def feature_extract(dt):
import tsfresh.feature_extraction.feature_calculators as fc
ft = {
'abs_energy': fc.abs_energy(dt),
'sum_values': fc.sum_values(dt),
'mean': fc.mean(dt),
'maximum': fc.maximum(dt),
'minimum': fc.minimum(dt),
'median': fc.median(dt),
'quantile_0.1': fc.quantile(dt, 0.1),
'quantile_0.2': fc.quantile(dt, 0.2),
'quantile_0.3': fc.quantile(dt, 0.3),
'quantile_0.4': fc.quantile(dt, 0.4),
'quantile_0.5': fc.quantile(dt, 0.5),
'quantile_0.6': fc.quantile(dt, 0.6),
'quantile_0.7': fc.quantile(dt, 0.7),
'quantile_0.8': fc.quantile(dt, 0.8),
'quantile_0.9': fc.quantile(dt, 0.9),
#
# TODO:
# Below functions dont works well -> need to be checked!!
#
#'fft_coefficient__coeff_0__attr_real': fc.fft_coefficient(dt {"coeff": 0, "attr": "real"}),
#'fft_coefficient__coeff_0__attr_imag': fc.fft_coefficient(dt {"coeff": 0, "attr": "imag"}),
#'fft_coefficient__coeff_0__attr_abs': fc.fft_coefficient(dt {"coeff": 0, "attr": "abs"}),
#'fft_coefficient__coeff_0__attr_angle': fc.fft_coefficient(dt {"coeff": 0, "attr": "angle"}),
#
#=> Mr. Huy just fix this issue with above function fft_ft !!
}
ft.update(fft_ft(dt))
return ft
def get_feature_date(daydf):
daydf = daydf.drop("charttime", axis=1)
daydf = daydf.drop("subject_id", axis=1)
daydf = daydf.drop("icustay_id", axis=1)
retval = np.zeros([1, 0], dtype=float)
retval = pd.DataFrame(retval)
for column in daydf:
#print(retval.shape)
if ((column == "heart_rate") | (column == "abp_systolic")
| (column == "gcs_total") | (column == "platelets") |
(column == "creatinine")):
t = scalar_feature_extraction(daydf[column])
t = pd.DataFrame(t)
retval = pd.concat([retval, t], axis=1)
elif ((column == "weight") | (column == "age")):
t = np.zeros([1, 1], dtype=float)
t[0][0] = daydf.iloc[0][0]
t = pd.DataFrame(t)
retval = pd.concat([retval, t], axis=1)
elif (column == "is_dead_in_n_hours"):
t = np.zeros([1, 1], dtype=float)
t[0][0] = tffe.maximum(daydf[column].values)
t = pd.DataFrame(t)
retval = pd.concat([retval, t], axis=1)
else:
t = logistic_feature_extraction(daydf[column])
t = pd.DataFrame(t)
retval = pd.concat([retval, t], axis=1)
return retval
def get_sta_features(self, data):
"""
Calculate the value of 9 kinds of selected statistical features
:param data:
:return:
"""
def _cal_trend(data):
time_list = np.arange(len(data))
# create linear regression object
regr = linear_model.LinearRegression()
regr.fit(time_list.reshape(-1, 1), np.array(data).reshape(-1, 1))
return regr.coef_[0][0]
E = ts.abs_energy(data)
S = ts.binned_entropy(data, max_bins=5)
ro = ts.autocorrelation(data, lag=4)
skewness = ts.skewness(data)
kurtosis = ts.kurtosis(data)
trend = _cal_trend(data)
mean = ts.mean(data)
min = ts.minimum(data)
max = ts.maximum(data)
return [E, S, ro, skewness, kurtosis, trend, mean, min, max]
def TS_feature3(signal):
max_ts = ts.maximum(signal)
mean_rs = ts.mean(signal)
mean_abs_change = ts.mean_abs_change(signal)
mean_change = ts.mean_change(signal)
median_ts = ts.median(signal)
minimum_ts = ts.minimum(signal)
return max_ts, mean_rs, mean_abs_change, mean_change, median_ts, minimum_ts
def time_series_maximum(x):
"""
:param x: the time series to calculate the feature of
:type x: pandas.Series
:return: the value of this feature
:return type: float
"""
return ts_feature_calculators.maximum(x)
def time_series_maximum(x):
"""
序列x的最大值
:param x: x
:type x: pandas.Series
:return: the value of this feature
:return type: float
"""
return ts_feature_calculators.maximum(x)
def scalar_feature_extraction(column):
retval = np.zeros([1, 10], dtype=float)
retval[0][0] = tffe.count_above_mean(column.values)
retval[0][1] = tffe.mean(column.values)
retval[0][2] = tffe.maximum(column.values)
retval[0][3] = tffe.median(column.values)
retval[0][4] = tffe.minimum(column.values)
retval[0][5] = tffe.sample_entropy(column.values)
if (isNaN(retval[0][5])):
retval[0][5] = 0
retval[0][6] = tffe.skewness(column.values)
retval[0][7] = tffe.variance(column.values)
retval[0][8] = tffe.longest_strike_above_mean(column.values)
retval[0][9] = tffe.longest_strike_below_mean(column.values)
return retval
def translate_to_hourly(df):
col_names = df.columns.values
new_df = pd.DataFrame(columns=col_names)
interim_df = pd.DataFrame(columns=col_names)
curr_icu_stay_id = df.iloc[0].loc['icustay_id']
curr_hour = datetime.strptime(str(df.iloc[0].loc['charttime']),
"%Y-%m-%d %H:%M:%S")
for index in range(len(df.index)):
print("index: ", index)
row_icu_stay_id = df.iloc[index].loc['icustay_id']
row_hour = datetime.strptime(str(df.iloc[index].loc['charttime']),
"%Y-%m-%d %H:%M:%S")
if ((row_icu_stay_id == curr_icu_stay_id) &
(is_same_hour(curr_hour, row_hour))):
print(df.iloc[index].loc['charttime'])
interim_df = interim_df.append(df.iloc[index], ignore_index=True)
else:
new_row = interim_df.mean(axis=0)
new_row.loc['charttime'] = curr_hour.strftime("%Y-%m-%d %H:%M:%S")
new_row.loc['is_dead_in_n_hours'] = tffe.maximum(
interim_df['is_dead_in_n_hours'].values)
curr_hour = row_hour
curr_icu_stay_id = row_icu_stay_id
new_df = new_df.append(new_row, ignore_index=True)
interim_df = pd.DataFrame(columns=col_names)
interim_df = interim_df.append(df.iloc[index], ignore_index=True)
print("completed hour")
''' # commented stuff used during testing
print("-----------------------------------------------------------------")
print("average: \n", new_row)
print("end of average")
for index, row in interim_df.iterrows():
print(row)
#x = input("Pause:")'''
#print(new_df.head())
new_row = interim_df.mean(axis=0)
new_row.loc['charttime'] = curr_hour.strftime("%Y-%m-%d %H:%M:%S")
new_df = new_df.append(new_row, ignore_index=True)
return new_df
def extract_features(data):
day = 24 * 60
return list(
numpy.nan_to_num(
numpy.array([
feature.symmetry_looking(data, [{
'r': 0.3
}])[0][1],
feature.variance_larger_than_standard_deviation(data).bool(),
feature.ratio_beyond_r_sigma(data, 2),
feature.has_duplicate_max(data),
feature.has_duplicate_min(data),
feature.has_duplicate(data),
feature.agg_autocorrelation(numpy.array(data.value),
[{
'f_agg': 'mean',
'maxlag': day
}])[0][1],
feature.partial_autocorrelation(data, [{
'lag': day
}])[0][1],
feature.abs_energy(numpy.array(data.value)),
feature.mean_change(data),
feature.mean_second_derivative_central(data),
feature.median(data),
float(feature.mean(data)),
float(feature.standard_deviation(data)),
float(feature.longest_strike_below_mean(data)),
float(feature.longest_strike_above_mean(data)),
int(feature.number_peaks(data, 10)),
feature.linear_trend(numpy.array(data.value), [{
'attr': 'rvalue'
}])[0][1],
feature.c3(data, day),
float(feature.maximum(data)),
float(feature.minimum(data))
])))
def get_maximum(arr):
res = np.array([maximum(arr)])
res = np.nan_to_num(res)
return res