.test_period == test_period) &

(df.n_months_back == n_months_back) &

(df.units_X == units_X) &

(df.units_y == units_y) &

(df.alpha == alpha) &

(df.l1_ratio == l1_ratio)

)

subset = df.loc[mask]

assert len(subset) == 1, subset

row = dict(subset.iloc[0])

y[tick_index] = row[loss_metric]

x_label.append('%3.1f-%4.2f' % (row['alpha'], row['l1_ratio']))

plt.plot(y / 1000.0,

label='%3s-%3s' % (units_X, units_y),

linestyle=[':', '-.', '--', '-'][line_index % 4],

color='bgrcmykw'[line_index % 8],

)

plt.xticks(range(len(y)), x_label, size='xx-small', rotation='vertical')

plt.yticks(size='xx-small')

plt.title('yr-mo %s-%s bk %d' % (test_period[:4], test_period[4:], n_months_back),

loc='left',

fontdict={'fontsize': 'xx-small', 'style': 'italic'},

)

def make_figure(year, months):

print 'make_figure', year, months

test_periods_typical = [str(year * 100 + month)

for month in months

]

test_periods = ('200902',) if year == 2009 else test_periods_typical

plt.figure() # new figure

# plt.suptitle('Loss by Test Period, Tree Max Depth, N Trees') # overlays the subplots

loss_metric = 'rmse'

loss_metric = 'mae'

axes_number = 0

n_months_backs = range(1, 7, 1)

last_test_period_index = len(test_periods) - 1

last_n_months_back_index = len(n_months_backs) - 1

for test_period_index, test_period in enumerate(test_periods):

for n_months_back_index, n_months_back in enumerate(n_months_backs):

axes_number += 1 # count across rows

plt.subplot(len(test_periods), len(n_months_backs), axes_number)

make_subplot(test_period, n_months_back, loss_metric)

if test_period_index == last_test_period_index:

# annotate the bottom row only

if n_months_back_index == 0:

plt.xlabel('alpha-l1_ratio')

plt.ylabel('%s x $1000' % loss_metric)

if n_months_back_index == last_n_months_back_index:

plt.legend(loc='best', fontsize=5)

plt.tight_layout(pad=0.4, w_pad=0.5, h_pad=1.0)

out_suffix = '-%02d' % months if len(months) == 1 else ''

plt.savefig(control.path_chart_base + str(year) + out_suffix + '.pdf')

plt.close()

for year in (2004, 2005, 2006, 2007, 2008, 2009):

months = (2,) if year == 2009 else (2, 5, 8, 11)

for month in months:

make_figure(year, (month,))

make_figure(year, months)

if control.test:

'write one txt file for each n_months_back'

def make_subplot(year, month, n_months_back):

'mutate the default axes'

test_period = str(year * 100 + month) # yyyymm

mask = (

(df.test_period == test_period) &

(df.n_months_back == n_months_back)

)

subset = df.loc[mask]

y = subset.mae

y.index = subset.learning_rate.values

y = y.sort_index()

plt.plot(y / 1000.0)

x_label = ['%4.2f' % x for x in y.index]

plt.xticks(range(len(y)), x_label, size='xx-small', rotation='vertical')

plt.yticks(size='xx-small')

plt.title('%4d-%02d %1d' % (year, month, n_months_back),

loc='right',

fontdict={'fontsize': 'xx-small', 'style': 'italic'},

)

return

learning_rates = sorted(set(df.learning_rates))

n_ticks = len(learning_rates)

# OLD BELOW ME

l1_ratios = sorted(set(df.l1_ratio))

n_ticks = len(l1_ratios)

line_index = -1

for units_X in ('natural', 'log'):

for units_y in ('natural', 'log'):

line_index += 1

x_label = []

y = np.empty((n_ticks,), dtype=float)

tick_index = -1

for l1_ratio in l1_ratios:

tick_index += 1

mask = (

(df.test_period == test_period) &

(df.n_months_back == n_months_back) &

(df.units_X == units_X) &

(df.units_y == units_y) &

(df.alpha == alpha) &

(df.l1_ratio == l1_ratio)

)

if sum(mask) != 1:

# something is wrong

print 'mask', sum(mask)

print 'test_period', sum(df.test_period == test_period)

print 'n_months_back', sum(df.n_months_back == n_months_back)

print 'units_X', sum(df.units_X == units_X)

print 'units_y', sum(df.units_y == units_y)

print 'alpha', sum(df.alpha == alpha)

print 'l1_ratio', sum(df.l1_ratio == l1_ratio)

print test_period, n_months_back, units_X, units_y, alpha, l1_ratio

pdb.set_trace()

subset = df.loc[mask]

assert len(subset) == 1, subset.shape

row = dict(subset.iloc[0])

y[tick_index] = row[loss_metric]

x_label.append('%4.2f' % row['l1_ratio'])

plt.plot(y / 1000.0,

label='%3s-%3s' % (units_X, units_y),

linestyle=[':', '-.', '--', '-'][line_index % 4],

color='bgrcmykw'[line_index % 8],

)

plt.xticks(range(len(y)), x_label, size='xx-small', rotation='vertical')

plt.yticks(size='xx-small')

plt.title('%4.2f %s-%s %d' % (alpha, test_period[:4], test_period[4:], n_months_back),

loc='left',

fontdict={'fontsize': 'xx-small', 'style': 'italic'},

)

def make_figure(year):

print 'make_figure', year

plt.figure() # new figure

# plt.suptitle('Loss by Test Period, Tree Max Depth, N Trees') # overlays the subplots

axes_number = 0

n_months_backs = (1, 2, 3, 4, 5, 6)

month_seq = (2, 5, 8, 11)

for month in month_seq:

for n_months_back in n_months_backs:

axes_number += 1 # count across rows

plt.subplot(len(month_seq), len(n_months_backs), axes_number)

make_subplot(year, month, n_months_back)

# annotate the bottom row only

if month == 11:

if n_months_back == 1:

plt.xlabel('learning_rate')

plt.ylabel('mae x $1000')

if n_months_back == 6:

plt.legend(loc='best', fontsize=5)

plt.tight_layout(pad=0.4, w_pad=0.5, h_pad=1.0)

plt.savefig(control.path_chart_base + str(year) + '.pdf')

plt.close()

for year in (2004, 2005, 2006, 2007, 2008, 2009):

make_figure(year)

if control.test:

break

'return data frame, ege_control'

def process_file(path, rows_list):

'mutate rows_list to include gscv object info at path'

print 'reducing', path

with open(path, 'rb') as f:

val_result, ege_control = pickle.load(f)

for k, v in val_result.iteritems():

actuals = v.actuals.values

predictions = v.predictions

if predictions is None:

print k

print 'predictions is missing'

pdb.set_trace()

errors = actuals - predictions

root_mean_squared_error = np.sqrt(np.sum(errors * errors) / (1.0 * len(errors)))

median_absolute_error = np.median(np.abs(errors))

row = {

'n_months_back': k.n_months_back,

'learning_rate': k.learning_rate,

'test_period': str(k.yyyymm),

'rmse': root_mean_squared_error,

'mae': median_absolute_error,

}

rows_list.append(row)

return ege_control # return last ege_control value (they should all be the same)

rows_list = []

for path in glob.glob(control.path_in_ege):

ege_control = process_file(path, rows_list)

if control.test:

break

df = pd.DataFrame(rows_list)

control = make_control(argv)

sys.stdout = Logger(base_name=control.arg.base_name)

print control

if control.arg.data:

df, ege_control = make_data(control)

with open(control.path_reduction, 'wb') as f:

pickle.dump((df, ege_control, control), f)

else:

with open(control.path_reduction, 'rb') as f:

df, ege_control, data_control = pickle.load(f)

make_charts(df, control, ege_control)

print control

if control.test:

print 'DISCARD OUTPUT: test'

print 'done'