def output_all_plots():
use_log_scale = False
ymax_dict = {'rosenbrock_2': 4}
data = load_data("soo.csv")
data = load_data("logo.csv", data)
data = load_data("bamsoo.csv", data)
for fn in all_fns(data):
regrets = data_filter(data, fn, REGRETS)
plt.clf()
plot_vals_err(fn, regrets)
out_name = fn + '_regrets.png'
plt.savefig(out_name, bbox_inches='tight')
for fn in all_fns(data):
ws = data_filter(data, fn, WS)
plt.clf()
plot_vals_err(fn, ws, ylabel="W")
out_name = fn + '_ws.png'
plt.savefig(out_name, bbox_inches='tight')
for fn in all_fns(data):
dists = data_filter(data, fn, DISTS)
plt.clf()
plot_vals_err(fn, dists, ylabel="Min. Distance")
out_name = fn + '_dists.png'
plt.savefig(out_name, bbox_inches='tight')
for fn in all_fns(data):
d = data_for_fn(fn, data)['LOGO']
slope_ws = reduce_sublists(pair_slope_val(1, d, WS))
plt.clf()
plot_scatter(fn, slope_ws, 'W', alpha=0.05)
out_name = fn + '_ws_scatter.png'
plt.savefig(out_name, bbox_inches='tight')
for fn in all_fns(data):
d = data_for_fn(fn, data)['BO1']
slope_dists = reduce_sublists(pair_slope_val(1, d, DISTS))
plt.clf()
plot_scatter(fn, slope_dists, 'Min. Distance')
out_name = fn + '_dists_scatter.png'
plt.savefig(out_name, bbox_inches='tight')
def kfold(k=10):
print "Loading data."
videos, users, reviews = load_data()
print "Extracting features."
orig_X = np.array([(x['date'], x['text'], x['user']) for x in reviews])
feats = create_features(orig_X, users)
#y = np.array([1 if x['spam'] == 'true' else 0 for x in reviews])
y = np.array([1 if x['adult'] == 'true' else 0 for x in reviews])
print "Vectorizing features."
v = DictVectorizer(sparse=False)
feats = v.fit_transform(feats)
print "Starting K-fold cross validation."
cv = cross_validation.KFold(len(feats), k=k, indices=True, shuffle=True, random_state=1234)
cls = LogisticRegression(penalty='l2', tol=0.00001, fit_intercept=False, dual=False, C=2.4105, class_weight=None)
if PRINT_COEFS:
cls.fit(feats, y)
c = v.inverse_transform(cls.coef_)
for key, val in sorted(c[0].iteritems(), key=lambda x: x[1]):
# if isinstance(key, str) and key.startswith("_"):
print key, val
quit()
f1sum = 0
for i, (train_idx, test_idx) in enumerate(cv):
train_X, train_y, test_X, test_y = feats[train_idx], \
y[train_idx], feats[test_idx], y[test_idx]
cls.fit(train_X, train_y)
preds = cls.predict(test_X)
if PRINT_ERRORS:
# worst = np.argsort(np.abs(test_y - preds))
#for j in worst[-1:-10:-1]:
orig_test = orig_X[test_idx]
# for j in worst:
for j in range(len(orig_test)):
if test_y[j] != preds[j]:
print j, orig_test[j][1], test_y[j], preds[j]
#quit()
f1 = metrics.f1_score(test_y, preds)
print "Fold %d F1 score: %.5f" % (i, f1)
f1sum += f1
avgf1 = (f1sum / k)
print "Mean F1 score: %.5f" % (f1sum / k)
# scores = cross_validation.cross_val_score(cls, feats, y, cv=10, score_func=metrics.f1_score)
# for i, score in enumerate(scores):
# print "Fold %d: %.5f" % (i, score)
# print "Mean score: %0.5f (+/- %0.2f)" % (scores.mean(), scores.std() / 2)
return avgf1
def output_singles():
max_out = 20
data = load_data("random.csv")
for fn in all_fns(data):
print fn
d = data_for_fn(fn, data)['RANDOM']
pairs = zip(d[REGRETS], d[DISTS])
for idx, (top, bot) in enumerate(pairs[:max_out]):
idx += 1
if idx % 10 == 0:
print idx
plt.clf()
plot_singles(fn, top, bot, bottom_log=True)
out_name = fn + '_single_' + str(idx) + '.png'
plt.savefig(out_name, bbox_inches='tight')
"""
import numpy as np
import tensorflow as tf
import time
from parse import FLAGS, load_data, build_embed
from srn import HLSTM
from srn_tool import train_srn, evaluate_srn, inference_srn
from pn import PolicyGradient
from pn_tool import pretrain_pn, train_pn, develop_pn, evaluate_pn, inference_pn
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
if FLAGS.log_parameters:
print(FLAGS.__flags)
label_train, text_train, sentence_len_train, keyword_train = load_data(FLAGS.data_dir, FLAGS.train_filename)
label_dev, text_dev, sentence_len_dev, keyword_dev = load_data(FLAGS.data_dir, FLAGS.valid_filename)
label_test, text_test, sentence_len_test, keyword_test = load_data(FLAGS.data_dir, FLAGS.test_filename)
embed = build_embed(FLAGS.data_dir, FLAGS.word_vector_filename)
SRN_graph = tf.Graph()
PN_graph = tf.Graph()
with SRN_graph.as_default():
SRN = HLSTM(FLAGS.symbols,
FLAGS.embed_units,
FLAGS.hidden_units,
FLAGS.labels,
embed,
FLAGS.learning_rate_srn)
if FLAGS.log_parameters:
def kfold(k=10):
print "Loading data."
videos, users, reviews = load_data()
print "Extracting features."
orig_X = np.array([(x['date'], x['text'], x['user']) for x in reviews])
feats = create_features(orig_X, users)
#y = np.array([1 if x['spam'] == 'true' else 0 for x in reviews])
y = np.array([1 if x['adult'] == 'true' else 0 for x in reviews])
print "Vectorizing features."
v = DictVectorizer(sparse=False)
feats = v.fit_transform(feats)
print "Starting K-fold cross validation."
cv = cross_validation.KFold(len(feats),
k=k,
indices=True,
shuffle=True,
random_state=1234)
cls = LogisticRegression(penalty='l2',
tol=0.00001,
fit_intercept=False,
dual=False,
C=2.4105,
class_weight=None)
if PRINT_COEFS:
cls.fit(feats, y)
c = v.inverse_transform(cls.coef_)
for key, val in sorted(c[0].iteritems(), key=lambda x: x[1]):
# if isinstance(key, str) and key.startswith("_"):
print key, val
quit()
f1sum = 0
for i, (train_idx, test_idx) in enumerate(cv):
train_X, train_y, test_X, test_y = feats[train_idx], \
y[train_idx], feats[test_idx], y[test_idx]
cls.fit(train_X, train_y)
preds = cls.predict(test_X)
if PRINT_ERRORS:
# worst = np.argsort(np.abs(test_y - preds))
#for j in worst[-1:-10:-1]:
orig_test = orig_X[test_idx]
# for j in worst:
for j in range(len(orig_test)):
if test_y[j] != preds[j]:
print j, orig_test[j][1], test_y[j], preds[j]
#quit()
f1 = metrics.f1_score(test_y, preds)
print "Fold %d F1 score: %.5f" % (i, f1)
f1sum += f1
avgf1 = (f1sum / k)
print "Mean F1 score: %.5f" % (f1sum / k)
# scores = cross_validation.cross_val_score(cls, feats, y, cv=10, score_func=metrics.f1_score)
# for i, score in enumerate(scores):
# print "Fold %d: %.5f" % (i, score)
# print "Mean score: %0.5f (+/- %0.2f)" % (scores.mean(), scores.std() / 2)
return avgf1
import numpy as np import cPickle from features import create_features, PROJECT from parse import load_data from dict_vectorizer import DictVectorizer videos, users, reviews = load_data() orig_X = np.array([(x['date'], x['text'], x['user']) for x in reviews]) feats = create_features(orig_X, None) v = DictVectorizer(sparse=False) feats = v.fit_transform(feats) # feats is now in vectorized format # v.transform() is the transformation that needs to be used on test data cPickle.dump(v, open(PROJECT + "db/dictvectorizer.pickle", "wb"))
from parse import load_data
if __name__ == '__main__':
load_data()
def main():
data = load_data("output.json")
d = data_for_fn('hartman_3', data)['BO1']
print d
