DiCE
DiCE copied to clipboard
interpretml
when using a trained pytorch model,the error:running_mean should contain xx elements not xx
I have already trained a pytorch model, the error occurs in the following sentence: exp = dice_ml.Dice(d, m)
the ERROR show as follows: RuntimeError: running_mean should contain 69 elements not 41 I have saw some explainations,saying that the solution is to align the output dimensions of BN layer and conv layer. But when I trained my model,it did not show this problem.My model trained successfully. the problem occurs when trying to use dice to generate a exp. So I don`t know how to solve it.
@gaugup
thx for the reply. Im new for DL,so maybe cant figure out the problem.
codes are as follows:
`
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
import dice_ml
from dice_ml.utils import helpers # helper functions
from torch.autograd import Variable
import torch.nn as nn
import torch
from sklearn.model_selection import train_test_split
from torch.autograd import Variable
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
from torch.utils.data import Dataset, DataLoader
hidden_layer = 100
output_data = 4
epoch_n = 500
learning_rate = 1e-4
batch_size = 50
class tabularDataset(Dataset):
def __init__(self, X, Y):
self.x = X.values
self.y = Y
def __len__(self):
return len(self.y)
def __getitem__(self, idx):
return (self.x[idx], self.y[idx])
class twolayer(nn.Module):
def __init__(self, input_data, hidden_layer, output_data):
super(twolayer, self).__init__()
self.linear1 = nn.Linear(input_data, hidden_layer)
self.ReLU = nn.ReLU()
self.linear2 = nn.Linear(hidden_layer, output_data)
self.bn_in = nn.BatchNorm1d(input_data)
self.bn1 = nn.BatchNorm1d(hidden_layer)
self.bn2 = nn.BatchNorm1d(output_data)
def forward(self, x):
x = self.bn_in(x)
a1 = self.linear1(x)
self.ReLU(a1)
a1 = self.bn1(a1)
y_pred = self.linear2(a1)
y_pred = self.bn2(y_pred)
return y_pred
def train(model, train_dl, optimizer, loss_fn):
model.train()
total_epochs = 5
losses = []
for epoch in range(total_epochs):
for i, (x, y) in enumerate(train_dl):
x-x.float().to(DEVICE)
y = y.long().to(DEVICE)
optimizer.zero_grad()
outputs = model(x.to(torch.float32))
loss = loss_fn(outputs, y)
loss.backward()
optimizer.step()
losses.append(loss.cpu().data.item())
print('Epoch : %d/%d, Loss: %.4f' %
(epoch+1, total_epochs, np.mean(losses)))
model.eval()
correct = 0
total = 0
for i, (x, y) in enumerate(train_dl):
x = x.float().to(DEVICE)
y = y.long()
outputs = model(x).cpu()
_, predicted = torch.max(outputs.data, 1)
total += y.size(0)
correct += (predicted == y).sum()
print('auc: %.4f %%' % (100 * correct / total))
if __name__ == '__main__':
dataset = pd.read_csv('./mykddcup.csv', names=['duration', 'protocol_type' , 'service', 'flag', 'src_bytes', 'dst_bytes' , 'land' , 'wrong_fragment' , 'urgent' , 'hot', 'numfailedlogins' , 'loggedin' , 'numcompromised', 'rootshell', 'suattempted' , 'numroot', 'numfilecreations', 'numshells' , 'numaccessfiles' , 'numoutboundcmds', 'ishotlogin' , 'isguestlogin', 'count', 'srv_count', 'serror_rate',
'srv_serror_rate', 'rerror_rate' , 'srv_rerror_rate', 'same_srv_rate' , 'diff_srv_rate' , 'srv_diff_host_rate', 'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate', 'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate', 'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate' , 'label'])
y = dataset['label']
print(torch.__version__)
for each in y:
if each == 1:
y[each] = 1
else:
y[each] = 0
y_label = LabelEncoder()
y = y_label.fit_transform(y)
print(y)
input_data = dataset.drop('label', axis=1).shape[1]
train_dataset, test_dataset, y_train, y_test = train_test_split(dataset,
y,
test_size=0.2,
random_state=0,
stratify=y)
train_ds = tabularDataset(train_dataset.drop('label', axis=1), y_train)
test_ds = tabularDataset(test_dataset, y_test)
d = dice_ml.Data(dataframe=train_dataset, continuous_features=['duration', 'src_bytes', 'dst_bytes', 'wrong_fragment', 'urgent', 'hot', 'numfailedlogins', 'numcompromised', 'numroot', 'numfilecreations', 'numshells', 'numaccessfiles', 'numoutboundcmds', 'count', 'srv_count', 'serror_rate', 'srv_serror_rate', 'rerror_rate', 'srv_rerror_rate',
'same_srv_rate', 'diff_srv_rate', 'srv_diff_host_rate', 'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate', 'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate', 'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate'], outcome_name='label')
DEVICE = torch.device("cpu")
if torch.cuda.is_available():
DEVICE = torch.device("cuda")
model = twolayer(input_data, hidden_layer, output_data).to(DEVICE)
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(
model.parameters(), lr=learning_rate)
train_dl = DataLoader(train_ds, batch_size=batch_size, shuffle=True)
train(model, train_dl, optimizer, loss_fn)
torch.save(model, './model.tar')
#!DICE
backend = 'PYT'
m = dice_ml.Model(model_path='./model.tar', backend=backend)
exp = dice_ml.Dice(d, m, method="random")
query_instance = {'duration': 0,
'dst_bytes': 21,
'wrong_fragment': 5,
'src_bytes': 0,
}
dice_exp = exp.generate_counterfactuals(
query_instance, total_CFs=4, desired_class="opposite")
dice_exp.visualize_as_dataframe(show_only_changes=True)
`
@gaugup
thx for the reply. Im new for DL,so maybe cant figure out the problem.
codes are as follows:
`
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
import dice_ml
from dice_ml.utils import helpers # helper functions
from torch.autograd import Variable
import torch.nn as nn
import torch
from sklearn.model_selection import train_test_split
from torch.autograd import Variable
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
from sklearn.preprocessing import LabelEncoder
from torch.utils.data import Dataset, DataLoader
hidden_layer = 100
output_data = 4
epoch_n = 500
learning_rate = 1e-4
batch_size = 50
class tabularDataset(Dataset):
def __init__(self, X, Y):
self.x = X.values
self.y = Y
def __len__(self):
return len(self.y)
def __getitem__(self, idx):
return (self.x[idx], self.y[idx])
class twolayer(nn.Module):
def __init__(self, input_data, hidden_layer, output_data):
super(twolayer, self).__init__()
self.linear1 = nn.Linear(input_data, hidden_layer)
self.ReLU = nn.ReLU()
self.linear2 = nn.Linear(hidden_layer, output_data)
self.bn_in = nn.BatchNorm1d(input_data)
self.bn1 = nn.BatchNorm1d(hidden_layer)
self.bn2 = nn.BatchNorm1d(output_data)
def forward(self, x):
x = self.bn_in(x)
a1 = self.linear1(x)
self.ReLU(a1)
a1 = self.bn1(a1)
y_pred = self.linear2(a1)
y_pred = self.bn2(y_pred)
return y_pred
def train(model, train_dl, optimizer, loss_fn):
model.train()
total_epochs = 5
losses = []
for epoch in range(total_epochs):
for i, (x, y) in enumerate(train_dl):
x-x.float().to(DEVICE)
y = y.long().to(DEVICE)
optimizer.zero_grad()
outputs = model(x.to(torch.float32))
loss = loss_fn(outputs, y)
loss.backward()
optimizer.step()
losses.append(loss.cpu().data.item())
print('Epoch : %d/%d, Loss: %.4f' %
(epoch+1, total_epochs, np.mean(losses)))
model.eval()
correct = 0
total = 0
for i, (x, y) in enumerate(train_dl):
x = x.float().to(DEVICE)
y = y.long()
outputs = model(x).cpu()
_, predicted = torch.max(outputs.data, 1)
total += y.size(0)
correct += (predicted == y).sum()
print('auc: %.4f %%' % (100 * correct / total))
if __name__ == '__main__':
dataset = pd.read_csv('./mykddcup.csv', names=['duration', 'protocol_type' , 'service', 'flag', 'src_bytes', 'dst_bytes' , 'land' , 'wrong_fragment' , 'urgent' , 'hot', 'numfailedlogins' , 'loggedin' , 'numcompromised', 'rootshell', 'suattempted' , 'numroot', 'numfilecreations', 'numshells' , 'numaccessfiles' , 'numoutboundcmds', 'ishotlogin' , 'isguestlogin', 'count', 'srv_count', 'serror_rate',
'srv_serror_rate', 'rerror_rate' , 'srv_rerror_rate', 'same_srv_rate' , 'diff_srv_rate' , 'srv_diff_host_rate', 'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate', 'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate', 'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate' , 'label'])
y = dataset['label']
print(torch.__version__)
for each in y:
if each == 1:
y[each] = 1
else:
y[each] = 0
y_label = LabelEncoder()
y = y_label.fit_transform(y)
print(y)
input_data = dataset.drop('label', axis=1).shape[1]
train_dataset, test_dataset, y_train, y_test = train_test_split(dataset,
y,
test_size=0.2,
random_state=0,
stratify=y)
train_ds = tabularDataset(train_dataset.drop('label', axis=1), y_train)
test_ds = tabularDataset(test_dataset, y_test)
d = dice_ml.Data(dataframe=train_dataset, continuous_features=['duration', 'src_bytes', 'dst_bytes', 'wrong_fragment', 'urgent', 'hot', 'numfailedlogins', 'numcompromised', 'numroot', 'numfilecreations', 'numshells', 'numaccessfiles', 'numoutboundcmds', 'count', 'srv_count', 'serror_rate', 'srv_serror_rate', 'rerror_rate', 'srv_rerror_rate',
'same_srv_rate', 'diff_srv_rate', 'srv_diff_host_rate', 'dst_host_count', 'dst_host_srv_count', 'dst_host_same_srv_rate', 'dst_host_diff_srv_rate', 'dst_host_same_src_port_rate', 'dst_host_srv_diff_host_rate', 'dst_host_serror_rate', 'dst_host_srv_serror_rate', 'dst_host_rerror_rate', 'dst_host_srv_rerror_rate'], outcome_name='label')
DEVICE = torch.device("cpu")
if torch.cuda.is_available():
DEVICE = torch.device("cuda")
model = twolayer(input_data, hidden_layer, output_data).to(DEVICE)
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(
model.parameters(), lr=learning_rate)
train_dl = DataLoader(train_ds, batch_size=batch_size, shuffle=True)
train(model, train_dl, optimizer, loss_fn)
torch.save(model, './model.tar')
#!DICE
backend = 'PYT'
m = dice_ml.Model(model_path='./model.tar', backend=backend)
exp = dice_ml.Dice(d, m, method="random")
query_instance = {'duration': 0,
'dst_bytes': 21,
'wrong_fragment': 5,
'src_bytes': 0,
}
dice_exp = exp.generate_counterfactuals(
query_instance, total_CFs=4, desired_class="opposite")
dice_exp.visualize_as_dataframe(show_only_changes=True)
`