i just learn data science from DataCamp about 2 months
it's my 1st step inside the data world
Environment
create a custom dockerfile from gw000/keras and modify it to add packages
# Dockerfile
FROM gw000/keras:2.0.5-py3-tf-cpu
# install dependencies from debian packages
RUN apt-get update -qq \
&& apt-get install --no-install-recommends -y \
python-matplotlib \
python-pillow
# install dependencies from python packages
RUN pip3 --no-cache-dir install \
pandas \
scikit-learn \
statsmodels
build it with tag name gw000/keras:2.0.5-py3-tf-cpu-datascience
$ docker build -t gw000/keras:2.0.5-py3-tf-cpu-datascience - < Dockerfile
Neural Network
download data from https://www.kaggle.com/c/digit-recognizer/data
create a sample code file called nn.py
model
- Rectified linear unit (ReLU) activation function for the first 3 layers
- Softmax activation function for the output layer
- Stochastic Optimization - Adam for optimizer
- categorical_crossentropy loss function for classification
# nn.py
import numpy as np
import pandas as pd
import keras
from keras.utils import to_categorical
from keras.models import Sequential
from keras.layers import Dense
# load train data
df = pd.read_csv('train.csv')
X = df.drop(['label'], axis=1).as_matrix()
y = to_categorical(df['label'])
# load test data
X_test = pd.read_csv('test.csv').as_matrix()
model = Sequential()
# hidden layer
model.add(Dense(25, activation='relu', input_shape=(X.shape[1],)))
model.add(Dense(25, activation='relu'))
model.add(Dense(25, activation='relu'))
# output layer
model.add(Dense(10, activation='softmax'))
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(X, y, validation_split=0.3)
predict = np.argmax(model.predict(X_test), axis=1)
# write predict data to a csv file
result = pd.DataFrame({'ImageId': np.arange(1, predict.shape[0]+1), 'Label': predict})
result.to_csv('submission.csv', index=False)
run with docker
# run a keras container to execute bash inside
$ docker run -it --rm -v <volume-folder>:/srv gw000/keras:2.0.5-py3-tf-cpu-datascience bash
# execute nn.py in the volume folder of container
$ python3 nn.py
Using TensorFlow backend.
Train on 29399 samples, validate on 12601 samples
Epoch 1/10
2018-01-03 14:42:29.906464: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2018-01-03 14:42:29.908490: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2018-01-03 14:42:29.908599: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-01-03 14:42:29.908667: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-01-03 14:42:29.908725: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
29399/29399 [==============================] - 28s - loss: 7.4748 - acc: 0.5132 - val_loss: 5.7031 - val_acc: 0.6215
Epoch 2/10
29399/29399 [==============================] - 10s - loss: 5.1255 - acc: 0.6508 - val_loss: 3.9194 - val_acc: 0.6898
Epoch 3/10
29399/29399 [==============================] - 9s - loss: 1.4371 - acc: 0.8121 - val_loss: 0.4783 - val_acc: 0.8856
Epoch 4/10
...
Rank
go to https://www.kaggle.com/c/digit-recognizer/submit
upload your submission csv file
@2018.01.03
it is a very simple example so the rank is not good!
you can improve it by CNN - Introduction to CNN Keras - 0.997 (top 6%)