ikleni

def feature_creation(temp, i):
    names = []
    temp[i+'_'+'mean' + '_' + '60'] = temp[i].shift(1 ).rolling(60, min_periods=10).mean()
    names.append(i+'_'+'mean' + '_' + '60')

    temp[i+'_'+'std' + '_' + '60'] = temp[i].shift(1 ).rolling(60, min_periods=10).std()
    names.append(i+'_'+'std' + '_' + '60')
    
    temp[i+'_'+'std' + '_' + '10'] = temp[i].shift(1 ).rolling(10, min_periods=6).std()

ikleni

def visual(X, iot = ['sensor1'], batch =60 , look_b = 50 ):
  # obtain indexes
  stops = X[(X['Response']== 1)].Time.values
  non_stops = X[(X['delta_nxt'] >100)&(X['Y']== 1)].Time.values # for good work
  idx = np.random.choice(stops, size=batch)
  for i in idx:
    temp = X[(X.Time == i)|((i - X.Time ) < look_b )&((i - X.Time) > 0 )]
    plt.figure(figsize=(30,10))
    for n,j in enumerate(iot):
      ax = plt.subplot(4,5, n+1) # here x*y should equal len(iot)

ikleni

def class_score(dt, feature, q): 
    # checks how well does a particular feature
    # split the data based on quantiles
    
    q0, q1 = np.quantile(dt[feature], q[0] ), np.quantile(dt[feature], q[1] )
    dt['pred'] = ((dt[feature]< q0) |(dt[feature]> q1)).astype(int)
    if dt[(dt['pred']==1)].shape[0] > 2:
        score = dt[(dt['Response']==1)&(dt['pred']==1)].shape[0]/ \
        dt[(dt['pred']==1)].shape[0]
    else:

ikleni

def agg_dev(X, feats, cat = True):
    X = X.set_index('Id')
    
    # for lists of discrete features
    if cat == True:

        # count how many features changed in value
        alpha = np.zeros(len(tmp_mini3)) # placeholder
        for i in feats:
            # get lagged values

ikleni

def y_prep(tmp, interval = [10,5], x = 'Time', type = 'A' ):       
    # forward fill current stops    
    tmp[['Cur_Stop', 'Cur_Stop_end']]  =tmp[['Stop_t','End']].fillna(method='ffill') 
    # back fill future stops    
    tmp[['Nxt_Stop', 'Nxt_Stop_start']] = tmp[['Stop_t','Start']].fillna(method='bfill')   
    # time since last and next stop    
    tmp['delta_cur'] = ((tmp['Cur_Stop_end'] - tmp[x])/np.timedelta64(1, 'm'))                                                                                          
    tmp['delta_nxt'] = ((tmp['Nxt_Stop_start'] -tmp[x])/np.timedelta64(1, 'm'))
    tmp['Y'] =  (tmp[x]<=interval[0]).astype(int)*(tmp[x]>=interval[1]).astype(int) 
    tmp = tmp[(tmp['Y'] == 0)&(tmp[x] >interval[1])|

ikleni

def del_const_feat(X, thres = 5, dtype = int):
    constant_cols = []
    if dtype == int:
        for i in X.dtypes[(X.dtypes == int)].index:
            if  X.shape[0]- X[i].value_counts().sort_values(ascending = False).values[0] <= thres:
                constant_cols.append(i)
    if dtype == float:
        for i in X.dtypes[(X.dtypes == float)].index:
            if  X[i].std() == 0:
                constant_cols.append(i)