Real world exmpl

In [1]:

from river.datasets import Insects
from river.preprocessing import MinMaxScaler 
from river.metrics import Accuracy
from river.utils import Rolling
from river.tree import HoeffdingTreeClassifier
from deep_river.classification import Classifier
from torch import nn 
from tqdm import tqdm 
import matplotlib.pyplot as plt
import matplotlib
import torch

from river.datasets import Insects from river.preprocessing import MinMaxScaler from river.metrics import Accuracy from river.utils import Rolling from river.tree import HoeffdingTreeClassifier from deep_river.classification import Classifier from torch import nn from tqdm import tqdm import matplotlib.pyplot as plt import matplotlib import torch

In [2]:

insects = Insects("abrupt_balanced")

def test_train_eval(model, stream, update_interval=100):
    results = []
    steps = []
    step = 0
    metric = Rolling(Accuracy(), window_size=400)
    scaler = MinMaxScaler()
    for x, y in tqdm(list(stream)):
        x = scaler.learn_one(x).transform_one(x)
        y_pred = model.predict_one(x)
        model.learn_one(x, y)
        metric.update(y, y_pred)
        step += 1
        if step % update_interval == 0:
            results.append(metric.get())
            steps.append(step)
    return steps, results

insects = Insects("abrupt_balanced") def test_train_eval(model, stream, update_interval=100): results = [] steps = [] step = 0 metric = Rolling(Accuracy(), window_size=400) scaler = MinMaxScaler() for x, y in tqdm(list(stream)): x = scaler.learn_one(x).transform_one(x) y_pred = model.predict_one(x) model.learn_one(x, y) metric.update(y, y_pred) step += 1 if step % update_interval == 0: results.append(metric.get()) steps.append(step) return steps, results

In [3]:

class SimpleMLP(nn.Module):
    def __init__(self, n_features):
        super().__init__()
        self.hidden1 = nn.Linear(n_features, 30)
        self.hidden2 = nn.Linear(30, 30)
        self.logits = nn.Linear(30, 6)

    def forward(self, x):
        h = torch.relu(self.hidden1(x))
        h = torch.relu(self.hidden2(h))
        return self.logits(h)


mlp = Classifier(
    SimpleMLP,
    loss_fn="binary_cross_entropy_with_logits",
    optimizer_fn="sgd",
    lr=0.05,
    seed=42,
)
steps, results_mlp = test_train_eval(mlp, insects)

class SimpleMLP(nn.Module): def __init__(self, n_features): super().__init__() self.hidden1 = nn.Linear(n_features, 30) self.hidden2 = nn.Linear(30, 30) self.logits = nn.Linear(30, 6) def forward(self, x): h = torch.relu(self.hidden1(x)) h = torch.relu(self.hidden2(h)) return self.logits(h) mlp = Classifier( SimpleMLP, loss_fn="binary_cross_entropy_with_logits", optimizer_fn="sgd", lr=0.05, seed=42, ) steps, results_mlp = test_train_eval(mlp, insects)

100%|██████████| 52848/52848 [00:41<00:00, 1270.52it/s]

In [4]:

tree = HoeffdingTreeClassifier()
steps, results_tree = test_train_eval(tree, insects)

tree = HoeffdingTreeClassifier() steps, results_tree = test_train_eval(tree, insects)

100%|██████████| 52848/52848 [00:46<00:00, 1142.66it/s]

In [5]:

fig, ax = plt.subplots(figsize=(8, 4))
ax.plot(steps, results_mlp, label="MLP")
ax.plot(steps, results_tree, label="Hoeffding Tree")

change_points = [0, 14352, 19500, 33240, 38682, 39510, 52848]
temps = [30, 20, 36, 26, 34, 30]
cmap = plt.cm.plasma
norm = matplotlib.colors.Normalize(vmin=20, vmax=36)
sm = plt.cm.ScalarMappable(cmap=cmap, norm=norm)
for i in range(len(change_points)-1):
    ax.axvspan(change_points[i], change_points[i+1], alpha=0.4, color=cmap(norm(temps[i])), lw=0)
ax.set_xlim(0, change_points[-1])
cbar = fig.colorbar(sm, ax=ax)
cbar.ax.set_ylabel("Temperature in °C")
ax.set_xlabel("Steps")
ax.set_ylabel("Moving Accuracy")
ax.legend()

fig, ax = plt.subplots(figsize=(8, 4)) ax.plot(steps, results_mlp, label="MLP") ax.plot(steps, results_tree, label="Hoeffding Tree") change_points = [0, 14352, 19500, 33240, 38682, 39510, 52848] temps = [30, 20, 36, 26, 34, 30] cmap = plt.cm.plasma norm = matplotlib.colors.Normalize(vmin=20, vmax=36) sm = plt.cm.ScalarMappable(cmap=cmap, norm=norm) for i in range(len(change_points)-1): ax.axvspan(change_points[i], change_points[i+1], alpha=0.4, color=cmap(norm(temps[i])), lw=0) ax.set_xlim(0, change_points[-1]) cbar = fig.colorbar(sm, ax=ax) cbar.ax.set_ylabel("Temperature in °C") ax.set_xlabel("Steps") ax.set_ylabel("Moving Accuracy") ax.legend()

Out[5]:

<matplotlib.legend.Legend at 0x142fbceb0>