def train(self, x, y):
""" Constructs a decision tree classifier from data
Parameters
----------
x : numpy.array
An N by K numpy array (N is the number of instances, K is the
number of attributes)
y : numpy.array
An N-dimensional numpy array
Returns
-------
DecisionTreeClassifier
A copy of the DecisionTreeClassifier instance
"""
# Make sure that x and y have the same number of instances
assert x.shape[0] == len(y), \
"Training failed. x and y must have the same number of instances."
#######################################################################
# ** TASK 2.1: COMPLETE THIS METHOD **
#######################################################################
dataset = ds.ClassifierDataset()
dataset.initFromData(x, y)
rootSplitObject = dataset.getSplitObjectForRoot()
treeStats = ClassifierTreeStats()
self.classifierTree = ClassifierTree(dataset, rootSplitObject,
treeStats)
# set a flag so that we know that the classifier has been trained
self.is_trained = True
return self
import prune
import cProfile
repeats = 3
pathToSimple1 = './data/simple1.txt'
pathToSimple2 = './data/simple2.txt'
pathToTest = './data/test.txt'
pathToToy = './data/toy.txt'
pathToToy2 = './data/toy2.txt'
pathToFull = './data/train_full.txt'
pathToNoisy = './data/train_noisy.txt'
pathToSub = './data/train_sub.txt'
pathToValid = './data/validation.txt'
dataset = ds.ClassifierDataset()
dataset.initFromFile(pathToFull)
dtc = cs.DecisionTreeClassifier()
print("FULL")
for i in range(repeats):
cProfile.run('dtc.train(dataset.attrib, dataset.labels)', None, 'time')
dataset.initFromFile(pathToSub)
print("\n\n\n=====\n\n\n\n")
print("SUB")
for i in range(repeats):
cProfile.run('dtc.train(dataset.attrib, dataset.labels)', None, 'time')
dataset.initFromFile(pathToNoisy)
print("\n\n\n=====\n\n\n\n")
import matplotlib.pyplot as plt
import numpy as np
import torch
from torch import optim
import torch.nn.functional as F
from model import BYOL, ClassifierBYOL, ClassifierScratch
import dataset
from tqdm import tqdm
import glob
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
BATCH_SIZE = 16
EPOCHS = 50
# %%
train_data = torchvision.datasets.ImageFolder('data/label/train')
train_dataset = dataset.ClassifierDataset(train_data)
train_dataloader = DataLoader(train_dataset,
shuffle=True,
num_workers=NUM_WORKERS,
batch_size=BATCH_SIZE)
# %%
def acc_fn(logits, targets):
logits = torch.argmax(logits, dim=1).detach().cpu()
targets = targets.detach().cpu()
acc = torch.mean((logits == targets) * 1.0)
return acc.numpy()
def train_batch(batch, model, optimizer):
import numpy as np import dataset as ds from prune import Prune import classification as cs pathToSimple1 = './data/simple1.txt' pathToSimple2 = './data/simple2.txt' pathToTest = './data/test.txt' pathToToy = './data/toy.txt' pathToToy2 = './data/toy2.txt' pathToFull = './data/train_full.txt' pathToNoisy = './data/train_noisy.txt' pathToSub = './data/train_sub.txt' pathToValid = './data/validation.txt' pathToToyValid = "./data/toyvalid.txt" dataset = ds.ClassifierDataset() dataset.initFromFile(pathToNoisy) dtc = cs.DecisionTreeClassifier() dtc.train(dataset.attrib, dataset.labels) validationDataset = ds.ClassifierDataset() validationDataset.initFromFile(pathToValid) Prune(dtc, validationDataset.attrib, validationDataset.labels)