Refer to the jupyter notebook for rendered code.
This is handy when you want to train multiple classifiers and store the results in a structured way. Similar to lists in R.
classifiers = {
'LR1': LogisticRegression(multi_class='multinomial', C=10, penalty='l1', solver='saga'),
'LR2': LogisticRegression(multi_class='multinomial', C=10, penalty='l2', solver='saga')
}Can check some information on the dictionary, with len(classificers) and .items(), .keys() and values().
To access the items inside a dictionary, use dict['key1'].
classifiers['LR1'].fit(Xtrain, ytrain)
classifiers['LR1'].predict(Xtest)Print out the information
for name, model in classifiers.items():
print(name)
print(model)Want to run the classifiers together: set up the result dictionary to hold the outputs of potentially different structures. Remember to save them into the right name!
results = {}
# name extracts the key (can be x, i, whatever)
# model refers to the content
for name, model in classifiers.items():
model.fit(Xtrain, y_train)
yhat = model.predict(Xtest)
accuracy = accuracy_score(y_test, yhat)
# save results. it has to be saved in the proper name
results[name] = {
'model': name,
'accuracy': accuracy,
'coefficients': model.coef_,
'intercept': model.intercept_,
'yhat': yhat
}Say that the results look like the following format.
import numpy as np
import pandas as pd
model1 = {
'accuracy': 0.95,
'yhat': np.array([1,3,5])
}
model2 = {
'accuracy': 0.97,
'yhat': np.array([1,4,6])
}
# put them together. remember dictionary must have a key, can not simply do {model1, model2}!
results = {
'model1': model1,
'model2': model2
}
results{'model1': {'accuracy': 0.95, 'yhat': array([1, 3, 5])},
'model2': {'accuracy': 0.97, 'yhat': array([1, 4, 6])}}Quick glance: results.items(), results.keys()
results.keys()dict_keys(['model1', 'model2'])results.items()dict_items([('model1', {'accuracy': 0.95, 'yhat': array([1, 3, 5])}), ('model2', {'accuracy': 0.97, 'yhat': array([1, 4, 6])})])Get all the results from the first key (‘model1’)
results['model1']{'accuracy': 0.95, 'yhat': array([1, 3, 5])}More specific, just get accuracy.
results['model1']['accuracy']0.95Here we need to use list comprehension and dictionary comprehension. Recall that with LC, the syntax goes like [<expression> for <item> in <iterable>].
# all the elements
[x for x in results.values()][{'accuracy': 0.95, 'yhat': array([1, 3, 5])},
{'accuracy': 0.97, 'yhat': array([1, 4, 6])}]Combine with dictionary syntax (the expression), to get model_x['accuracy'] for both models:
# certain item only
[x['accuracy'] for x in results.values()][0.95, 0.97]This can also be presented as a dataframe too.
accuracy = pd.DataFrame([x['accuracy'] for x in results.values()])
accuracy # can do accuracy.T to change the layout| 0 | |
|---|---|
| 0 | 0.95 |
| 1 | 0.97 |
Can do it to arrays too.
[x['yhat'] for x in results.values()][array([1, 3, 5]), array([1, 4, 6])]Do it with df.rename(columns = {'old':'new'})
accuracy.rename(columns = {0:'accuracies'})| accuracies | |
|---|---|
| 0 | 0.95 |
| 1 | 0.97 |
Dictionary comprehension is similar to LC, and it’s more handy when keys are involved for later. Note that here we are extracting from results.items() rather than values().
{key: {'accuracy': value['accuracy']} for key, value in results.items()}{'model1': {'accuracy': 0.95}, 'model2': {'accuracy': 0.97}}The results here can be put directly into a dataframe.
a = {key: {'accuracy': value['accuracy']} for key, value in results.items()}
pd.DataFrame(a)| model1 | model2 | |
|---|---|---|
| accuracy | 0.95 | 0.97 |
yhats = pd.DataFrame({key: {'yhat': value['yhat']} for key, value in results.items()})
yhats| model1 | model2 | |
|---|---|---|
| yhat | [1, 3, 5] | [1, 4, 6] |
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.datasets import make_blobs
# from code_py.sklearn_1 import Xtrain
X, y = make_blobs(n_samples=300,
centers=4,
random_state=0,
cluster_std=1)
# plot the two dimensions of X; color with the class in y
sns.scatterplot(x = X[:,0], y = X[:,1], hue=y, palette=sns.color_palette("hls", 4))
plt.show();
# split the data
from sklearn.model_selection import train_test_split
Xtrain, Xtest, ytrain, ytest = train_test_split(X, y, test_size=0.3,random_state=42)
# fit a decision tree
from sklearn.tree import DecisionTreeClassifier
tree = DecisionTreeClassifier().fit(Xtrain, ytrain)
# make prediction
ytest_pred = tree.predict(Xtest)
ytest_pred[0:5]array([2, 0, 0, 3, 1])Now we try to visualise the results. First put the predictions along with the original data, and then add a label for whether there is a mismatch.
# tt = pd.DataFrame(Xtest, columns=['x1', 'x2'])
# tt['new'] = ytest
mat = np.column_stack((Xtest, ytest, ytest_pred))
test_df = pd.DataFrame(mat, columns=['x1', 'x2', 'y', 'pred'])
# add a new column where y and pred do not match
test_df['mismatch'] = np.where(test_df['y'] != test_df['pred'], 1, 0)
test_df.head()| x1 | x2 | y | pred | mismatch | |
|---|---|---|---|---|---|
| 0 | -1.993750 | 1.500976 | 2.0 | 2.0 | 0 |
| 1 | 1.840706 | 3.561622 | 0.0 | 0.0 | 0 |
| 2 | -0.170058 | 5.276275 | 3.0 | 0.0 | 1 |
| 3 | -0.352996 | 9.210424 | 3.0 | 3.0 | 0 |
| 4 | 0.118988 | 1.086442 | 1.0 | 1.0 | 0 |
Now visualize
# visualize based on mismatch status
sns.relplot(data = test_df,
x = 'x1',
y = 'x2',
col = 'mismatch',
hue = 'y',
palette=sns.color_palette("hls", 4))
plt.show();
Another way to visualize
sns.scatterplot(data = test_df,
x = 'x1',
y = 'x2',
hue = 'y',
style = 'mismatch',
palette=sns.color_palette("hls", 4))
plt.show();