-
Notifications
You must be signed in to change notification settings - Fork 0
/
tackling_mnsit_dataset.py
281 lines (187 loc) · 57.5 KB
/
tackling_mnsit_dataset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
# -*- coding: utf-8 -*-
"""Tackling MNSIT Dataset.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1idl0Mo1PlX9p3kfJsOrQMGkc_jWUrM5u
## Handwritten Digit Recogonition (MNSIT Dataset)
**History of Handwritten Digit dataset**
Modified National Institute of Standards and Technology database (MNIST dataset) is a large dataset of handwritten digits which is widely used in image processing and machine learning. The set of images in the MNIST database is a combination of two of NIST's databases: Special Database 1 and Special Database 3. Special Database 1 and Special Database 3 consist of digits written by high school students and employees of the United States Census Bureau, respectively.
**Why to start with MNIST dataset?**
The MNIST dataset is a well-known dataset for image classification. Tensorflow and Keras also provide MNIST dataset directly through their APIs. For the learning purpose, the MNIST dataset is easy to use and experiment with different machine learning techniques.
**About the Notebook**
* In this notebook, I have covered the necessary steps to approach any Machine Learning Classification Problem.
* Included Image Visualization for better understanding.
* Quick Links to the functions I have used to explore it in depth.
* Basic techniques such as k-folds, using Callbacks, etc.
I have tried to make this notebook as simple as possible, along with covering the basic approach to tackle any classification task.
**Task**
The task is to classify the images in 10 class, i.e., [0-9], inclusively.
## Required Imports
**Imports:**
1. pandas : For handeling csv dataset
2. numpy : Support for Pandas and calculations
3. Matplotlib - For visualization (Plotting graphs)
4. keras - Prediction Models
5. tensorflow - Prediction Models
"""
# Commented out IPython magic to ensure Python compatibility.
import pandas as pd
import numpy as np
import tensorflow as tf
import keras
import os
import matplotlib.pyplot as plt
import pickle
# %matplotlib inline
"""## Loading and Visualizing Dataset
---
**About Dataset**
MNIST dataset has the following features:
* Dataset size 60,000 samples of handwritten images.
* The size of each image is 28x28 pixels.
* Each image has only 1 color channel, i.e., grayscale image.
* Each pixel has value in the range of [0,255] where 0 represents black, and 255 represents white.
* Each image has labeled from 0-9.
**Loading Training and testing Data**
"""
(X_train_full, y_train_full), (X_test, y_test) = keras.datasets.mnist.load_data()
print("The shape of training dataset -", X_train_full.shape)
print("Each pixel intensity is also represented as a byte(0-255) -", X_train_full.dtype)
"""Prepairing **Validation data** and **Normalizing the data**
**It requires a few steps:**
* Assuming the validation set size. I am taking it 10% of the training set.
* Splitting training set into a training set (90% original training set) and validation set (10% original training set) from the training dataset.
* scaling down the pixel intensities down to the 0-1 range and convert them to floats, by dividing by 255
* Splitting the labels for both training set and validation set.
"""
X_valid, X_train = X_train_full[:5000] / 255., X_train_full[5000:] / 255.
y_valid, y_train = y_train_full[:5000], y_train_full[5000:]
X_test = X_test / 255.
"""Shape of training set"""
X_train.shape, y_train.shape
"""## Visualize Digits dataset
---
**Visualizing the digits by plotting Images**
This will plot the first 40 images of diits with the label.
"""
n_rows = 4
n_cols = 10
plt.figure(figsize=(n_cols * 1.2, n_rows * 1.2))
for row in range(n_rows):
for col in range(n_cols):
index = n_cols * row + col
plt.subplot(n_rows, n_cols, index + 1)
plt.imshow(X_train[index], cmap="binary", interpolation="nearest")
plt.axis("off")
plt.title(y_train[index], fontsize=12)
plt.subplots_adjust(wspace=0.2, hspace=0.5)
plt.show()
"""**Visualising one digit**
---
This will plot an image of the first digit in the dataset.
"""
plt.imshow(X_train[0], cmap="binary")
plt.axis("off")
plt.show()
print(y_train[0])
"""## **Building Model**
---
**Model Using Keras**
There are two different ways of defining the Model in Keras:
* Sequential Model
* Function API
Functional API is used to build a more complicated Model such as for multi-output Models, directed acyclic graphs, or models with shared layers. I am using the Sequential Model in this notebook to keep things simple.
In Sequential Model, you can add each layer sequentially.
Let's build a simple dense network and find the optimal learning rate. We will need a callback to grow the learning rate at each iteration. It will also record the learning rate and the loss at each iteration:
"""
K = keras.backend
class ExponentialLearningRate(keras.callbacks.Callback):
def __init__(self, factor):
self.factor = factor
self.rates = []
self.losses = []
def on_batch_end(self, batch, logs):
self.rates.append(K.get_value(self.model.optimizer.lr))
self.losses.append(logs["loss"])
K.set_value(self.model.optimizer.lr, self.model.optimizer.lr * self.factor)
keras.backend.clear_session()
np.random.seed(42)
tf.random.set_seed(42)
model = keras.models.Sequential([
keras.layers.Flatten(input_shape=[28, 28]),
keras.layers.Dense(300, activation="relu"),
keras.layers.Dense(100, activation="relu"),
keras.layers.Dense(10, activation="softmax")
])
"""We will start with a small learning rate of 1e-3, and grow it by 0.5% at each iteration:"""
model.compile(loss="sparse_categorical_crossentropy",
optimizer=keras.optimizers.SGD(lr=1e-3),
metrics=["accuracy"])
expon_lr = ExponentialLearningRate(factor=1.005) # increasing by 0.5% factor at each iterton
"""Now let's train the model for just 1 epoch:"""
history = model.fit(X_train, y_train, epochs=1,
validation_data=(X_valid, y_valid),
callbacks=[expon_lr])
"""We can now plot the loss as a functionof the learning rate:"""
plt.plot(expon_lr.rates, expon_lr.losses)
plt.gca().set_xscale('log')
plt.hlines(min(expon_lr.losses), min(expon_lr.rates), max(expon_lr.rates))
plt.axis([min(expon_lr.rates), max(expon_lr.rates), 0, expon_lr.losses[0]])
plt.xlabel("Learning rate")
plt.ylabel("Loss")
"""The loss starts shooting back up violently around 3e-1, so let's try using 2e-1 as our learning rate:"""
keras.backend.clear_session()
np.random.seed(42)
tf.random.set_seed(42)
model = keras.models.Sequential([
keras.layers.Flatten(input_shape=[28, 28]),
keras.layers.Dense(300, activation="relu"),
keras.layers.Dense(100, activation="relu"),
keras.layers.Dense(10, activation="softmax")
])
model.compile(loss="sparse_categorical_crossentropy",
optimizer=keras.optimizers.SGD(lr=2e-1),
metrics=["accuracy"])
run_index = 1 # increment this at every run
run_logdir = os.path.join(os.curdir, "my_mnsit_logs", "run_{:03d}".format(run_index))
run_logdir
early_stopping_cb = keras.callbacks.EarlyStopping(patience=20)
checkpoint_cb = keras.callbacks.ModelCheckpoint("my_mnsit_model.h5", save_best_only=True)
tenorboard_cb = keras.callbacks.TensorBoard(run_logdir)
history = model.fit(X_train, y_train, epochs=100,
validation_data=(X_valid, y_valid),
callbacks=[early_stopping_cb, checkpoint_cb, tenorboard_cb])
"""See, our model compilation stopped at around 23-29 epochs that means our best model is somewhere here and there"""
model = keras.models.load_model("my_mnsit_model.h5") # rolling back to best model
model.evaluate(X_test, y_test)
"""We got over 97% accuracy. Finally 😃"""
valid_p = np.argmax(model.predict(X_valid), axis=1)
"""## Visualizing Result
---
#### All Errors in the Validation set
Let's see all the errors in the validation set. It seems that in most of the cases, the recognition of digits is difficult for even humans. So we can say that our model is performing well.
"""
rows = 10
cols = 10
f = plt.figure(figsize=(2*cols,2*rows))
sub_plot = 1
for i in range(X_valid.shape[0]):
if y_valid[i]!=valid_p[i]:
f.add_subplot(rows,cols,sub_plot)
sub_plot+=1
plt.imshow(X_valid[i].reshape([28,28]),cmap="binary", interpolation="nearest")
plt.axis("off")
plt.title("T: "+str(y_valid[i])+" P:"+str(valid_p[i]), y=-0.15,color="Red")
plt.show()
"""## Predict on Testset"""
y_test = np.argmax(model.predict(X_test), axis=1)
rows = 5
cols = 10
f = plt.figure(figsize=(2*cols,2*rows))
for i in range(rows*cols):
f.add_subplot(rows,cols,i+1)
plt.imshow(X_test[i].reshape([28,28]),cmap="binary")
plt.axis("off")
plt.title(str(y_test[i]))