davrot/pytutorial
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update README.md

Browse source 
Signed-off-by: David Rotermund <54365609+davrot@users.noreply.github.com>
This commit is contained in:
David Rotermund 2024-02-16 00:37:22 +01:00 committed by GitHub
parent 8acfe8ed77
commit 6cc6c9e48b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 33 additions and 41 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

70
numpy/roc/README.md
View file

@ -60,7 +60,9 @@ a_x = rng.normal(1.5, 1.0, size=(5000))
b_x = rng.normal(0.0, 1.0, size=(5000))
data_data = np.concatenate([a_x, b_x])
data_class = np.concatenate([np.full_like(a_x, -1), np.full_like(b_x, +1)])
data_class = np.concatenate(
[np.full_like(a_x, -1 / a_x.shape[0]), np.full_like(b_x, +1 / b_x.shape[0])]
)
idx = np.argsort(data_data)
data_data = data_data[idx]
@ -76,6 +78,7 @@ plt.ylabel("Cumsum of the classes")
plt.xlabel("Sorted sample id")
plt.show()
```
![Image2](image2.png)
## How to create an estimate from the ROC cumsum maximum
@ -90,41 +93,42 @@ a_x = rng.normal(1.5, 1.0, size=(5000))
b_x = rng.normal(0.0, 1.0, size=(5000))
data_data = np.concatenate([a_x, b_x])
data_class = np.concatenate([np.full_like(a_x, -1), np.full_like(b_x, +1)])
data_class = np.concatenate(
[np.full_like(a_x, -1 / a_x.shape[0]), np.full_like(b_x, +1 / b_x.shape[0])]
)
data_class_id = np.concatenate([np.full_like(a_x, -1), np.full_like(b_x, +1)])
idx = np.argsort(data_data)
data_data = data_data[idx]
data_class = data_class[idx]
data_class_id = data_class_id[idx]
data_cumsum = np.cumsum(data_class)
border = np.argmax(np.abs(data_cumsum))
variant_a = (data_class[:border] == -1).sum() + (data_class[border:] == +1).sum()
variant_b = (data_class[:border] == +1).sum() + (data_class[border:] == -1).sum()
estimate_a = np.concatenate(
(np.full_like(data_class[:border], -1), np.full_like(data_class[border:], +1))
if data_cumsum[border] < 0:
estimate = np.concatenate(
(
np.full_like(data_class[: border + 1], -1),
np.full_like(data_class[border + 1 :], +1),
)
estimate_b = np.concatenate(
(np.full_like(data_class[:border], +1), np.full_like(data_class[border:], -1))
)
if variant_a > variant_b:
print("We will use: Estimate A")
estimate = estimate_a
else:
print("We will use: Estimate B")
estimate = estimate_b
estimate = np.concatenate(
(
np.full_like(data_class[: border + 1], +1),
np.full_like(data_class[border + 1 :], -1),
)
)
performance = 100.0 * (data_class == estimate).sum() / data_class.shape[0]
performance = 100.0 * (data_class_id == estimate).sum() / data_class_id.shape[0]
print(f"Performance: {performance}% correct")
plt.subplot(3, 1, 1)
idx_a = np.where(data_class == -1)[0]
idx_b = np.where(data_class == +1)[0]
plt.subplot(2, 1, 1)
idx_a = np.where(data_class < 0)[0]
idx_b = np.where(data_class > 0)[0]
idx = np.arange(0, data_class.shape[0])
plt.plot(data_data[idx_a], np.zeros_like(idx_a), "c*")
@ -132,34 +136,22 @@ plt.plot(data_data[idx_b], np.zeros_like(idx_b), "m.")
plt.yticks([])
plt.title("Data")
plt.subplot(3, 1, 2)
idx_a = np.where(estimate_a == -1)[0]
idx_b = np.where(estimate_a == +1)[0]
idx = np.arange(0, estimate_a.shape[0])
plt.subplot(2, 1, 2)
idx_a = np.where(estimate < 0)[0]
idx_b = np.where(estimate > 0)[0]
idx = np.arange(0, estimate.shape[0])
plt.plot(data_data[idx_a], np.zeros_like(idx_a), "c*")
plt.plot(data_data[idx_b], np.zeros_like(idx_b), "m.")
plt.yticks([])
plt.title("Estimate A")
plt.subplot(3, 1, 3)
idx_a = np.where(estimate_b == -1)[0]
idx_b = np.where(estimate_b == +1)[0]
idx = np.arange(0, estimate_b.shape[0])
plt.plot(data_data[idx_a], np.zeros_like(idx_a), "c*")
plt.plot(data_data[idx_b], np.zeros_like(idx_b), "m.")
plt.yticks([])
plt.title("Estimate B")
plt.title("Estimate")
plt.xlabel("Data Value")
plt.show()
```
Output:
```python
We will use: Estimate B
Performance: 77.3% correct
Performance: 77.31% correct
```
![Image3](image3.png)