def get_data_for_float(self, dac_url, only_file=None,
surface_values_only=False):
'''Given a dac_url return a list of hashes of data for each
profile for the float specified in dac_url. Example dac_url
for float 1900722:
http://tds0.ifremer.fr/thredds/catalog/CORIOLIS-ARGO-GDAC-OBSaoml/1900722/profiles/catalog.xml
'''
pd = []
for profile_url in sorted(self.get_profile_opendap_urls(dac_url)):
if only_file:
if not profile_url.endswith(only_file):
continue
float = profile_url.split('/')[7]
prof = str(profile_url.split('/')[-1].split('.')[0].split('_')[1])
self.logger.info('Reading data from ' + profile_url[:20] + '...' +
profile_url[-50:])
try:
d = self.get_profile_data(profile_url,
surface_values_only=surface_values_only)
pd.append({float: {prof: d}})
except RequiredVariableNotPresent as e:
self.logger.warn(e)
except OpenDAPServerError as e:
self.logger.warn(e)
return pd
def get_values(lar):
st = []
pd = []
lar = str(lar)
for part in lar.split(' '):
if "kernel_size" in part[0:12]:
ker = part[-2]
if "MaxPool2d" in part:
ker = part[-2]
if 'AdaptiveAvgPool2d' in part:
ker = [i[-2] for i in part.split(' ') if 'output_size' in i][0]
s = 1
if 'stride' in part[0:7]:
s = part[-2]
if 'padding' in part:
pd.append(int(part[-2]))
else:
pd.append(0)
p = max(pd)
return int(ker), int(s), int(p)
def calc_embs(filepaths, batch_size=64):
pd = []
for start in tqdm(range(0, len(filepaths), batch_size)):
aligned_images = load_and_align_images(filepaths[start:start +
batch_size])
pd.append(nn4_small2.predict_on_batch(np.squeeze(aligned_images)))
embs = np.array(pd)
return np.array(embs)
def calc_embs(file_paths, batch_size=64):
pd = []
for start in tqdm(range(0, len(file_paths), batch_size)):
aligned_images = load_and_align_images(file_paths[start:start +
batch_size])
pd.append(model.predict_on_batch(np.squeeze(aligned_images)))
embs = l2_normalize(np.concatenate(pd))
# embs = np.array(pd)
return np.array(embs)
def ProbDis(vector): #Define Probability distribution
pd = []
m, n = vector.shape
for i in range(m):
v = 0
for j in range(n):
v = v + (vector[i, j]**2)
pd.append(v / n)
return pd
def calc_embs(filepaths, margin=10, batch_size=10):
pd = []
for start in tqdm(range(0, len(filepaths), batch_size)):
aligned_images = prewhiten(
load_and_align_images(filepaths[start:start + batch_size], margin))
pd.append(model.predict_on_batch(aligned_images))
embs = l2_normalize(np.concatenate(pd))
return embs
def calc_emb_test(faces):
pd = []
aligned_faces = align_faces(faces)
if (len(faces) == 1):
pd.append(nn4_small2.predict_on_batch(aligned_faces))
elif (len(faces) > 1):
pd.append(nn4_small2.predict_on_batch(np.squeeze(aligned_faces)))
embs = np.array(pd)
return np.array(embs)
def step_by_step_static_path_density(self, ende=None):
""" Returns list. [index=Aggregation depth: static path density]
"""
pd = []
for i, Cn in enumerate(self.step_by_step_aggregation(ende)):
print 'Static path density. Step ', i
pd.append(self.path_density_of_A(Cn))
return pd
def calc_emb_test(faces):
pd = []
aligned_faces = align_faces(faces)
if len(faces) == 1:
pd.append(model.predict_on_batch(aligned_faces))
elif len(faces) > 1:
pd.append(model.predict_on_batch(np.squeeze(aligned_faces)))
embs = l2_normalize(np.concatenate(pd))
# embs = np.array(pd)
return np.array(embs)
def find_closest_nodes(matrix):
nodes = list(matrix.columns)
pd = []
# find minimal distance in matrix
for n1 in nodes:
for n2 in nodes:
if n1==n2:
continue
pd.append((matrix[n1][n2], n1, n2))
pd.sort()
return pd[0]
def calc_emb_test(faces):
pd = []
aligned_faces = align_faces(faces)
if(len(faces)==1):
pd.append(nn4_small2.predict_on_batch(aligned_faces))
elif(len(faces)>1):
pd.append(nn4_small2.predict_on_batch(np.squeeze(aligned_faces)))
#embs = l2_normalize(np.concatenate(pd))
embs = np.array(pd)
#print(embs)
return np.array(embs)
def calc_embs(filepaths, margin=10, batch_size=1):
aligned_images = prewhiten(load_and_align_images(filepaths, margin))
pd = []
current = 0
for start in range(0, len(aligned_images), batch_size):
total = len(aligned_images)
if current % 5 == 0:
print('{} / {}'.format(current, total))
pd.append(
model.predict_on_batch(aligned_images[start:start + batch_size]))
current += 1
embs = l2_normalize(np.concatenate(pd))
return embs
def arima(df,time_id,lookback, p,d,q):
Log(LOG_INFO) << "Computing arima(%d,%d,%d) with lookback: %d " % (p,d,q,lookback)
pd=[]
for tid in time_id:
# pdb.set_trace()
series = np.log(df[OPEN_KEY][tid-lookback:tid].values)
model = ARIMA(series,order=(p,d,q))
model_fit = model.fit(method_kwargs={"warn_convergence": False})
output = model_fit.forecast()
p0 = np.log(df[OPEN_KEY][tid])
err = (output[0]-p0)/p0
pd.append(err)
pd = np.array(pd)
return pd.reshape(-1,1)
def s(self, q, pars=None):
if not hasattr(self, '_atomic_formfactors'):
self._atomic_formfactors = formFactor(q, self.Z)
if pars is None:
pars = self.par0
else:
# print(pars)
# print(self.par0.keys())
assert all([key in pars.keys() for key in self.par0.keys()]), \
'the input parameter dict does not contain all necessary parameter keys'
if self.reparameterized:
pars = self.convert(pars)
if not self.dispersed:
self.transform(pars)
return Debye(q, self, f=self._atomic_formfactors)
else:
pd = []
wd = []
for t in self._associated_transformation:
if t.dw:
_p, _w = t.dw.disperse(pars, t.name)
else:
_p, _w = pars[t.name], 1
pd.append(_p)
wd.append(_w)
pd_grid = [i.ravel() for i in np.meshgrid(*pd)]
wd_grid = [i.ravel() for i in np.meshgrid(*wd)]
n = len(pd_grid[0]) # number of combinations
# _bla = 0
_s = np.zeros(q.shape)
for i in range(n):
_p_dict = {}
_w = 1
for j, key in enumerate(self._t_keys):
_p_dict[key] = pd_grid[j][i]
_w *= wd_grid[j][i]
self.transform(_p_dict)
_s += _w * Debye(q, self, f=self._atomic_formfactors)
return _s
def test_compare_with_closed_form(self):
"""Test that compares the computed with the analytical CRPS."""
pd_single = norm(0, 1)
pd = []
for i in range(0, 3):
pd.append(pd_single)
meas = [-1, 0, 1]
mean_crps, single_crps = crps(pd, meas)
def crps_closed_form(pd, meas):
return meas * (2 * pd.cdf(meas) - 1) + 2 * pd.pdf(meas) - 1 / np.sqrt(np.pi)
crps_analytical = list(map(crps_closed_form, pd, meas))
is_good = np.isclose(np.array(single_crps), np.array(crps_analytical)).all()
assert_true(is_good, msg="Computed CRPS is not equal to analytical CRPS.")
def test_compare_different_expectations(self):
"""Test that compares same distance between meas and pd."""
pd_single = norm(0, 1)
pd = []
for i in range(0, 3):
pd.append(pd_single)
meas = [-1, 0, 1]
mean_crps1, single_crps1 = crps(pd, meas)
pd2 = []
for i in range(0, 3):
pd2.append(norm(i, 1))
meas2 = [-1, 1, 3]
mean_crps2, single_crps2 = crps(pd2, meas2)
is_good = np.equal(single_crps1, single_crps2).all()
assert_true(is_good, msg="Relation of individual CPRS values should return same value.")
def test_compare_different_expectations(self):
"""Test that compares same distance between meas and pd."""
pd_single = norm(0, 1)
pd = []
for i in range(0, 3):
pd.append(pd_single)
meas = [-1, 0, 1]
meanCRIGN1, singleCRIGN1 = crign.crign(pd, meas)
pd2 = []
for i in range(0, 3):
pd2.append(norm(i, 1))
meas2 = [-1, 1, 3]
meanCRIGN2, singleCRIGN2 = crign.crign(pd2, meas2)
is_good = np.isclose(singleCRIGN1, singleCRIGN2).all()
assert_true(
is_good,
msg=
"Relation of individual CRIGN values should return roughly the same value."
)
def getClusterCenters(data, k):
rows, cols = data.shape
centers = np.zeros((k, cols))
r = np.random.choice(rows, 1) #Initial Choice
centers[0, :] = data[r, :]
k_center = 1
while k_center < k:
pd = [] #pd represents probability distribution
dist_sum = 0
for i in range(rows):
max_dist = 1000
for j in range(k_center):
dist = np.linalg.norm(data[i] - centers[j])**2
if dist < max_dist:
max_dist = dist
d = max_dist * max_dist
dist_sum += d
pd.append(d)
for i in range(rows):
pd[i] = pd[i] / dist_sum
r = np.random.choice(rows, 1, p=pd)
centers[k_center, :] = data[r, :]
k_center = k_center + 1
return centers
def map_to_chart(candle, pd, stock):
temp = DailyData.chart(candle['date'], stock, candle['open'],
candle['high'], candle['low'], candle['close'],
candle['volume'], "NOT AVAILABLE",
"NOT AVAILABLE", "NOT AVAILABLE")
pd = pd.append(
{
'date': temp.get_date(),
'symbol': stock,
'open': temp.get_open(),
'high': temp.get_high(),
'low': temp.get_low(),
'close': temp.get_close(),
'volume': temp.get_volume(),
'change': temp.get_change(),
'changePercent': temp.get_changePercent(),
'vwap': temp.get_vwap()
},
ignore_index=True)
return pd
def get_pm25_df(html_path):
with open(html_path, encoding="utf-8") as html_file:
soup = BeautifulSoup(html_file, "html.parser")
pm25_df = pd.DataFrame(
columns=["hour", "pm25_value_max", "pm25_value_min"])
container_tag = soup.find_all("div", {"class": "whitebody"})[0]
parent_tag = container_tag.center.div.find("div", {
"class": "forecast-body"
}).find("div", {
"class": "forecast-body-table"
}).table
list_of_pm25 = parent_tag.find("tr", {
"class": "wf-row-pm25"
}).find_all("td")[1:]
hour = 0
for each_point in list_of_pm25:
try:
this_hour = hour
max_pm25 = int(
each_point.div.find("div", {
"class": "wf-cell-aqi-val-max"
}).text)
min_pm25 = int(
each_point.div.find("div", {
"class": "wf-cell-aqi-val-min"
}).text)
pm25_df = pm25_df.append(
{
"hour": this_hour,
"pm25_value_max": max_pm25,
"pm25_value_min": min_pm25
},
ignore_index=True)
hour += 3
hour = hour % 24
except:
print("this one dose not contain info")
return pm25_df
def condensematrix(dm, pd, names, key, hrf='canonical', op='mult'):
# returns condition with probe removed
import copy as cp
import numpy as np
delays = None
if hrf == 'fir':
delays = []
for i in dm.names:
if i == 'constant':
delays.append('-1')
else:
delays.append(i.split('_')[i.split('_').index('delay') + 1])
delays = np.array(delays, dtype=int)
if op == 'stack':
for i in dm.names:
if (i != 'constant'):
if (i.split('_')[i.split('_').index(key) + 1] != '0'):
pd.append(dm.matrix[:, dm.names.index(i)])
names.append(i.replace('glm_label_', ''))
else:
idx = []
for i in dm.names:
if i == 'constant':
idx.append('0')
else:
idx.append(i.split('_')[i.split('_').index(key)+1])
idx = np.array(idx, dtype=float)
if delays is not None:
for d in np.arange(np.max(delays)+1):
outkey = key + '_delay_' + str(d)
outidx = idx[delays == d]
pd.append(np.dot(dm.matrix[:, delays==d], outidx))
names.append(outkey)
else:
pd.append(np.dot(dm.matrix, idx))
names.append(key)
def condensematrix(dm, pd, names, key, hrf='canonical', op='mult'):
# returns condition with probe removed
import copy as cp
import numpy as np
delays = None
if hrf == 'fir':
delays = []
for i in dm.names:
if i == 'constant':
delays.append('-1')
else:
delays.append(i.split('_')[i.split('_').index('delay') + 1])
delays = np.array(delays, dtype=int)
if op == 'stack':
for i in dm.names:
if (i != 'constant'):
if (i.split('_')[i.split('_').index(key) + 1] != '0'):
pd.append(dm.matrix[:, dm.names.index(i)])
names.append(i.replace('glm_label_', ''))
else:
idx = []
for i in dm.names:
if i == 'constant':
idx.append('0')
else:
idx.append(i.split('_')[i.split('_').index(key) + 1])
idx = np.array(idx, dtype=float)
if delays is not None:
for d in np.arange(np.max(delays) + 1):
outkey = key + '_delay_' + str(d)
outidx = idx[delays == d]
pd.append(np.dot(dm.matrix[:, delays == d], outidx))
names.append(outkey)
else:
pd.append(np.dot(dm.matrix, idx))
names.append(key)
type(df.ix[['a','b'], 'Two']) type(df.ix[['a','b'],:]) df.ix[['a','b'],:] df.ix[['a','b'],] df.plot() matplotlib.show() matplotlib.Show() import matplotlib as mpl df.show() df1 = pd.DataFrame(np.random.randn(6,3), columns=['A','B','C']) df2 = pd.DataFrame(np.random.randn(6,3), columns=['D','E','F']) df3 = df1.copy() df1 df = pd.concat([df1, df2]) df df = pd.append([df1, df2]) help(pd.concat) df = pd.append([df1, df2], join='Inner') df = pd.concat([df1, df2], join='Inner') df = pd.concat([df1, df2], join=inner) df = pd.concat([df1, df2], join='inner') df df = pd.concat([df1, df2], join_axes='inner') df = df1.append(df2) df df = pd.concat([df1, df2], axes=1) df = df1.append(df2, axis=1) df = df1.append(df2, ignore_index=1) df df = df1.append(df2, ignore_index=0) df
baseUrl='https://movie.douban.com/top250?start='
headers = {
'User-Agent' : 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/88.0.4324.96 Safari/537.36 Edg/88.0.705.56'
}
movieDataList = []
for j in range(0,2):
star=j*25
url=baseUrl+str(star)
req = Request(url=url, headers=headers, method="GET")
response = urlopen(req).read().decode("utf-8")
soup = BeautifulSoup(response, 'html.parser')
movielist = soup.find_all("div", class_="item")
for i in range(0, 25):
moviename = re.search(r'<span class="title">(.*?)</span>', str(movielist[i]))
movevedio = movielist[i].select(".info > .hd > a")[0]
vedioname = re.search(r'<a class="" href="(.*?)">', str(movevedio))
movieimg = movielist[i].select(".pic>a>img")
img = re.search('src="(.*?)"', str(movieimg))
movedaoyan = movielist[i].select(".info>.bd>p")[0]
daoyan = re.search('<p class="">([\s\S]*?)</p>', str(movedaoyan))
jianjie = daoyan.group(1).strip().replace(" ", '').replace("\n", '').replace("<br/>", '').replace('/', '')
moviecomment = movielist[i].select(".star>.rating_num")[0].get_text()
jieshao = movielist[i].select('.quote>.inq')[0].get_text()
movedata = (moviename.group(1), vedioname.group(1), img.group(1), jianjie, moviecomment, jieshao)
# print(movedata)
movieDataList.append(movedata)
print(movieDataList)
print(len(movieDataList))
prob_df = pd.DataFrame()
prob_df['label'] = y_test
prob_df['prob_down'] = prob_down
global quantile
quantile = np.percentile(prob_down,list(list(range(0,101,10))))
prob_df['group'] = prob_df['prob_down'].apply(group_prob)
pd = []
td = []
for i in range(1,11):
group_df = prob_df[prob_df['group']==i]
true_down_prob = stats.itemfreq(group_df['label'])[0][1]/group_df.shape[0]
print('group:',i,'predict_prob:',group_df['prob_down'].median(),'true_prob:',true_down_prob)
pd.append(group_df['prob_down'].median())
td.append(true_down_prob)
print('mse',mean_squared_error(pd,td))
############ Lasso model ###############
from sklearn import linear_model
from sklearn.metrics import r2_score
df = pd.read_csv('/Users/vikasnatesh/Downloads/labeled_data_10s.csv')
# Train Test split (80% train, 20% test)
def make_parammat(dm, hrf='canonical', zscore=False):
# remove anything with a 0 and include probe as a feature
# assuming dm is a dict
import numpy as np
out = dm[dm.keys()[0]]
pd = []
names = []
for key in dm.keys():
if key == 'motion':
names.append('motion_0')
pd.append(np.dot(dm[key].matrix, np.array([1, 0, 0, 0, 0, 0, 0])))
names.append('motion_1')
pd.append(np.dot(dm[key].matrix, np.array([0, 1, 0, 0, 0, 0, 0])))
names.append('motion_2')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 1, 0, 0, 0, 0])))
names.append('motion_3')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 1, 0, 0, 0])))
names.append('motion_4')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 0, 1, 0, 0])))
names.append('motion_5')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 0, 0, 1, 0])))
# hardcode stim and verb
elif key == 'stim' or key == 'verb' or key == 'anim':
condensematrix(dm[key], pd, names, key, hrf, op='stack')
else:
condensematrix(dm[key], pd, names, key, hrf, op='mult')
# don't need constant because normalized data
# pd.append(np.ones(np.shape(pd[-1])))
# names.append('constant')
if zscore == True:
out.matrix = zs(np.array(pd).T)
else:
out.matrix = (np.array(pd).T)
out.names = names
return out
# winter
winter = joindf[(joindf.index.month == 12) | (joindf.index.month == 1) |
(joindf.index.month == 2)]
spring = joindf[(joindf.index.month > 2) & (joindf.index.month < 6)]
summer = joindf[(joindf.index.month > 5) & (joindf.index.month < 9)]
autumn = joindf[(joindf.index.month > 8) & (joindf.index.month < 12)]
correlation = {
'annual': joindf.corr().iloc[0, 1],
'winter': winter.corr().iloc[0, 1],
'spring': spring.corr().iloc[0, 1],
'summer': summer.corr().iloc[0, 1],
'autumn': autumn.corr().iloc[0, 1]
}
# funciona com xarray mas nao com o pandas...
nstorm_monthly_mean = nstorms.groupby(by=nstorms.index.month).mean()
for i in np.arange(1, 13, 1):
df = nstorms[nstorms.index.month == i] - nstorm_monthly_mean.year[i]
df.columns = ['norm']
result = pd.append([nstorms, df], axis=1, join='outer')
nstorms_norm = nstorms_norm.reset_index(drop=True)
indices_crop = indices_crop.reset_index(drop=True)
correlation = indices_crop.corr(nstorms_norm.values, method='pearson')
plt.figure()
def make_parammat(dm, hrf='canonical', zscore=False):
# remove anything with a 0 and include probe as a feature
# assuming dm is a dict
import numpy as np
out = dm[dm.keys()[0]]
pd = []
names = []
for key in dm.keys():
if key == 'motion':
names.append('motion_0')
pd.append(np.dot(dm[key].matrix, np.array([1, 0, 0, 0, 0, 0, 0])))
names.append('motion_1')
pd.append(np.dot(dm[key].matrix, np.array([0, 1, 0, 0, 0, 0, 0])))
names.append('motion_2')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 1, 0, 0, 0, 0])))
names.append('motion_3')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 1, 0, 0, 0])))
names.append('motion_4')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 0, 1, 0, 0])))
names.append('motion_5')
pd.append(np.dot(dm[key].matrix, np.array([0, 0, 0, 0, 0, 1, 0])))
# hardcode stim and verb
elif key == 'stim' or key == 'verb' or key == 'anim':
condensematrix(dm[key], pd, names, key, hrf, op='stack')
else:
condensematrix(dm[key], pd, names, key, hrf, op='mult')
# don't need constant because normalized data
# pd.append(np.ones(np.shape(pd[-1])))
# names.append('constant')
if zscore==True:
out.matrix = zs(np.array(pd).T)
else:
out.matrix = (np.array(pd).T)
out.names = names
return out
from scraper import *
import pandas as pd
link = ['https://www.ah.nl/producten/product/wi48405/ah-basic-havermout', 'https://www.ah.nl/producten/product/wi383520/ah-amandel-drink-ongezoet']
name = ['oats', 'almond milk']
portion = [60, 200]
for idx in range(len(link)):
get_nutrients(link[idx], name[idx])
foods = pd.append(get_nutrients(link, name), foods)
foods.to_csv('foods.csv')
#meal 1 oats
# meal 2 chicken and veg
# introduce class mealplan, subclasses ingredients/food, contains targets, ingredients, stock
import gc
def fix_time(df, time_cols):
for time_col in time_cols:
df[time_col] = pd.to_datetime(df[time_col],
errors='coerce',
format='%Y%m%d')
return df
gc.enable()
df_train = pd.read_csv('~/train.csv')
df_train = pd.append((df_train, pd.read_csv('~/train_v2.csv')),
axis=0,
ignore_index=True).reset_index(drop=True)
df_test = pd.read_csv('~/sample_submission_v2.csv')
df_members = pd.read_csv('~/members_v3.csv')
df_transactions = pd.read_csv('~/transactions.csv')
df_transactions = pd.append(
(df_transactions, pd.read_csv('~/transactions_v2.csv')),
axis=0,
ignore_index=True).reset_index(drop=True)
df_transactions = df_transactions.sort_values(
by=['transaction_date'], ascending=[False]).reset_index(drop=True)
df_transactions = df_transactions.drop_duplicates(subset=['msno'],
keep='first')
import csv
import pandas as pd
file1 = ("C:/Users/Krithi/Desktop/Python/bright_star.csv")
file2 = ("C:/Users/Krithi/Desktop/Python/convertedStars.csv")
d1 = []
d2 = []
with open(file1, "r", encoding="utf8") as f:
csvreader = csv.reader(f)
for i in csvreader:
d1.append(i)
with open(file2, "r", encoding="utf8") as f:
csvreader = csv.reader(f)
for i in csvreader:
d2.append(i)
h1 = d1[0]
h2 = d2[0]
pd1 = d1[1:]
pd2 = d2[1:]
h = h1 + h2
pd = []
for i in pd1:
pd.append(i)
for j in pd2:
pd.append(j)
with open("totalStars.csv", "w", encoding='utf8') as f:
csvwriter = csv.writer(f)
csvwriter.writerow(h)
csvwriter.writerows(pd)
#df = pd.read_csv("totalStars.csv")
#df.tail(8)
for i in range(1, number_of_steps):
if label == 'mlp' or label == 'cnn' or label == 'lstm':
model.set_weights(init_weights)
if i * data_batch > X_train.shape[0] - 1:
# if label == 'mlp' or label == 'cnn' or label == 'lstm':
history = model.fit(X_train[:X_train.shape[0] - 1, :], y_train[:y_train.shape[0] - 1], epochs=100,
validation_data=(X_test, y_test),
callbacks=[early_stopping], workers=8)
# else:
# history = model.fit(X_train[:X_train.shape[0] - 1, :], y_train[:y_train.shape[0] - 1]) # validation_data=(X_test, y_test), workers=8)
predictions['size_' + str(i * data_batch)] = model.predict(X_test, y_test)
temp = pd.DataFrame(history.history)
stats = pd.append([stats, temp.iloc[temp.shape[0] - 1, :]])
break
# if label == 'mlp' or label == 'cnn' or label == 'lstm':
history = model.fit(X_train[:i * data_batch], y_train[:i * data_batch], epochs=100,
validation_data=(X_test, y_test),
callbacks=[early_stopping], workers=8)
# else:
# history = model.fit(X_train[:i * data_batch], y_train[:i * data_batch]) # ,validation_data=(X_test, y_test), workers=8)
predictions['size_' + str(i)] = np.argmax(model.predict(X_test))#, axis=1)
temp = pd.DataFrame(history.history)
stats = stats.append(temp.iloc[temp.shape[0] - 1, :])
# Log of models data