In [1]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn import *
import scipy
%matplotlib inline
Datasource: http://yann.lecun.com/exdb/mnist/
The training dataset consists of 60,000 training digits and the test set contains 10,000 samples, respectively. The images in the MNIST dataset consist of pixels, and each pixel is represented by a gray scale intensity value. Here, we unroll the pixels into 1D row vectors, which represent the rows in our image array (784 per row or image). The second array (labels) returned by the load_mnist function contains the corresponding target variable, the class labels (integers 0-9) of the handwritten digits.
Csv version of the files are available in the following links. CSV training set http://www.pjreddie.com/media/files/mnist_train.csv CSV test set http://www.pjreddie.com/media/files/mnist_test.csv
In [2]:
training = pd.read_csv("/data/MNIST/mnist_train.csv", header = None)
testing = pd.read_csv("/data/MNIST/mnist_test.csv", header = None)
In [3]:
X_train, y_train = training.iloc[:, 1:].values, training.iloc[:, 0].values
X_test, y_test = testing.iloc[:, 1:].values, testing.iloc[:, 0].values
In [4]:
print("Shape of X_train: ", X_train.shape, "shape of y_train: ", y_train.shape)
print("Shape of X_test: ", X_test.shape, "shape of y_test: ", y_test.shape)
In [5]:
label_counts = pd.DataFrame({
"train": pd.Series(y_train).value_counts().sort_index(),
"test": pd.Series(y_test).value_counts().sort_index()
})
(label_counts / label_counts.sum()).plot.bar()
plt.xlabel("Class")
plt.ylabel("Frequency (normed)")
Out[5]:
In [6]:
scipy.stats.chisquare(label_counts.train, label_counts.test)
Out[6]:
In [7]:
fig, axes = plt.subplots(5, 5, figsize = (15, 9))
for i, ax in enumerate(fig.axes):
img = X_train[i, :].reshape(28, 28)
ax.imshow(img, cmap = "Greys", interpolation="nearest")
ax.set_title("True: %i" % y_train[i])
plt.tight_layout()
In [8]:
fig, axes = plt.subplots(10, 5, figsize = (15, 20))
for i, ax in enumerate(fig.axes):
img = X_train[y_train == 7][i, :].reshape(28, 28)
ax.imshow(img, cmap = "Greys", interpolation="nearest")
plt.tight_layout()
In [9]:
scaler = preprocessing.StandardScaler()
X_train_std = scaler.fit_transform(X_train.astype(np.float64))
X_test_std = scaler.transform(X_test.astype(np.float64))
In [10]:
%%time
lr = linear_model.LogisticRegression()
lr.fit(X_train_std, y_train)
print("accuracy:", lr.score(X_test_std, y_test))
In [11]:
y_test_pred = lr.predict(X_test_std)
miss_indices = (y_test != y_test_pred)
misses = X_test[miss_indices]
print("No of miss: ", misses.shape[0])
fig, axes = plt.subplots(10, 5, figsize = (15, 20))
misses_actual = y_test[miss_indices]
misses_pred = y_test_pred[miss_indices]
for i, ax in enumerate(fig.axes):
img = misses[i].reshape(28, 28)
ax.imshow(img, cmap = "Greys", interpolation="nearest")
ax.set_title("A: %s, P: %d" % (misses_actual[i], misses_pred[i]))
plt.tight_layout()
In [12]:
inits = np.random.randn(10, 784)
inits = inits / np.std(inits, axis=1).reshape(10, -1)
In [13]:
%%time
est = linear_model.SGDClassifier(n_jobs=4, tol=1e-5, eta0 = 0.15,
learning_rate = "invscaling",
alpha = 0.01, max_iter= 100)
est.fit(X_train_std, y_train, inits)
print("accuracy", est.score(X_test_std, y_test), "iterations:", est.n_iter_)
In [14]:
fig, _ = plt.subplots(3, 4, figsize = (15, 10))
for i, ax in enumerate(fig.axes):
if i < est.coef_.shape[0]:
ax.imshow(est.coef_[i, :].reshape(28, 28), cmap = "bwr", interpolation="nearest")
else:
ax.remove()
In [15]:
pd.DataFrame(est.coef_[0, :].reshape(28, 28))
Out[15]:
In [ ]: