Õunte suuruste seast anomaalia leidmine

Õunte andmed võeti veebiaadressilt

Anomaaliad püüti leida meetoditega

• One-class Vector support machine
• PCA-based anomaly detection

from azureml import Workspace
ws = Workspace()
experiment = ws.experiments['66e373b2084d4ffa9395c0e34ce9ccaa.f-id.e773e947bd7d4c68b4da26e992d0122f']
ds = experiment.get_intermediate_dataset(
    node_id='ebd1c86f-2ee7-47f3-b8d9-428740f5e5d8-36086',
    port_name='Results dataset',
    data_type_id='GenericCSV'
)
frame = ds.to_dataframe()

frame.head()

	august	september	Label	Scored Labels	Scored Probabilities
0	6.0	7.9	1	0	-1.043081e-07
1	4.0	5.7	1	0	-7.145673e-03
2	5.2	6.6	1	0	-5.228825e-03
3	4.1	5.4	1	0	-7.064961e-03
4	5.7	7.9	1	0	-9.832531e-04

frame.tail()

	august	september	Label	Scored Labels	Scored Probabilities
100	5.6	4.2	2	0	-0.002141
101	5.7	4.1	2	0	-0.001486
102	10.1	10.2	2	0	0.032764
103	0.1	1.2	2	0	0.019103
104	1.0	1.1	2	0	0.015089

import matplotlib.pyplot as plt
plt.scatter(frame.august, frame.september)
plt.show()

plt.scatter(frame.august, frame.september, c=frame.Label, edgecolors="none")
plt.show()

frame["Scored Probabilities"]
frame["Scored Probabilities"].mean()
tulp=frame["Scored Probabilities"]
frame.normskoor=(tulp-tulp.min())/(tulp.max()-tulp.min())
frame.normskoor.max()

1.0

plt.scatter(frame.august, frame.september, c=frame.Label, s=frame.normskoor*20+0.5, edgecolors="none")
plt.show()

tavalised=frame[frame.Label==1][["august", "september"]]
plt.scatter(tavalised.august, tavalised.september, edgecolor="none")
plt.title("Õunte suurused (cm)")
plt.xlabel("august")
plt.ylabel("september")
plt.xlim([0, 10])
plt.ylim([0, 10])
plt.plot([tavalised.august.mean(), tavalised.august.mean()], [0, tavalised.september.max()], 
            linewidth=0.6, linestyle="dotted", color="gray")
#Tõmmake septembri keskmist näitav horisontaaljoon
plt.plot( [0, tavalised.august.max()], [tavalised.september.mean(), tavalised.september.mean()], 
            linewidth=0.6, linestyle="dotted", color="gray")
plt.show()