def encode_by_tokens(self, graph_set):
sentence_tokens = graph_set[0].get("tokens", [])
sentence_encoded = [
utils.get_idx(t, self._word2idx) for t in sentence_tokens
]
edges_encoded = []
for g in graph_set:
first_edge = graph.get_graph_first_edge(g)
property_label = first_edge.get('label', '')
edge_ids = [
utils.get_idx(t, self._word2idx)
for t in property_label.split()
]
edges_encoded.append(edge_ids)
return sentence_encoded, edges_encoded
def get_interp_data(z_sample, maps, logfile=None, verbose=True):
"""
Finds the two projections with redshifts that are the nearest higher and
nearest lower redshifts and extracts the
Parameters
----------
z_sample : float
The redshift of interest in the interpolation.
maps : array or array-like
The filenames of the column density maps. These have the same indexing
as redshift_arr
logfile :
The file to write the logs.
Returns
-------
map_high : str
map_high : str
"""
z_exist = np.empty(len(maps))
for i in range(len(maps)):
with h5py.File(maps[i], "r") as ds:
z_exist[i] = ds["Header"].attrs["Redshift"]
# Get index
idx_low, idx_high = utils.get_idx(z_sample, z_exist)
# Get redshift of maps lower/higher than z_sample
z_low, z_high = z_exist[idx_low], z_exist[idx_high]
dist_low, dist_high = utils.z_to_mpc(z_low), utils.z_to_mpc(z_high)
map_low, map_high = maps[idx_low], maps[idx_high]
#data_low, data_high = h5py.File(map_low), h5py.File(map_high)
if logfile:
wlog(
"{0:<10} {1:>10}\
{2:<10} {3:>10}".format("idx_low", idx_low, "idx_high", idx_high),
logfile, verbose)
wlog(
"{0:<10} {1:>10.5}\
{2:<10} {3:>10.5}".format("z_low", z_low, "z_high", z_high),
logfile, verbose)
wlog(
"{0:<10} {1:>9.5}\
{2} {3:>9.5}".format("dist_low", dist_low, "dist_high",
dist_high), logfile, verbose)
wlog(
"{0:<10} {1}\
{2:<10} {3}\n".format("map_low", map_low, "map_high", map_high),
logfile, verbose)
return map_low, map_high
def encode_by_tokens(self, graph_set):
sentence_tokens = graph_set[0].get("tokens", [])
sentence_encoded = [
utils.get_idx(t, self._word2idx) for t in sentence_tokens
]
graphs_encoded = []
for g in graph_set:
edges_encoded = []
for edge in g.get('edgeSet', []):
property_label = edge.get('label', '')
edge_ids = [
utils.get_idx(t, self._word2idx)
for t in property_label.split()
]
edges_encoded.append(edge_ids)
graphs_encoded.append(edges_encoded)
return sentence_encoded, graphs_encoded
def encode_graphs(self, graph_set):
graphs_encoded = []
for g in graph_set:
edges_encoded = []
for edge in g.get('edgeSet', []):
property_label = edge.get('label', '')
edge_ids = [utils.get_idx(t, self._word2idx) for t in property_label.split()]
edges_encoded.append(edge_ids)
graphs_encoded.append(edges_encoded)
return graphs_encoded
def feature_extraction(features, filtered, valid_labels, fs, window):
# all feature calculations must end up the same size
X = np.empty((len(valid_labels[window:]), 0))
for i in features:
if i == 'linelength':
X = np.concatenate((X, linelength(filtered, window)), axis=1)
if i == 'delta':
X = np.concatenate(
(X,
bandpower(filtered, window,
utils.get_idx(delta_band, window, fs), fs)),
axis=1)
if i == 'theta':
X = np.concatenate(
(X,
bandpower(filtered, window,
utils.get_idx(theta_band, window, fs), fs)),
axis=1)
if i == 'alpha':
X = np.concatenate(
(X,
bandpower(filtered, window,
utils.get_idx(alpha_band, window, fs), fs)),
axis=1)
if i == 'beta':
X = np.concatenate(
(X,
bandpower(filtered, window,
utils.get_idx(beta_band, window, fs), fs)),
axis=1)
if i == 'gamma':
X = np.concatenate(
(X,
bandpower(filtered, window,
utils.get_idx(gamma_band, window, fs), fs)),
axis=1)
# size of labels should be consistent with X.shape[0]
y = valid_labels[window:]
return X, y
def train(self):
os.makedirs(self.summ_path+"model/")
optimizer = keras.optimizers.Adam(0.00001)
global_step = 0
print("Training Started. Results and summary files are stored at", self.summ_path)
for cur_epoch in trange(config.epoch):
trn_sup_idx, trn_qry_idx, trn_lbl = utils.get_idx(lbl=self.trn_lbl)
for cur_step in trange(0, config.iter_cnt, config.batch_size):
cur_sup_dat = self.trn_dat[trn_sup_idx[cur_step:cur_step+config.batch_size]]
cur_qry_dat = self.trn_dat[trn_qry_idx[cur_step:cur_step+config.batch_size]]
cur_lbl = trn_lbl[cur_step:cur_step+config.batch_size]
self.train_one_step(x_sup=cur_sup_dat, x_qry=cur_qry_dat, lbl=cur_lbl,
model=self.all_model, optim=optimizer, vars=self.all_model.trainable_variables,
step=global_step, log=global_step%10==0)
global_step+=1
val_sup_idx, val_qry_idx, val_lbl = utils.get_idx(lbl=self.val_lbl, iter_cnt=100)
self.logger(x_sup=self.val_dat[val_sup_idx], x_qry=self.val_dat[val_qry_idx], lbl=val_lbl,
model=self.all_model, step=global_step)
self.all_model.save_weights(self.summ_path+"model/%04d.h5"%cur_epoch)
def get_edge_feature_vector(self, edge):
edge_kbid = edge.get('kbID')[:-1] if 'kbID' in edge else utils.unknown_el
right_label_ids = [utils.get_idx(t, self._word2idx) for t in edge.get('canonical_right', "").split()][
:self._p.get('symbolic.features', {}).get("right.label", 0)]
feature_vector = [self._property2idx.get(edge_kbid, 0),
self._property2idx.get(edge['hopUp'][:-1] if 'hopUp' in edge else utils.all_zeroes, 0),
self._property2idx.get(edge['hopDown'][:-1] if 'hopDown' in edge else utils.all_zeroes, 0),
self._modifier2idx.get("argmax" if "argmax" in edge
else "argmin" if "argmin" in edge
else "num" if "num" in edge
else "filter" if "filter" in edge
else utils.all_zeroes, 0), self._type2idx.get(edge.get('type', utils.unknown_el), 0),
self._propertytype2idx.get(edge['kbID'][-1] if 'kbID' in edge else utils.unknown_el, 0),
] + right_label_ids
assert len(feature_vector) <= self._feature_vector_size
return feature_vector
def longest_increasing_seq(A): """ Longest Increasing Subsequence. Given a sequence of n real numbers A(1) ... A(n), determine a subsequence (not necessarily contiguous) of maximum length in which the values in the subsequence form a strictly increasing sequence. """ D, T = [1] * len(A), [-1] * len(A) maxLen, maxIdx = 1, -1 for i in range(1, len(A)): for j in range(i): if A[i] > A[j] and D[i] < D[j] + 1: D[i] = D[j] + 1 T[i] = j if D[i] > maxLen: maxLen = D[i] maxIdx = i l = utils.get_idx(maxIdx, T, []) return maxLen, [A[i] for i in l]
def test_get_idx(self):
self.assertEquals(utils.get_idx(['a','b','c'], 1), 'b')
self.assertEquals(utils.get_idx(pd.Series(['a','b','c']), 1), 'b')
def print_indices(f, freq_band, window, fs):
idxStartBin, idxEndBin = utils.get_idx(freq_band, window, fs)
f.write("val idxStartBin = %d\n" % idxStartBin)
f.write("val idxEndBin = %d\n" % idxEndBin)
f.write("\n\n")
def plot_distributions(samples,
output_dir,
bin_sizes,
plot_var,
sig_tag,
weight_type='None',
normalize=False,
density=True,
log=True,
file_name=''):
if 'top' in sig_tag: tag = r'$t\bar{t}$'
elif 'BSM' in sig_tag: tag = 'BSM'
elif 'OoD' in sig_tag: tag = 'OoD'
if 'OoD' in sig_tag:
labels = {0: [tag, 'QCD'], 1: [tag + ' (weighted)', 'QCD (weighted)']}
else:
labels = {0: [tag, 'QCD'], 1: [tag + ' (cut)', 'QCD (cut)']}
colors = ['tab:orange', 'tab:blue', 'tab:brown']
alphas = [1, 0.5]
xlabel = {
'pt': '$p_t$',
'm': '$m$',
'rljet_n_constituents': 'Number of constituents'
}[plot_var]
plt.figure(figsize=(13, 8))
pylab.grid(True)
axes = plt.gca()
if not isinstance(samples, list): samples = [samples]
for m in [0, 1]:
for n in range(len(samples)):
sample = samples[n]
condition = sample['JZW'] == -1 if m == 0 else sample[
'JZW'] >= 0 if m == 1 else sample['JZW'] >= -2
if not np.any(condition): continue
variable = np.float32(sample[plot_var][condition])
weights = sample['weights'][condition]
if 'flat' in weight_type:
min_val, max_val = max(0, np.min(variable)), np.max(variable)
else:
min_val, max_val = max(0, np.min(sample[plot_var])), np.max(
sample[plot_var])
bins = get_idx(max_val,
bin_size=bin_sizes[plot_var],
min_val=min_val,
integer=False,
tuples=False)
if normalize:
weights *= 100 / (np.sum(sample['weights'])
) #100/(np.sum(weights))
if density:
indices = np.searchsorted(bins, variable, side='right')
weights /= np.take(np.diff(bins),
np.minimum(indices,
len(bins) - 1) - 1)
pylab.hist(variable,
bins,
histtype='step',
weights=weights,
color=colors[m],
lw=2,
log=log,
alpha=alphas[n],
label=labels[n][m])
if 'OoD' in sig_tag:
if plot_var == 'm':
pylab.xlim(0, 1200)
pylab.ylim(1e0, 1e5)
elif plot_var == 'pt':
pylab.xlim(0, 3000)
pylab.ylim(1e0, 1e5)
elif 'Geneva' in sig_tag:
if plot_var == 'm':
pylab.xlim(0, 500)
pylab.ylim(1e-2, 1e5)
elif plot_var == 'pt':
pylab.xlim(0, 2000)
pylab.ylim(1e-2, 1e5)
else:
if plot_var == 'm':
pylab.xlim(0, 500)
pylab.ylim(1e0, 1e7)
elif plot_var == 'pt':
pylab.xlim(0, 2000)
pylab.ylim(1e0, 1e7)
axes.xaxis.set_minor_locator(ticker.AutoMinorLocator(10))
if not log: axes.yaxis.set_minor_locator(ticker.AutoMinorLocator(10))
plt.xlabel(xlabel + ' (GeV)', fontsize=24)
y_label = ' density' if density else ''
if normalize: y_label += ' (%)'
elif sig_tag in ['top-UFO', 'BSM']:
y_label += ' (' + r'58.5 fb$^{-1}$' + ')'
plt.ylabel('Distribution' + y_label, fontsize=24)
axes.tick_params(axis='both', which='major', labelsize=14)
plt.legend(loc='upper right',
ncol=1 if len(samples) == 1 else 2,
fontsize=18)
if file_name == '':
file_name = (plot_var if plot_var == 'pt' else 'mass') + '_dist.png'
file_name = output_dir + '/' + file_name
print('Saving', format(plot_var, '2s'), 'distributions to:', file_name)
plt.savefig(file_name)
supports,
fmt=fmt,
delimiter=',')
np.savetxt("generated_files/alpha_vectors.csv",
alpha_vector,
fmt=fmt,
delimiter=',')
np.savetxt("generated_files/intercepts.csv",
intercept,
fmt=fmt,
delimiter=',')
for i in features:
if i == 'delta':
np.savetxt("generated_files/delta_index.csv",
utils.get_idx(fe.delta_band, window, fs),
fmt='%d',
delimiter=',')
if i == 'theta':
np.savetxt("generated_files/theta_index.csv",
utils.get_idx(fe.theta_band, window, fs),
fmt='%d',
delimiter=',')
if i == 'alpha':
np.savetxt("generated_files/alpha_index.csv",
utils.get_idx(fe.alpha_band, window, fs),
fmt='%d',
delimiter=',')
if i == 'beta':
np.savetxt("generated_files/beta_index.csv",
utils.get_idx(fe.beta_band, window, fs),
def get_data_for_interpolation(z_sample,
redshift_arr,
projections,
logfile=None):
"""
Finds the two projections with redshifts that are the nearest higher and
nearest lower redshifts and extracts the
Parameters
----------
z_sample : float
The redshift of interest in the interpolation.
redshift_arr: array or array-like
The redshifts from the projections ordered by snapshot number.
projections : array or array-like
The filenames of the projections. These have the same indexing as
redshift_arr
logfile :
The file to write the logs.
Returns
-------
data_low :
The data of the projection with the nearest lower redshift to z_sample.
dist_low :
The comoving distance to the projection with the nearest lower
redshift to z_sample.
data_high :
The data of the projection with the nearest higher redshift to z_sample.
dist_high :
The comoving distance to the projection with the nearest higher
redshift to z_sample.
"""
if logfile:
logfile.write("\n-----------------")
logfile.write(
"\nGetting Interpolation Data: z = {0:.5f}".format(z_sample))
logfile.write("\n-----------------\n")
idx_low, idx_high = utils.get_idx(z_sample, redshift_arr)
z_low, z_high = redshift_arr[idx_low], redshift_arr[idx_high]
dist_low, dist_high = utils.z_to_mpc(z_low), utils.z_to_mpc(z_high)
proj_low, proj_high = projections[idx_low], projections[idx_high]
data_low, data_high = h5py.File(proj_low), h5py.File(proj_high)
if logfile:
logfile.write("{0:<10} {1:10}\n{2:<10} {3:10}\n".format(
"idx_low", idx_low, "idx_high", idx_high))
logfile.write("{0:<10} {1:10.5}\n{2:<10} {3:10.5}\n".format(
"z_low", z_low, "z_high", z_high))
logfile.write("{0:<10} {1:10.5}\n{2:<10} {3:10.5}\n".format(
"dist_low", dist_low, "dist_high", dist_high))
logfile.write("{0:<10} {1}\n{2:<10} {3}".format(
"proj_low", proj_low, "proj_high", proj_high))
return data_low, dist_low, data_high, dist_high
def test_get_idx(self):
self.assertEquals(utils.get_idx(['a', 'b', 'c'], 1), 'b')
self.assertEquals(utils.get_idx(pd.Series(['a', 'b', 'c']), 1), 'b')
def encode_question(self, graph_set):
sentence_tokens = graph_set[0].get("tokens", [])
sentence_encoded = [utils.get_idx(t, self._word2idx) for t in sentence_tokens]
return sentence_encoded