Newer
Older
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<img width=\"800px\" src=\"../fidle/img/00-Fidle-header-01.svg\"></img>\n",
"\n",
"# <!-- TITLE --> [PMNIST1] - Simple classification with DNN\n",
"<!-- DESC -->Example of classification with a fully connected neural network, using Pytorch\n",
"<!-- AUTHOR : Laurent Risser (CNRS/IMT) -->\n",
"\n",
"## Objectives :\n",
" - Recognizing handwritten numbers\n",
" - Understanding the principle of a classifier DNN network \n",
" - Implementation with PyTorch \n",
"\n",
"\n",
"The [MNIST dataset](http://yann.lecun.com/exdb/mnist/) (Modified National Institute of Standards and Technology) is a must for Deep Learning. \n",
"It consists of 60,000 small images of handwritten numbers for learning and 10,000 for testing.\n",
"\n",
"\n",
"## What we're going to do :\n",
"\n",
" - Retrieve data\n",
" - Preparing the data\n",
" - Create a model\n",
" - Train the model\n",
" - Evaluate the result\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 1 - Init python stuff"
]
},
{
"cell_type": "code",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"import torch.nn as nn\n",
"import torch.nn.functional as F\n",
"from torch.autograd import Variable\n",
"import torchvision #to get the MNIST dataset\n",
"\n",
"\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"import sys,os\n",
"\n",
"import fidle\n",
"from fidle_pwk_additional import convergence_history_CrossEntropyLoss\n",
"# Init Fidle environment\n",
"run_id, run_dir, datasets_dir = fidle.init('PMNIST1')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 2 - Retrieve data\n",
"MNIST is one of the most famous historic dataset. \n",
"Include in [torchvision datasets](https://pytorch.org/docs/stable/torchvision/datasets.html)"
]
},
{
"cell_type": "code",
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
"source": [
"#get and format the training set\n",
"mnist_trainset = torchvision.datasets.MNIST(root='./data', train=True, download=True, transform=None)\n",
"x_train=mnist_trainset.data.type(torch.DoubleTensor)\n",
"y_train=mnist_trainset.targets\n",
"\n",
"\n",
"#get and format the test set\n",
"mnist_testset = torchvision.datasets.MNIST(root='./data', train=False, download=True, transform=None)\n",
"x_test=mnist_testset.data.type(torch.DoubleTensor)\n",
"y_test=mnist_testset.targets\n",
"\n",
"#check data shape and format\n",
"print(\"Size of the train and test observations\")\n",
"print(\" -> x_train : \",x_train.shape)\n",
"print(\" -> y_train : \",y_train.shape)\n",
"print(\" -> x_test : \",x_test.shape)\n",
"print(\" -> y_test : \",y_test.shape)\n",
"\n",
"print(\"\\nRemark that we work with torch tensors and not numpy arrays:\")\n",
"print(\" -> x_train.dtype = \",x_train.dtype)\n",
"print(\" -> y_train.dtype = \",y_train.dtype)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 3 - Preparing the data"
]
},
{
"cell_type": "code",
"source": [
"print('Before normalization : Min={}, max={}'.format(x_train.min(),x_train.max()))\n",
"\n",
"xmax=x_train.max()\n",
"x_train = x_train / xmax\n",
"x_test = x_test / xmax\n",
"\n",
"print('After normalization : Min={}, max={}'.format(x_train.min(),x_train.max()))\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Have a look"
]
},
{
"cell_type": "code",
"source": [
"np_x_train=x_train.numpy().astype(np.float64)\n",
"np_y_train=y_train.numpy().astype(np.uint8)\n",
"\n",
"fidle.scrawler.images(np_x_train,np_y_train , [27], x_size=5,y_size=5, colorbar=True)\n",
"fidle.scrawler.images(np_x_train,np_y_train, range(5,41), columns=12)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 4 - Create model\n",
"About informations about : \n",
" - [Optimizer](https://pytorch.org/docs/stable/optim.html)\n",
" - [Basic neural-network blocks](https://pytorch.org/docs/stable/nn.html)\n",
" - [Loss](https://pytorch.org/docs/stable/nn.html#loss-functions)"
]
},
{
"cell_type": "code",
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
"metadata": {},
"outputs": [],
"source": [
"class MyModel(nn.Module):\n",
" \"\"\"\n",
" Basic fully connected neural-network\n",
" \"\"\"\n",
" def __init__(self):\n",
" hidden1 = 100\n",
" hidden2 = 100\n",
" super(MyModel, self).__init__()\n",
" self.hidden1 = nn.Linear(784, hidden1)\n",
" self.hidden2 = nn.Linear(hidden1, hidden2)\n",
" self.hidden3 = nn.Linear(hidden2, 10)\n",
"\n",
" def forward(self, x):\n",
" x = x.view(-1,784) #flatten the images before using fully-connected layers\n",
" x = self.hidden1(x)\n",
" x = F.relu(x)\n",
" x = self.hidden2(x)\n",
" x = F.relu(x)\n",
" x = self.hidden3(x)\n",
" x = F.softmax(x, dim=0)\n",
" return x\n",
"\n",
" \n",
" \n",
"model = MyModel()\n",
" \n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Step 5 - Train the model\n",
"\n",
"#### 5.1 - stochastic gradient descent strategy to fit the model\n"
]
},
{
"cell_type": "code",
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
"metadata": {},
"outputs": [],
"source": [
"def fit(model,X_train,Y_train,X_test,Y_test, EPOCHS = 5, BATCH_SIZE = 32):\n",
" \n",
" loss = nn.CrossEntropyLoss()\n",
" optimizer = torch.optim.Adam(model.parameters(),lr=1e-3) #lr is the learning rate\n",
" model.train()\n",
" \n",
" history=convergence_history_CrossEntropyLoss()\n",
" \n",
" history.update(model,X_train,Y_train,X_test,Y_test)\n",
" \n",
" n=X_train.shape[0] #number of observations in the training data\n",
" \n",
" #stochastic gradient descent\n",
" for epoch in range(EPOCHS):\n",
" \n",
" batch_start=0\n",
" epoch_shuffler=np.arange(n) \n",
" np.random.shuffle(epoch_shuffler) #remark that 'utilsData.DataLoader' could be used instead\n",
" \n",
" while batch_start+BATCH_SIZE < n:\n",
" #get mini-batch observation\n",
" mini_batch_observations = epoch_shuffler[batch_start:batch_start+BATCH_SIZE]\n",
" var_X_batch = Variable(X_train[mini_batch_observations,:,:]).float() #the input image is flattened\n",
" var_Y_batch = Variable(Y_train[mini_batch_observations])\n",
" \n",
" #gradient descent step\n",
" optimizer.zero_grad() #set the parameters gradients to 0\n",
" Y_pred_batch = model(var_X_batch) #predict y with the current NN parameters\n",
" \n",
" curr_loss = loss(Y_pred_batch, var_Y_batch) #compute the current loss\n",
" curr_loss.backward() #compute the loss gradient w.r.t. all NN parameters\n",
" optimizer.step() #update the NN parameters\n",
" \n",
" #prepare the next mini-batch of the epoch\n",
" batch_start+=BATCH_SIZE\n",
" \n",
" history.update(model,X_train,Y_train,X_test,Y_test)\n",
" \n",
" return history\n",
"\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"##### 5.2 - fit the model"
]
},
{
"cell_type": "code",
"metadata": {},
"outputs": [],
"source": [
"\n",
"batch_size = 512\n",
"epochs = 128\n",
"\n",
"\n",
"history=fit(model,x_train,y_train,x_test,y_test,EPOCHS=epochs,BATCH_SIZE = batch_size)\n",
"\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 6 - Evaluate\n",
"### 6.1 - Final loss and accuracy"
]
},
{
"cell_type": "code",
"source": [
"var_x_test = Variable(x_test[:,:,:]).float()\n",
"var_y_test = Variable(y_test[:])\n",
"y_pred = model(var_x_test)\n",
"\n",
"loss = nn.CrossEntropyLoss()\n",
"curr_loss = loss(y_pred, var_y_test)\n",
"\n",
"val_loss = curr_loss.item()\n",
"val_accuracy = float( (torch.argmax(y_pred, dim= 1) == var_y_test).float().mean() )\n",
"\n",
"\n",
"print('Test loss :', val_loss)\n",
"print('Test accuracy :', val_accuracy)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 6.2 - Plot history"
]
},
{
"cell_type": "code",
"fidle.scrawler.history(history, figsize=(6,4))\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 6.3 - Plot results"
]
},
{
"cell_type": "code",
"source": [
"y_pred = model(var_x_test)\n",
"np_y_pred_label = torch.argmax(y_pred, dim= 1).numpy().astype(np.uint8)\n",
"\n",
"np_x_test=x_test.numpy().astype(np.float64)\n",
"np_y_test=y_test.numpy().astype(np.uint8)\n",
"\n",
"fidle.scrawler.images(np_x_test, np_y_test, range(0,60), columns=12, x_size=1, y_size=1, y_pred=np_y_pred_label)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 6.4 - Plot some errors"
]
},
{
"cell_type": "code",
"source": [
"errors=[ i for i in range(len(np_y_test)) if np_y_pred_label[i]!=np_y_test[i] ]\n",
"errors=errors[:min(24,len(errors))]\n",
"fidle.scrawler.images(np_x_test, np_y_test, errors[:15], columns=6, x_size=2, y_size=2, y_pred=np_y_pred_label)\n"
"fidle.scrawler.confusion_matrix(np_y_test,np_y_pred_label, range(10))\n"
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<div class=\"todo\">\n",
" A few things you can do for fun:\n",
" <ul>\n",
" <li>Changing the network architecture (layers, number of neurons, etc.)</li>\n",
" <li>Display a summary of the network</li>\n",
" <li>Retrieve and display the softmax output of the network, to evaluate its \"doubts\".</li>\n",
" </ul>\n",
"</div>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"---\n",
"<img width=\"80px\" src=\"../fidle/img/00-Fidle-logo-01.svg\"></img>"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3.9.2 ('fidle-env')",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.2"
},
"vscode": {
"interpreter": {
"hash": "b3929042cc22c1274d74e3e946c52b845b57cb6d84f2d591ffe0519b38e4896d"
}
}
},
"nbformat": 4,
"nbformat_minor": 4
}