71  outliers graphs

import matplotlib.pyplot as plt
import warnings
import pandas as pd
import numpy as np 
import seaborn as sns
sns.set(style="ticks", font_scale=1.5)  # white graphs, with large and legible letters
warnings.simplefilter(action='ignore', category=FutureWarning)
import matplotlib.gridspec as gridspec
from matplotlib.dates import DateFormatter
import matplotlib.dates as mdates
from scipy.stats import median_abs_deviation

https://www.google.com/imgres?imgurl=https%3A%2F%2Fcxl.com%2Fwp-content%2Fuploads%2F2017%2F01%2Fchart-1.png&tbnid=RClfEFYNWm0WWM&vet=12ahUKEwjsx9KiqtKDAxXwh_0HHQR0CPoQMygSegQIARB1..i&imgrefurl=https%3A%2F%2Fcxl.com%2Fblog%2Foutliers%2F&docid=EVGldozQCNsxLM&w=1662&h=722&q=outliers%20different%20distributions&client=safari&ved=2ahUKEwjsx9KiqtKDAxXwh_0HHQR0CPoQMygSegQIARB1

source: https://github.com/erykml/medium_articles/blob/master/Machine%20Learning/outlier_detection_hampel_filter.ipynb

71.1 define functions

def random_walk_with_outliers(origin, n_steps, perc_outliers=0.0, outlier_mult=10, seed=42):
    '''
    Function for generating a random time series based on random walk.
    It adds a specified percentage of outliers by multiplying the random walk step by a scalar.
    
    Parameters
    ------------
    origin : int
        The starting point of the series
    n_steps : int
        Lenght of the series
    perc_outliers : float
        Percentage of outliers to introduce to the series [0.0-1.0]
    outlier_mult : float
        Scalar by which to multiply the RW increment to create an outlier
    seed : int
        Random seed

    Returns
    -----------
    rw : np.ndarray
        The generated random walk series with outliers
    indices : np.ndarray
        The indices of the introduced outliers 
    '''
    assert (perc_outliers >= 0.0) & (perc_outliers <= 1.0)
    
    #set seed for reproducibility
    np.random.seed(seed)
    
    # possible steps
    steps = [-1, 1]

    # simulate steps
    steps = np.random.choice(a=steps, size=n_steps-1)
    rw = np.append(origin, steps).cumsum(0)
    
    # add outliers
    n_outliers = int(np.round(perc_outliers * n_steps, 0))
    indices = np.random.randint(0, len(rw), n_outliers)
    rw[indices] = rw[indices] + steps[indices + 1] * outlier_mult
    
    return rw, indices

def concise(ax):
    locator = mdates.AutoDateLocator(minticks=3, maxticks=7)
    formatter = mdates.ConciseDateFormatter(locator)
    ax.xaxis.set_major_locator(locator)
    ax.xaxis.set_major_formatter(formatter)

71.2 load and process data

start = '2023-01-10 00:00:00'
n_steps = 1000
rw39, outlier_ind39 = random_walk_with_outliers(origin=0,
                                            n_steps=n_steps,
                                            perc_outliers=0.0031,
                                            outlier_mult=50,
                                            seed=39)
date_range = pd.date_range(start, periods=n_steps, freq='1min')
df = pd.DataFrame({'date': date_range, 'signal': rw39}).set_index('date')
start = df.index[0]
end = df.index[-1]

rw40, outlier_ind40 = random_walk_with_outliers(origin=0,
                                            n_steps=n_steps,
                                            perc_outliers=0.0031,
                                            outlier_mult=50,
                                            seed=40)
df['signal40'] = rw40

df.loc['2023-01-10 04:40:00', 'signal40'] = 43.0

# rw41, outlier_ind41 = random_walk_with_outliers(origin=0,
#                                             n_steps=n_steps,
#                                             perc_outliers=0.0031,
#                                             outlier_mult=50,
#                                             seed=41)
# df['signal41'] = rw41

71.3 stationary signal

fig, ax = plt.subplots(figsize=(8,4))
# plot signal
ax.plot(df['signal'], color="tab:blue")
# make graph look nice
ax.set(ylabel='values',
       xlim=[start,end],
       title="a stationary signal",
       ylim=[-45, 80])
concise(ax)
fig.savefig("signal_39_stationary.png", bbox_inches='tight')

71.4 visual inspection

fig, ax = plt.subplots(figsize=(8,4))
# plot signal
ax.plot(df['signal'], color="tab:blue")
# plot horizontal lines
ax.plot([start, end], [40]*2, color="black", alpha=0.4)
ax.text(end, 40, " 40", va="center")
ax.plot([start, end], [-30]*2, color="black", alpha=0.4)
ax.text(end, -30, " -30", va="center")
# find and plot outliers
outliers_index = df.index[(df['signal'] > 40) | (df['signal'] < -30)]
ax.plot(df.loc[outliers_index, 'signal'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax.set(ylabel='values',
       xlim=[start,end],
       title="visual inspection",
       ylim=[-45, 80])
concise(ax)
fig.savefig("outliers_visual_inspection.png", bbox_inches='tight')

71.5 mean +- 3 std

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

avg = df['signal'].mean()
std = df['signal'].std()

# plot signal
ax0.plot(df['signal'], color="tab:blue")

sns.kdeplot(data=df, y='signal', shade=True, ax=ax1)

pdf_xlim = ax1.get_xlim()
# plot horizontal lines
# mean
ax0.plot([start, end], [avg]*2, color="black", zorder=-10, alpha=0.7)
ax1.plot(pdf_xlim, [avg]*2, color="black", alpha=0.7)
ax1.text(1.1*pdf_xlim[1], avg, "mean", va="center")
# mean + std
ax0.plot([start, end], [avg+std]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [avg+std]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], avg+std, r"mean$+$std", va="center", alpha=0.4)
# mean - std
ax0.plot([start, end], [avg-std]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [avg-std]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], avg-std, r"mean$-$std", va="center", alpha=0.4)

n_sigma = 3
# mean + 3std
ax0.plot([start, end], [avg+n_sigma*std]*2, color="tab:red")
ax1.plot(pdf_xlim, [avg+n_sigma*std]*2, color="tab:red")
ax1.text(1.1*pdf_xlim[1], avg+n_sigma*std, r"mean$+3\cdot$std", va="center", color="tab:red")
# mean - 3std
ax0.plot([start, end], [avg-n_sigma*std]*2, color="tab:red")
ax1.plot(pdf_xlim, [avg-n_sigma*std]*2, color="tab:red")
ax1.text(1.1*pdf_xlim[1], avg-n_sigma*std, r"mean$-3\cdot$std", va="center", color="tab:red")

# find and plot outliers
outliers_index = df.index[(df['signal'] > avg + n_sigma*std) | 
                          (df['signal'] < avg - n_sigma*std)
                         ]
ax0.plot(df.loc[outliers_index, 'signal'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 80],
       title=r"threshold: mean $\pm3$ std",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 80],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )
fig.savefig("outliers_3sigma.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/653833997.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])

71.6 IQR

# get kdeplot data
fig, ax = plt.subplots(figsize=(8,4))
my_kde = sns.kdeplot(df['signal'], bw_adjust=0.5)
line = my_kde.lines[0]
kde_vals, kde_pdf = line.get_data()
kde_cdf = np.cumsum(kde_pdf) / np.sum(kde_pdf)

def find_nearest(array, value):
    return (np.abs(array - value)).argmin()

# Find the boundaries where the KDE is 25% and 75% of the area
Q1_index = find_nearest(kde_cdf, 0.25)
Q1_boundary = kde_vals[Q1_index]
Q3_index = find_nearest(kde_cdf, 0.75)
Q3_boundary = kde_vals[Q3_index]
IQR = Q3_boundary - Q1_boundary

fig, ax = plt.subplots(figsize=(8,4))

sns.kdeplot(df['signal'], ax=ax, shade=True, bw_adjust=0.5)

ax.fill_between(x=kde_vals[Q1_index:Q3_index],
                y1=kde_pdf[Q1_index:Q3_index],
                color="tab:pink"
                )

h = 0.02
ax.annotate("",
            xy=(Q1_boundary, h), xycoords='data',
            xytext=(Q3_boundary, h), textcoords='data',
            size=20,
            arrowprops=dict(arrowstyle="<->",
                            connectionstyle="arc3,rad=0.0",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5
                            ),
            )
ax.annotate("",
            xy=(Q1_boundary, h), xycoords='data',
            xytext=(Q3_boundary, h), textcoords='data',
            size=20,
            arrowprops=dict(arrowstyle="<->",
                            connectionstyle="arc3,rad=0.0",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5
                            ),
            )

ax.annotate("Q1\nquantile 0.25",
            xy=(Q1_boundary, 0.025), xycoords='data',
            xytext=(Q1_boundary-IQR, 0.040), textcoords='data',
            size=20,
            ha="right",
            va="top",
            arrowprops=dict(arrowstyle="->",
                            connectionstyle="angle,angleA=0,angleB=90,rad=5",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5,
                            color="black"
                            ),
            )

ax.annotate("Q3\nquantile 0.75",
            xy=(Q3_boundary, 0.025), xycoords='data',
            xytext=(Q3_boundary+IQR, 0.040), textcoords='data',
            size=20,
            ha="left",
            va="top",
            arrowprops=dict(arrowstyle="->",
                            connectionstyle="angle,angleA=0,angleB=90,rad=5",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5,
                            color="black"
                            ),
            )

ax.annotate(r"Q3 + 1.5$\cdot$IQR",
            xy=(Q3_boundary+1.5*IQR, 0.00), xycoords='data',
            xytext=(Q3_boundary+1.5*IQR, 0.015), textcoords='data',
            size=20,
            ha="left",
            arrowprops=dict(arrowstyle="->",
                            connectionstyle="angle,angleA=0,angleB=90,rad=5",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5,
                            color="black"
                            ),
            )

ax.annotate(r"Q1 - 1.5$\cdot$IQR",
            xy=(Q1_boundary-1.5*IQR, 0.00), xycoords='data',
            xytext=(Q1_boundary-1.5*IQR, 0.015), textcoords='data',
            size=20,
            ha="right",
            arrowprops=dict(arrowstyle="->",
                            connectionstyle="angle,angleA=0,angleB=90,rad=5",
                            shrinkA=0, shrinkB=0,
                            linewidth=2.5,
                            color="black"
                            ),
            )

ax.text(Q1_boundary+IQR/2, 0.018, "IQR",
        ha="center", va="top", color="white")
ax.text(Q1_boundary+IQR/2, 0.013, r"50% of"+"\nthe area",
        ha="center", va="top", color="black")

ax.set(xlim=[Q1_boundary-5*IQR, Q3_boundary+5*IQR],
       ylabel="pdf",)
fig.savefig("IQR_pdf.png", bbox_inches='tight')

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

median = df['signal'].quantile(0.50)
Q1 = df['signal'].quantile(0.25)
Q3 = df['signal'].quantile(0.75)
IQR = Q3 - Q1

# plot signal
ax0.plot(df['signal'], color="tab:blue")

my_kde = sns.kdeplot(data=df, y='signal', shade=True, ax=ax1, bw_adjust=0.5)
pdf_xlim =ax1.get_xlim()

kde_q1_idx = find_nearest(kde_vals, Q1)
kde_q3_idx = find_nearest(kde_vals, Q3)
ax1.fill_betweenx(y=kde_vals[kde_q1_idx:kde_q3_idx],
                  x1=kde_pdf[kde_q1_idx:kde_q3_idx],
                  color="tab:pink")
# plot horizontal lines
# Q3 + 1.5 IQR
ax0.plot([start, end], [Q3+1.5*IQR]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [Q3+1.5*IQR]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], Q3+1.5*IQR, "Q3+1.5*IQR", va="center", alpha=0.4)
# Q1 - 1.5 IQR
ax0.plot([start, end], [Q1-1.5*IQR]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [Q1-1.5*IQR]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], Q1-1.5*IQR, "Q1-1.5*IQR", va="center", alpha=0.4)

# find and plot outliers
outliers_index = df.index[(df['signal'] > Q3+1.5*IQR) | 
                          (df['signal'] < Q1-1.5*IQR)
                         ]
ax0.plot(df.loc[outliers_index, 'signal'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 80],
       title=r"threshold: Q$\pm1.5$ IQR",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 80],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )
fig.savefig("outliers_1.5IQR.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/3077230861.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])

71.7 non stationary signal

fig, ax = plt.subplots(figsize=(8,4))
# plot signal
ax.plot(df['signal40'], color="tab:blue")
# make graph look nice
ax.set(ylabel='values',
       xlim=[start,end],
       title="a non-stationary signal",
       ylim=[-45, 80])
concise(ax)
fig.savefig("signal_40_non_stationary.png", bbox_inches='tight')

71.8 running +- 3 std

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

avg = df['signal40'].mean()
std = df['signal40'].std()

# plot signal
ax0.plot(df['signal40'], color="tab:blue")

sns.kdeplot(data=df, y='signal40', shade=True, ax=ax1)


# plot horizontal lines
# mean
ax0.plot([start, end], [avg]*2, color="black", zorder=-10, alpha=0.7)
ax1.plot(pdf_xlim, [avg]*2, color="black", alpha=0.7)
ax1.text(1.1*pdf_xlim[1], avg, "mean", va="center")
# mean + std
ax0.plot([start, end], [avg+std]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [avg+std]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], avg+std, "mean+std", va="center", alpha=0.4)
# mean - std
ax0.plot([start, end], [avg-std]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [avg-std]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], avg-std, "mean-std", va="center", alpha=0.4)

n_sigma = 3
# mean + 3std
ax0.plot([start, end], [avg+n_sigma*std]*2, color="tab:red")
ax1.plot(pdf_xlim, [avg+n_sigma*std]*2, color="tab:red")
ax1.text(1.1*pdf_xlim[1], avg+n_sigma*std, r"mean $+3\times$std", va="center", color="tab:red")
# mean - 3std
ax0.plot([start, end], [avg-n_sigma*std]*2, color="tab:red")
ax1.plot(pdf_xlim, [avg-n_sigma*std]*2, color="tab:red")
ax1.text(1.1*pdf_xlim[1], avg-n_sigma*std, r"mean $-3\times$std", va="center", color="tab:red")

# find and plot outliers
outliers_index = df.index[(df['signal40'] > avg + n_sigma*std) | 
                          (df['signal40'] < avg - n_sigma*std)
                         ]
ax0.plot(df.loc[outliers_index, 'signal40'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 120],
       title=r"threshold: mean $\pm3$ std",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 120],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )
fig.savefig("outliers_3sigma_seed40.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/372016966.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])

df['signal40_rol_mean'] = df['signal40'].rolling('60min', center=True).mean()
df['signal40_rol_std'] = df['signal40'].rolling('60min', center=True).std()

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

avg = df['signal40'].mean()
std = df['signal40'].std()

# plot signal
ax0.plot(df['signal40'], color="tab:blue")

sns.kdeplot(data=df, y='signal40', shade=True, ax=ax1)

# mean
ax0.plot(df['signal40_rol_mean'], color="black", alpha=0.7, label="mean")
# mean +- 3 std
n_sigma = 3
ax0.plot(df['signal40_rol_mean']+ n_sigma*df['signal40_rol_std'], color="tab:red")
plot_threshold, = ax0.plot(df['signal40_rol_mean']- n_sigma*df['signal40_rol_std'],
                           color="tab:red", label="threshold")

# find and plot outliers
outliers_index = df.index[(df['signal40'] > df['signal40_rol_mean']+ n_sigma*df['signal40_rol_std']) | 
                          (df['signal40'] < df['signal40_rol_mean']- n_sigma*df['signal40_rol_std'])
                         ]
ax0.plot(df.loc[outliers_index, 'signal40'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 120],
       title=r"1-hour rolling window: mean $\pm3$ std",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 120],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )
ax0.legend(frameon=False, loc="upper left")

fig.savefig("outliers_rolling_3std.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/142534422.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])

71.9 running: Q +- IQR

fig, ax = plt.subplots(figsize=(8,4))

my_kde = sns.kdeplot(df['signal40'], bw_adjust=0.5)
line = my_kde.lines[0]
kde_vals, kde_pdf = line.get_data()
kde_cdf = np.cumsum(kde_pdf) / np.sum(kde_pdf)

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

median = df['signal40'].quantile(0.50)
Q1 = df['signal40'].quantile(0.25)
Q3 = df['signal40'].quantile(0.75)
IQR = Q3 - Q1

# plot signal
ax0.plot(df['signal40'], color="tab:blue")

my_kde = sns.kdeplot(data=df, y='signal40', shade=True, ax=ax1, bw_adjust=0.5)
pdf_xlim =ax1.get_xlim()

kde_q1_idx = find_nearest(kde_vals, Q1)
kde_q3_idx = find_nearest(kde_vals, Q3)
ax1.fill_betweenx(y=kde_vals[kde_q1_idx:kde_q3_idx],
                  x1=kde_pdf[kde_q1_idx:kde_q3_idx],
                  color="tab:pink")
# plot horizontal lines
# Q3 + 1.5 IQR
ax0.plot([start, end], [Q3+1.5*IQR]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [Q3+1.5*IQR]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], Q3+1.5*IQR, "Q3+1.5*IQR", va="center", alpha=0.4)
# Q1 - 1.5 IQR
ax0.plot([start, end], [Q1-1.5*IQR]*2, color="black", alpha=0.4)
ax1.plot(pdf_xlim, [Q1-1.5*IQR]*2, color="black", alpha=0.4)
ax1.text(1.1*pdf_xlim[1], Q1-1.5*IQR, "Q1-1.5*IQR", va="center", alpha=0.4)

# find and plot outliers
outliers_index = df.index[(df['signal40'] > Q3+1.5*IQR) | 
                          (df['signal40'] < Q1-1.5*IQR)
                         ]
ax0.plot(df.loc[outliers_index, 'signal40'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-50, 140],
       title=r"threshold: Q$\pm1.5$ IQR",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-50, 140],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )
fig.savefig("outliers_1.5IQR_seed40.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/678430211.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])

def Q1(window):
    return window.quantile(0.25)
def Q3(window):
    return window.quantile(0.75)

df['signal40_rol_Q1'] = df['signal40'].rolling('60min', center=True).apply(Q1)
df['signal40_rol_Q3'] = df['signal40'].rolling('60min', center=True).apply(Q3)
df['signal40_rol_IQR'] = df['signal40_rol_Q3'] - df['signal40_rol_Q1']

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

avg = df['signal40'].mean()
std = df['signal40'].std()

# plot signal
ax0.plot(df['signal40'], color="tab:blue")

sns.kdeplot(data=df, y='signal40', shade=True, ax=ax1)

# median
# ax0.plot(df['signal40_rol_median'], color="black", alpha=0.7, label="median")
# Q1 - 1.5 IQR
threshold_bottom = df['signal40_rol_Q1'] - 1.5*df['signal40_rol_IQR']
ax0.plot(threshold_bottom, color="tab:red")
# Q3 + 1.5 IQR
threshold_top = df['signal40_rol_Q3'] + 1.5*df['signal40_rol_IQR']
ax0.plot(threshold_top, color="tab:red")

# find and plot outliers
outliers_index = df.index[(df['signal40'] > threshold_top) | 
                          (df['signal40'] < threshold_bottom)
                         ]
ax0.plot(df.loc[outliers_index, 'signal40'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 120],
       title=r"1-hour rolling threshold: Q $\pm1.5$ IQR",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 120],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )

ax0.legend(frameon=False, loc="upper left")


fig.savefig("outliers_rolling_IQR.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/3108172465.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])
No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.

71.10 Hampel, running MAD

# def MAD(window):
#     return np.median(                                # 3. compute median
#                      np.abs(                         # 2. take absolute values
#                          window - np.median(window)  # 1. calculate residuals
#                          )
#                     )

k = 1.4826 # scale factor for Gaussian distribution
def MAD(window):
    return (window - np.median(window)).abs().median()

df['signal40_rol_mad'] = k * df['signal40'].rolling('60min', center=True).apply(MAD)
df['signal40_rol_median'] = df['signal40'].rolling('60min', center=True).median()

fig, ax = plt.subplots(figsize=(8,4))
gs = gridspec.GridSpec(1, 2, width_ratios=[1, 0.2], height_ratios=[1])
gs.update(left=0.10, right=0.90, top=0.95, bottom=0.13,
          hspace=0.02, wspace=0.02)

ax0 = plt.subplot(gs[0, 0])
ax1 = plt.subplot(gs[0, 1])

# plot signal
ax0.plot(df['signal40'], color="tab:blue")

sns.kdeplot(data=df, y='signal40', shade=True, ax=ax1)

# median
# ax0.plot(df['signal40_rol_median'], color="black", alpha=0.7, label="median")
# Q1 - 1.5 IQR
threshold_bottom = df['signal40_rol_median'] - 3 * df['signal40_rol_mad']
ax0.plot(threshold_bottom, color="tab:red")
# Q3 + 1.5 IQR
threshold_top = df['signal40_rol_median'] + 3 * df['signal40_rol_mad']
ax0.plot(threshold_top, color="tab:red")

# find and plot outliers
outliers_index = df.index[(df['signal40'] > threshold_top) | 
                          (df['signal40'] < threshold_bottom)
                         ]
ax0.plot(df.loc[outliers_index, 'signal40'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")
# make graph look nice
ax0.set(ylabel='values',
       xlim=[start,end],
       ylim=[-45, 120],
       title=r"1-hour rolling threshold: median $\pm3$ MAD",
       )
concise(ax0)
ax1.set(xlabel='pdf',
        ylabel='',
        ylim=[-45, 120],
        yticks=[],
        xticks=[0, 0.03],
        xticklabels=['0', '0.03']
        )

ax0.legend(frameon=False, loc="upper left")


fig.savefig("outliers_rolling_MAD.png", bbox_inches='tight')

/var/folders/c3/7hp0d36n6vv8jc9hm2440__00000gn/T/ipykernel_73326/3446982039.py:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.
  ax0 = plt.subplot(gs[0, 0])
No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.

71.11 stationary MAD

fig, ax = plt.subplots(figsize=(8,4))
# plot signal
ax.plot(df['signal'], color="tab:blue")

k = 1.4826 # scale factor for Gaussian distribution
mad = median_abs_deviation(df['signal'])
median = df['signal'].median()

xlim = ax.get_xlim()

# median +- 3*k*mad
ax.plot([start, end], [median]*2, color="black")
# ax.text(1.1, median, "median", va="center", transform=ax.transAxes,)

threshold_top = median+3*k*mad
threshold_bottom = median-3*k*mad

ax.plot([start, end], [threshold_bottom]*2, color="black", alpha=0.4)
ax.plot([start, end], [threshold_top]*2, color="black", alpha=0.4)

ax.annotate("median", xy=(1.02, median), xycoords=('axes fraction', 'data'),
            va='center')
ax.annotate(r"median$+3k\cdot$MAD", xy=(1.02, threshold_top), xycoords=('axes fraction', 'data'),
            va='center')
ax.annotate(r"median$-3k\cdot$MAD", xy=(1.02, threshold_bottom), xycoords=('axes fraction', 'data'),
            va='center')

# find and plot outliers
outliers_index = df.index[(df['signal'] > threshold_top) | 
                          (df['signal'] < threshold_bottom)
                         ]
ax.plot(df.loc[outliers_index, 'signal'], ls='None',
        marker='o', markerfacecolor='yellow', markersize=5,
        markeredgecolor="black")

# make graph look nice
ax.set(ylabel='values',
       xlim=[start,end],
       title="a stationary signal",
       ylim=[-45, 80])
concise(ax)
fig.savefig("outliers_MAD_stationary.png", bbox_inches='tight')

outliers_index

DatetimeIndex(['2023-01-10 04:09:00', '2023-01-10 14:05:00',
               '2023-01-10 15:14:00'],
              dtype='datetime64[ns]', name='date', freq=None)

71.12 save data as csv for later usage

start = '2023-01-01 00:00:00'
end = '2023-12-31 23:55:00'
# date_range = pd.date_range(start, end, freq='5min')
date_range = pd.date_range(start, end, freq='1D')
n_steps = len(date_range)
rw39, outlier_ind39 = random_walk_with_outliers(origin=0,
                                            n_steps=n_steps,
                                            perc_outliers=0.0031,
                                            outlier_mult=50,
                                            seed=39)
# date_range = pd.date_range(start, periods=n_steps, freq='1min')
df = pd.DataFrame({'date': date_range, 'A': rw39}).set_index('date')

%matplotlib widget
fig, ax = plt.subplots(figsize=(8,4))

ax.plot(df['A'])

df.loc['2023-03-02':'2023-03-10', 'A'] = -20.0

%matplotlib widget
fig, ax = plt.subplots(figsize=(8,4))
df['Adiff'] = df['A'].diff()
ax.plot(df['A'])
ax.plot(df['Adiff'])

(df.loc['2023-03-03':'2023-03-06', 'Adiff'] == np.zeros(4)).all()

True

df.loc['2023-03-03':'2023-03-06', 'Adiff']

date
2023-03-03    0.0
2023-03-04    0.0
2023-03-05    0.0
2023-03-06    0.0
Name: Adiff, dtype: float64

np.zeros(4)

array([0., 0., 0., 0.])

n_consecutive = 2

def n_zeros(series, N):
    """
    True if all series equals np.zeros(N)
    False otherwise
    """
    return (series == np.zeros(N)).all()

df['mask1'] = df['Adiff'].rolling(n_consecutive).apply(n_zeros, args=(n_consecutive,))
df['mask'] = 0.0
for i in range(len(df)):
    if df['mask1'][i] == 1.0:
        df['mask'][i-n_consecutive:i+1] = 1.0

/var/folders/kv/9cqw3y_s6c75xmgqm9n0t5d40000gn/T/ipykernel_6180/2921310703.py:5: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['mask'][i-n_consecutive:i+1] = 1.0

%matplotlib widget
fig, ax = plt.subplots(figsize=(8,4))
ax.plot(df['mask1'])
ax.plot(df['mask'])

df['mask'] = 0.0
for j in range(len(df)-n_consecutive):
    if (df['Adiff'][j:j+n_consecutive] == np.zeros(n_consecutive)).all():
        df['mask'][j:j+n_consecutive] = 1.0

/var/folders/kv/9cqw3y_s6c75xmgqm9n0t5d40000gn/T/ipykernel_6180/2364603426.py:4: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['mask'][j:j+n_consecutive] = 1.0

%matplotlib widget
fig, ax = plt.subplots(figsize=(8,4))
# ax.plot(df['mask1'])
ax.plot(df['mask'])
ax.plot(df['Adiff'])
ax.plot(df['A'])

71.13 generate datasets

def random_walk_with_outliers2(origin, steps_dist, outlier_dist, n_steps, perc_outliers=0.0, outlier_mult=10, seed=42):
    '''
    Function for generating a random time series based on random walk.
    It adds a specified percentage of outliers by multiplying the random walk step by a scalar.
    
    Parameters
    ------------
    origin : int
        The starting point of the series
    steps_dist : list of int or float
        step distribution
    outlier_dist : list of int or float
        outlier distribution
    n_steps : int
        Lenght of the series
    perc_outliers : float
        Percentage of outliers to introduce to the series [0.0-1.0]
    outlier_mult : float
        Scalar by which to multiply the RW increment to create an outlier
    seed : int
        Random seed

    Returns
    -----------
    rw : np.ndarray
        The generated random walk series with outliers
    indices : np.ndarray
        The indices of the introduced outliers 
    '''
    assert (perc_outliers >= 0.0) & (perc_outliers <= 1.0)
    
    #set seed for reproducibility
    np.random.seed(seed)
    
    # possible steps
    # steps = [-1, 1]

    # simulate steps
    steps = np.random.choice(a=steps_dist, size=n_steps-1)
    rw = np.append(origin, steps).cumsum(0)
    
    # add outliers
    n_outliers = int(np.round(perc_outliers * n_steps, 0))
    indices = np.random.randint(0, len(rw), n_outliers)
    outlier_jumps = np.random.choice(a=outlier_dist, size=n_outliers)
    # rw[indices] = rw[indices] + steps[indices + 1] * outlier_mult
    rw[indices] = rw[indices] + outlier_jumps
    
    return rw, indices

rw39test, _ = random_walk_with_outliers2(origin=0,
                                       steps_dist=np.random.normal(size=1000),
                                       outlier_dist=10*np.random.normal(loc=5.0, size=100),
                                       n_steps=n_steps,
                                       perc_outliers=0.0002,
                                       outlier_mult=50,
                                       seed=206)

df['B'] = rw39test

start = '2023-01-01 00:00:00'
end = '2023-12-31 23:00:00'
# date_range = pd.date_range(start, end, freq='5min')
date_range = pd.date_range(start, end, freq='1D')
n_steps = len(date_range)

# date_range = pd.date_range(start, periods=n_steps, freq='1min')
df = pd.DataFrame({'date': date_range, 'A': rw39test}).set_index('date')

df

	A
date
2023-01-01	0.000000
2023-01-02	-0.032027
2023-01-03	-0.586351
2023-01-04	-1.575972
2023-01-05	-2.726800
...	...
2023-12-27	6.651301
2023-12-28	6.415175
2023-12-29	7.603140
2023-12-30	8.668182
2023-12-31	8.472768

365 rows × 1 columns

df['unix_time'] = df.index.timestamp()

AttributeError: 'DatetimeIndex' object has no attribute 'timestamp'

df

	A
date
2023-01-01	0.000000
2023-01-02	-0.032027
2023-01-03	-0.586351
2023-01-04	-1.575972
2023-01-05	-2.726800
...	...
2023-12-27	6.651301
2023-12-28	6.415175
2023-12-29	7.603140
2023-12-30	8.668182
2023-12-31	8.472768

365 rows × 1 columns

rw02, outlier_ind02 = random_walk_with_outliers(origin=0,
                                            n_steps=n_steps,
                                            perc_outliers=0.0001,
                                            outlier_mult=500,
                                            seed=2)

df.loc['2023-01-02 23:00:00':'2023-01-03 03:00:00', 'E'] = np.nan

df.iloc[np.random.randint(0, high=len(df), size=600)]['E'] = np.nan

/var/folders/kv/9cqw3y_s6c75xmgqm9n0t5d40000gn/T/ipykernel_9222/2255542716.py:1: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df.iloc[np.random.randint(0, high=len(df), size=600)]['E'] = np.nan

%matplotlib widget
fig, ax = plt.subplots(figsize=(8,4))
# ax.plot(df['B'])
ax.plot(rw39test)

df['date'] = df.index.strftime('%d%m%Y')
df['time'] = df.index.strftime('%H:%M:%S')

df['unix time (s)'] = df.index.view('int64') / 1e9
df

	A	date	unix time (s)
date
2023-01-01	0.000000	2023-01-01	1.672531e+09
2023-01-02	-0.032027	2023-01-02	1.672618e+09
2023-01-03	-0.586351	2023-01-03	1.672704e+09
2023-01-04	-1.575972	2023-01-04	1.672790e+09
2023-01-05	-2.726800	2023-01-05	1.672877e+09
...	...	...	...
2023-12-27	6.651301	2023-12-27	1.703635e+09
2023-12-28	6.415175	2023-12-28	1.703722e+09
2023-12-29	7.603140	2023-12-29	1.703808e+09
2023-12-30	8.668182	2023-12-30	1.703894e+09
2023-12-31	8.472768	2023-12-31	1.703981e+09

365 rows × 3 columns

df.drop(columns=['date'], inplace=True)
df

	A	unix time (s)
date
2023-01-01	0.000000	1.672531e+09
2023-01-02	-0.032027	1.672618e+09
2023-01-03	-0.586351	1.672704e+09
2023-01-04	-1.575972	1.672790e+09
2023-01-05	-2.726800	1.672877e+09
...	...	...
2023-12-27	6.651301	1.703635e+09
2023-12-28	6.415175	1.703722e+09
2023-12-29	7.603140	1.703808e+09
2023-12-30	8.668182	1.703894e+09
2023-12-31	8.472768	1.703981e+09

365 rows × 2 columns

df.loc['2023-10-11 03:00:00', 'B'] = '-'

df.to_csv('cleaning3.csv', index=False, sep=' ')

df['A']

date
2023-01-01 00:00:00     0.000000
2023-01-01 01:00:00    -2.027536
2023-01-01 02:00:00    -2.690617
2023-01-01 03:00:00    -1.985990
2023-01-01 04:00:00    -2.290898
                         ...    
2023-12-31 19:00:00   -74.514645
2023-12-31 20:00:00   -74.738058
2023-12-31 21:00:00   -75.848425
2023-12-31 22:00:00   -77.272183
2023-12-31 23:00:00   -76.557400
Name: A, Length: 8760, dtype: float64

import datetime
import time

dt = datetime.datetime.now()
timestamp = time.mktime(dt.timetuple())
print(timestamp)

1705330442.0

timestamp

1705330442.0

dt

datetime.datetime(2024, 1, 15, 16, 54, 2, 420872)

dt.timetuple()

time.struct_time(tm_year=2024, tm_mon=1, tm_mday=15, tm_hour=16, tm_min=54, tm_sec=2, tm_wday=0, tm_yday=15, tm_isdst=-1)

pd.to_datetime(1705330442.0, unit='s')

Timestamp('2024-01-15 14:54:02')

(df.index.values).apply(lambda x: x.timetuple())

AttributeError: 'numpy.ndarray' object has no attribute 'apply'

df['date'] = df.index.strftime('%Y-%m-%d')

df['date'].apply(lambda x: x.timetuple())

AttributeError: 'str' object has no attribute 'timetuple'

df.index.to_pydatetime()

array([datetime.datetime(2023, 1, 1, 0, 0),
       datetime.datetime(2023, 1, 2, 0, 0),
       datetime.datetime(2023, 1, 3, 0, 0),
       datetime.datetime(2023, 1, 4, 0, 0),
       datetime.datetime(2023, 1, 5, 0, 0),
       datetime.datetime(2023, 1, 6, 0, 0),
       datetime.datetime(2023, 1, 7, 0, 0),
       datetime.datetime(2023, 1, 8, 0, 0),
       datetime.datetime(2023, 1, 9, 0, 0),
       datetime.datetime(2023, 1, 10, 0, 0),
       datetime.datetime(2023, 1, 11, 0, 0),
       datetime.datetime(2023, 1, 12, 0, 0),
       datetime.datetime(2023, 1, 13, 0, 0),
       datetime.datetime(2023, 1, 14, 0, 0),
       datetime.datetime(2023, 1, 15, 0, 0),
       datetime.datetime(2023, 1, 16, 0, 0),
       datetime.datetime(2023, 1, 17, 0, 0),
       datetime.datetime(2023, 1, 18, 0, 0),
       datetime.datetime(2023, 1, 19, 0, 0),
       datetime.datetime(2023, 1, 20, 0, 0),
       datetime.datetime(2023, 1, 21, 0, 0),
       datetime.datetime(2023, 1, 22, 0, 0),
       datetime.datetime(2023, 1, 23, 0, 0),
       datetime.datetime(2023, 1, 24, 0, 0),
       datetime.datetime(2023, 1, 25, 0, 0),
       datetime.datetime(2023, 1, 26, 0, 0),
       datetime.datetime(2023, 1, 27, 0, 0),
       datetime.datetime(2023, 1, 28, 0, 0),
       datetime.datetime(2023, 1, 29, 0, 0),
       datetime.datetime(2023, 1, 30, 0, 0),
       datetime.datetime(2023, 1, 31, 0, 0),
       datetime.datetime(2023, 2, 1, 0, 0),
       datetime.datetime(2023, 2, 2, 0, 0),
       datetime.datetime(2023, 2, 3, 0, 0),
       datetime.datetime(2023, 2, 4, 0, 0),
       datetime.datetime(2023, 2, 5, 0, 0),
       datetime.datetime(2023, 2, 6, 0, 0),
       datetime.datetime(2023, 2, 7, 0, 0),
       datetime.datetime(2023, 2, 8, 0, 0),
       datetime.datetime(2023, 2, 9, 0, 0),
       datetime.datetime(2023, 2, 10, 0, 0),
       datetime.datetime(2023, 2, 11, 0, 0),
       datetime.datetime(2023, 2, 12, 0, 0),
       datetime.datetime(2023, 2, 13, 0, 0),
       datetime.datetime(2023, 2, 14, 0, 0),
       datetime.datetime(2023, 2, 15, 0, 0),
       datetime.datetime(2023, 2, 16, 0, 0),
       datetime.datetime(2023, 2, 17, 0, 0),
       datetime.datetime(2023, 2, 18, 0, 0),
       datetime.datetime(2023, 2, 19, 0, 0),
       datetime.datetime(2023, 2, 20, 0, 0),
       datetime.datetime(2023, 2, 21, 0, 0),
       datetime.datetime(2023, 2, 22, 0, 0),
       datetime.datetime(2023, 2, 23, 0, 0),
       datetime.datetime(2023, 2, 24, 0, 0),
       datetime.datetime(2023, 2, 25, 0, 0),
       datetime.datetime(2023, 2, 26, 0, 0),
       datetime.datetime(2023, 2, 27, 0, 0),
       datetime.datetime(2023, 2, 28, 0, 0),
       datetime.datetime(2023, 3, 1, 0, 0),
       datetime.datetime(2023, 3, 2, 0, 0),
       datetime.datetime(2023, 3, 3, 0, 0),
       datetime.datetime(2023, 3, 4, 0, 0),
       datetime.datetime(2023, 3, 5, 0, 0),
       datetime.datetime(2023, 3, 6, 0, 0),
       datetime.datetime(2023, 3, 7, 0, 0),
       datetime.datetime(2023, 3, 8, 0, 0),
       datetime.datetime(2023, 3, 9, 0, 0),
       datetime.datetime(2023, 3, 10, 0, 0),
       datetime.datetime(2023, 3, 11, 0, 0),
       datetime.datetime(2023, 3, 12, 0, 0),
       datetime.datetime(2023, 3, 13, 0, 0),
       datetime.datetime(2023, 3, 14, 0, 0),
       datetime.datetime(2023, 3, 15, 0, 0),
       datetime.datetime(2023, 3, 16, 0, 0),
       datetime.datetime(2023, 3, 17, 0, 0),
       datetime.datetime(2023, 3, 18, 0, 0),
       datetime.datetime(2023, 3, 19, 0, 0),
       datetime.datetime(2023, 3, 20, 0, 0),
       datetime.datetime(2023, 3, 21, 0, 0),
       datetime.datetime(2023, 3, 22, 0, 0),
       datetime.datetime(2023, 3, 23, 0, 0),
       datetime.datetime(2023, 3, 24, 0, 0),
       datetime.datetime(2023, 3, 25, 0, 0),
       datetime.datetime(2023, 3, 26, 0, 0),
       datetime.datetime(2023, 3, 27, 0, 0),
       datetime.datetime(2023, 3, 28, 0, 0),
       datetime.datetime(2023, 3, 29, 0, 0),
       datetime.datetime(2023, 3, 30, 0, 0),
       datetime.datetime(2023, 3, 31, 0, 0),
       datetime.datetime(2023, 4, 1, 0, 0),
       datetime.datetime(2023, 4, 2, 0, 0),
       datetime.datetime(2023, 4, 3, 0, 0),
       datetime.datetime(2023, 4, 4, 0, 0),
       datetime.datetime(2023, 4, 5, 0, 0),
       datetime.datetime(2023, 4, 6, 0, 0),
       datetime.datetime(2023, 4, 7, 0, 0),
       datetime.datetime(2023, 4, 8, 0, 0),
       datetime.datetime(2023, 4, 9, 0, 0),
       datetime.datetime(2023, 4, 10, 0, 0),
       datetime.datetime(2023, 4, 11, 0, 0),
       datetime.datetime(2023, 4, 12, 0, 0),
       datetime.datetime(2023, 4, 13, 0, 0),
       datetime.datetime(2023, 4, 14, 0, 0),
       datetime.datetime(2023, 4, 15, 0, 0),
       datetime.datetime(2023, 4, 16, 0, 0),
       datetime.datetime(2023, 4, 17, 0, 0),
       datetime.datetime(2023, 4, 18, 0, 0),
       datetime.datetime(2023, 4, 19, 0, 0),
       datetime.datetime(2023, 4, 20, 0, 0),
       datetime.datetime(2023, 4, 21, 0, 0),
       datetime.datetime(2023, 4, 22, 0, 0),
       datetime.datetime(2023, 4, 23, 0, 0),
       datetime.datetime(2023, 4, 24, 0, 0),
       datetime.datetime(2023, 4, 25, 0, 0),
       datetime.datetime(2023, 4, 26, 0, 0),
       datetime.datetime(2023, 4, 27, 0, 0),
       datetime.datetime(2023, 4, 28, 0, 0),
       datetime.datetime(2023, 4, 29, 0, 0),
       datetime.datetime(2023, 4, 30, 0, 0),
       datetime.datetime(2023, 5, 1, 0, 0),
       datetime.datetime(2023, 5, 2, 0, 0),
       datetime.datetime(2023, 5, 3, 0, 0),
       datetime.datetime(2023, 5, 4, 0, 0),
       datetime.datetime(2023, 5, 5, 0, 0),
       datetime.datetime(2023, 5, 6, 0, 0),
       datetime.datetime(2023, 5, 7, 0, 0),
       datetime.datetime(2023, 5, 8, 0, 0),
       datetime.datetime(2023, 5, 9, 0, 0),
       datetime.datetime(2023, 5, 10, 0, 0),
       datetime.datetime(2023, 5, 11, 0, 0),
       datetime.datetime(2023, 5, 12, 0, 0),
       datetime.datetime(2023, 5, 13, 0, 0),
       datetime.datetime(2023, 5, 14, 0, 0),
       datetime.datetime(2023, 5, 15, 0, 0),
       datetime.datetime(2023, 5, 16, 0, 0),
       datetime.datetime(2023, 5, 17, 0, 0),
       datetime.datetime(2023, 5, 18, 0, 0),
       datetime.datetime(2023, 5, 19, 0, 0),
       datetime.datetime(2023, 5, 20, 0, 0),
       datetime.datetime(2023, 5, 21, 0, 0),
       datetime.datetime(2023, 5, 22, 0, 0),
       datetime.datetime(2023, 5, 23, 0, 0),
       datetime.datetime(2023, 5, 24, 0, 0),
       datetime.datetime(2023, 5, 25, 0, 0),
       datetime.datetime(2023, 5, 26, 0, 0),
       datetime.datetime(2023, 5, 27, 0, 0),
       datetime.datetime(2023, 5, 28, 0, 0),
       datetime.datetime(2023, 5, 29, 0, 0),
       datetime.datetime(2023, 5, 30, 0, 0),
       datetime.datetime(2023, 5, 31, 0, 0),
       datetime.datetime(2023, 6, 1, 0, 0),
       datetime.datetime(2023, 6, 2, 0, 0),
       datetime.datetime(2023, 6, 3, 0, 0),
       datetime.datetime(2023, 6, 4, 0, 0),
       datetime.datetime(2023, 6, 5, 0, 0),
       datetime.datetime(2023, 6, 6, 0, 0),
       datetime.datetime(2023, 6, 7, 0, 0),
       datetime.datetime(2023, 6, 8, 0, 0),
       datetime.datetime(2023, 6, 9, 0, 0),
       datetime.datetime(2023, 6, 10, 0, 0),
       datetime.datetime(2023, 6, 11, 0, 0),
       datetime.datetime(2023, 6, 12, 0, 0),
       datetime.datetime(2023, 6, 13, 0, 0),
       datetime.datetime(2023, 6, 14, 0, 0),
       datetime.datetime(2023, 6, 15, 0, 0),
       datetime.datetime(2023, 6, 16, 0, 0),
       datetime.datetime(2023, 6, 17, 0, 0),
       datetime.datetime(2023, 6, 18, 0, 0),
       datetime.datetime(2023, 6, 19, 0, 0),
       datetime.datetime(2023, 6, 20, 0, 0),
       datetime.datetime(2023, 6, 21, 0, 0),
       datetime.datetime(2023, 6, 22, 0, 0),
       datetime.datetime(2023, 6, 23, 0, 0),
       datetime.datetime(2023, 6, 24, 0, 0),
       datetime.datetime(2023, 6, 25, 0, 0),
       datetime.datetime(2023, 6, 26, 0, 0),
       datetime.datetime(2023, 6, 27, 0, 0),
       datetime.datetime(2023, 6, 28, 0, 0),
       datetime.datetime(2023, 6, 29, 0, 0),
       datetime.datetime(2023, 6, 30, 0, 0),
       datetime.datetime(2023, 7, 1, 0, 0),
       datetime.datetime(2023, 7, 2, 0, 0),
       datetime.datetime(2023, 7, 3, 0, 0),
       datetime.datetime(2023, 7, 4, 0, 0),
       datetime.datetime(2023, 7, 5, 0, 0),
       datetime.datetime(2023, 7, 6, 0, 0),
       datetime.datetime(2023, 7, 7, 0, 0),
       datetime.datetime(2023, 7, 8, 0, 0),
       datetime.datetime(2023, 7, 9, 0, 0),
       datetime.datetime(2023, 7, 10, 0, 0),
       datetime.datetime(2023, 7, 11, 0, 0),
       datetime.datetime(2023, 7, 12, 0, 0),
       datetime.datetime(2023, 7, 13, 0, 0),
       datetime.datetime(2023, 7, 14, 0, 0),
       datetime.datetime(2023, 7, 15, 0, 0),
       datetime.datetime(2023, 7, 16, 0, 0),
       datetime.datetime(2023, 7, 17, 0, 0),
       datetime.datetime(2023, 7, 18, 0, 0),
       datetime.datetime(2023, 7, 19, 0, 0),
       datetime.datetime(2023, 7, 20, 0, 0),
       datetime.datetime(2023, 7, 21, 0, 0),
       datetime.datetime(2023, 7, 22, 0, 0),
       datetime.datetime(2023, 7, 23, 0, 0),
       datetime.datetime(2023, 7, 24, 0, 0),
       datetime.datetime(2023, 7, 25, 0, 0),
       datetime.datetime(2023, 7, 26, 0, 0),
       datetime.datetime(2023, 7, 27, 0, 0),
       datetime.datetime(2023, 7, 28, 0, 0),
       datetime.datetime(2023, 7, 29, 0, 0),
       datetime.datetime(2023, 7, 30, 0, 0),
       datetime.datetime(2023, 7, 31, 0, 0),
       datetime.datetime(2023, 8, 1, 0, 0),
       datetime.datetime(2023, 8, 2, 0, 0),
       datetime.datetime(2023, 8, 3, 0, 0),
       datetime.datetime(2023, 8, 4, 0, 0),
       datetime.datetime(2023, 8, 5, 0, 0),
       datetime.datetime(2023, 8, 6, 0, 0),
       datetime.datetime(2023, 8, 7, 0, 0),
       datetime.datetime(2023, 8, 8, 0, 0),
       datetime.datetime(2023, 8, 9, 0, 0),
       datetime.datetime(2023, 8, 10, 0, 0),
       datetime.datetime(2023, 8, 11, 0, 0),
       datetime.datetime(2023, 8, 12, 0, 0),
       datetime.datetime(2023, 8, 13, 0, 0),
       datetime.datetime(2023, 8, 14, 0, 0),
       datetime.datetime(2023, 8, 15, 0, 0),
       datetime.datetime(2023, 8, 16, 0, 0),
       datetime.datetime(2023, 8, 17, 0, 0),
       datetime.datetime(2023, 8, 18, 0, 0),
       datetime.datetime(2023, 8, 19, 0, 0),
       datetime.datetime(2023, 8, 20, 0, 0),
       datetime.datetime(2023, 8, 21, 0, 0),
       datetime.datetime(2023, 8, 22, 0, 0),
       datetime.datetime(2023, 8, 23, 0, 0),
       datetime.datetime(2023, 8, 24, 0, 0),
       datetime.datetime(2023, 8, 25, 0, 0),
       datetime.datetime(2023, 8, 26, 0, 0),
       datetime.datetime(2023, 8, 27, 0, 0),
       datetime.datetime(2023, 8, 28, 0, 0),
       datetime.datetime(2023, 8, 29, 0, 0),
       datetime.datetime(2023, 8, 30, 0, 0),
       datetime.datetime(2023, 8, 31, 0, 0),
       datetime.datetime(2023, 9, 1, 0, 0),
       datetime.datetime(2023, 9, 2, 0, 0),
       datetime.datetime(2023, 9, 3, 0, 0),
       datetime.datetime(2023, 9, 4, 0, 0),
       datetime.datetime(2023, 9, 5, 0, 0),
       datetime.datetime(2023, 9, 6, 0, 0),
       datetime.datetime(2023, 9, 7, 0, 0),
       datetime.datetime(2023, 9, 8, 0, 0),
       datetime.datetime(2023, 9, 9, 0, 0),
       datetime.datetime(2023, 9, 10, 0, 0),
       datetime.datetime(2023, 9, 11, 0, 0),
       datetime.datetime(2023, 9, 12, 0, 0),
       datetime.datetime(2023, 9, 13, 0, 0),
       datetime.datetime(2023, 9, 14, 0, 0),
       datetime.datetime(2023, 9, 15, 0, 0),
       datetime.datetime(2023, 9, 16, 0, 0),
       datetime.datetime(2023, 9, 17, 0, 0),
       datetime.datetime(2023, 9, 18, 0, 0),
       datetime.datetime(2023, 9, 19, 0, 0),
       datetime.datetime(2023, 9, 20, 0, 0),
       datetime.datetime(2023, 9, 21, 0, 0),
       datetime.datetime(2023, 9, 22, 0, 0),
       datetime.datetime(2023, 9, 23, 0, 0),
       datetime.datetime(2023, 9, 24, 0, 0),
       datetime.datetime(2023, 9, 25, 0, 0),
       datetime.datetime(2023, 9, 26, 0, 0),
       datetime.datetime(2023, 9, 27, 0, 0),
       datetime.datetime(2023, 9, 28, 0, 0),
       datetime.datetime(2023, 9, 29, 0, 0),
       datetime.datetime(2023, 9, 30, 0, 0),
       datetime.datetime(2023, 10, 1, 0, 0),
       datetime.datetime(2023, 10, 2, 0, 0),
       datetime.datetime(2023, 10, 3, 0, 0),
       datetime.datetime(2023, 10, 4, 0, 0),
       datetime.datetime(2023, 10, 5, 0, 0),
       datetime.datetime(2023, 10, 6, 0, 0),
       datetime.datetime(2023, 10, 7, 0, 0),
       datetime.datetime(2023, 10, 8, 0, 0),
       datetime.datetime(2023, 10, 9, 0, 0),
       datetime.datetime(2023, 10, 10, 0, 0),
       datetime.datetime(2023, 10, 11, 0, 0),
       datetime.datetime(2023, 10, 12, 0, 0),
       datetime.datetime(2023, 10, 13, 0, 0),
       datetime.datetime(2023, 10, 14, 0, 0),
       datetime.datetime(2023, 10, 15, 0, 0),
       datetime.datetime(2023, 10, 16, 0, 0),
       datetime.datetime(2023, 10, 17, 0, 0),
       datetime.datetime(2023, 10, 18, 0, 0),
       datetime.datetime(2023, 10, 19, 0, 0),
       datetime.datetime(2023, 10, 20, 0, 0),
       datetime.datetime(2023, 10, 21, 0, 0),
       datetime.datetime(2023, 10, 22, 0, 0),
       datetime.datetime(2023, 10, 23, 0, 0),
       datetime.datetime(2023, 10, 24, 0, 0),
       datetime.datetime(2023, 10, 25, 0, 0),
       datetime.datetime(2023, 10, 26, 0, 0),
       datetime.datetime(2023, 10, 27, 0, 0),
       datetime.datetime(2023, 10, 28, 0, 0),
       datetime.datetime(2023, 10, 29, 0, 0),
       datetime.datetime(2023, 10, 30, 0, 0),
       datetime.datetime(2023, 10, 31, 0, 0),
       datetime.datetime(2023, 11, 1, 0, 0),
       datetime.datetime(2023, 11, 2, 0, 0),
       datetime.datetime(2023, 11, 3, 0, 0),
       datetime.datetime(2023, 11, 4, 0, 0),
       datetime.datetime(2023, 11, 5, 0, 0),
       datetime.datetime(2023, 11, 6, 0, 0),
       datetime.datetime(2023, 11, 7, 0, 0),
       datetime.datetime(2023, 11, 8, 0, 0),
       datetime.datetime(2023, 11, 9, 0, 0),
       datetime.datetime(2023, 11, 10, 0, 0),
       datetime.datetime(2023, 11, 11, 0, 0),
       datetime.datetime(2023, 11, 12, 0, 0),
       datetime.datetime(2023, 11, 13, 0, 0),
       datetime.datetime(2023, 11, 14, 0, 0),
       datetime.datetime(2023, 11, 15, 0, 0),
       datetime.datetime(2023, 11, 16, 0, 0),
       datetime.datetime(2023, 11, 17, 0, 0),
       datetime.datetime(2023, 11, 18, 0, 0),
       datetime.datetime(2023, 11, 19, 0, 0),
       datetime.datetime(2023, 11, 20, 0, 0),
       datetime.datetime(2023, 11, 21, 0, 0),
       datetime.datetime(2023, 11, 22, 0, 0),
       datetime.datetime(2023, 11, 23, 0, 0),
       datetime.datetime(2023, 11, 24, 0, 0),
       datetime.datetime(2023, 11, 25, 0, 0),
       datetime.datetime(2023, 11, 26, 0, 0),
       datetime.datetime(2023, 11, 27, 0, 0),
       datetime.datetime(2023, 11, 28, 0, 0),
       datetime.datetime(2023, 11, 29, 0, 0),
       datetime.datetime(2023, 11, 30, 0, 0),
       datetime.datetime(2023, 12, 1, 0, 0),
       datetime.datetime(2023, 12, 2, 0, 0),
       datetime.datetime(2023, 12, 3, 0, 0),
       datetime.datetime(2023, 12, 4, 0, 0),
       datetime.datetime(2023, 12, 5, 0, 0),
       datetime.datetime(2023, 12, 6, 0, 0),
       datetime.datetime(2023, 12, 7, 0, 0),
       datetime.datetime(2023, 12, 8, 0, 0),
       datetime.datetime(2023, 12, 9, 0, 0),
       datetime.datetime(2023, 12, 10, 0, 0),
       datetime.datetime(2023, 12, 11, 0, 0),
       datetime.datetime(2023, 12, 12, 0, 0),
       datetime.datetime(2023, 12, 13, 0, 0),
       datetime.datetime(2023, 12, 14, 0, 0),
       datetime.datetime(2023, 12, 15, 0, 0),
       datetime.datetime(2023, 12, 16, 0, 0),
       datetime.datetime(2023, 12, 17, 0, 0),
       datetime.datetime(2023, 12, 18, 0, 0),
       datetime.datetime(2023, 12, 19, 0, 0),
       datetime.datetime(2023, 12, 20, 0, 0),
       datetime.datetime(2023, 12, 21, 0, 0),
       datetime.datetime(2023, 12, 22, 0, 0),
       datetime.datetime(2023, 12, 23, 0, 0),
       datetime.datetime(2023, 12, 24, 0, 0),
       datetime.datetime(2023, 12, 25, 0, 0),
       datetime.datetime(2023, 12, 26, 0, 0),
       datetime.datetime(2023, 12, 27, 0, 0),
       datetime.datetime(2023, 12, 28, 0, 0),
       datetime.datetime(2023, 12, 29, 0, 0),
       datetime.datetime(2023, 12, 30, 0, 0),
       datetime.datetime(2023, 12, 31, 0, 0)], dtype=object)

unix = df.index.view('int64') / 1e9

unix[0]

1672531200.0

pd.to_datetime(unix, unit='s')

DatetimeIndex(['2023-01-01', '2023-01-02', '2023-01-03', '2023-01-04',
               '2023-01-05', '2023-01-06', '2023-01-07', '2023-01-08',
               '2023-01-09', '2023-01-10',
               ...
               '2023-12-22', '2023-12-23', '2023-12-24', '2023-12-25',
               '2023-12-26', '2023-12-27', '2023-12-28', '2023-12-29',
               '2023-12-30', '2023-12-31'],
              dtype='datetime64[ns]', length=365, freq=None)