add: py2, py3 support

1. setup tox

evaluation.py

@@ -22,10 +22,10 @@ def evaluate(prediction, true_labels):
 
     Parameters
     ----------
-    prediction : 2-d array, shape (n_sample, n_classes)
+    prediction : 2-D array, shape (n_sample, n_classes)
         Onehot encoded predicted array
 
-    true_labels : 2-d array, shape (n_sample, n_classes)
+    true_labels : 2-D array, shape (n_sample, n_classes)
         Onehot encoded true array
 
     Returns
@@ -53,13 +53,13 @@ def main():
 
         # Check a single model accuracy
         acc = evaluate(pred, y_test)
-        print(f"Model-{idx}: {acc:>.5%}")
+        print("Model-{}: {:>.5%}".format(idx, acc))
 
     pred_list = np.asarray(pred_list)
     pred_mean = np.mean(pred_list, 0)
 
     accuracy = evaluate(pred_mean, y_test)
-    print(f"Final Test Accuracy: {accuracy:>.5%}")
+    print("Final Test Accuracy: {:>.5%}".format(accuracy))
 
 
 if __name__ == '__main__':

model.py

@@ -1,12 +1,12 @@
 import os
 
 from keras.layers import Input
-from keras.models import Model, load_model
+from keras.models import Model
 from keras.callbacks import ModelCheckpoint
 from utils import train_generator
 
 
-class BaseModel:
+class BaseModel(object):
     """Base Model Interface
 
     Methods

requirements.txt

@@ -0,0 +1,4 @@
+Keras==2.0.4
+numpy==1.12.1
+tensorflow-gpu==1.1.0
+h5py

resnet.py

@@ -24,7 +24,7 @@ def resnet(input_tensor):
     y : softmax output
     """
     def name_builder(type, stage, block, name):
-        return f"{type}{stage}{block}_branch{name}"
+        return "{}{}{}_branch{}".format(type, stage, block, name)
 
     def identity_block(input_tensor, kernel_size, filters, stage, block):
         F1, F2, F3 = filters

setup.py

@@ -0,0 +1,9 @@
+#!/usr/bin/env python
+
+from distutils.core import setup
+
+setup(name="mnist-competition",
+      version=1.0,
+      description="mnist competition file",
+      )
+

tox.ini

@@ -0,0 +1,8 @@
+[tox]
+envlist = py2,py3,py35,py36
+
+[testenv]
+deps = -r{toxinidir}/requirements.txt
+commands = python evaluation.py
+usedevelop = True
+

utils.py

@@ -88,7 +88,7 @@ def train_generator():
     return train_gen, val_gen
 
 
-def plot_(model_path: str, file_path: str):
+def plot_(model_path, file_path):
     """Visualize a model
 
     Parameters

vgg16.py

@@ -23,27 +23,27 @@ def vgg(input_tensor):
     y : softmax output tensor
     """
     def two_conv_pool(x, F1, F2, name):
-        x = Conv2D(F1, (3, 3), activation=None, padding='same', name=f'{name}_conv1')(x)
+        x = Conv2D(F1, (3, 3), activation=None, padding='same', name='{}_conv1'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F2, (3, 3), activation=None, padding='same', name=f'{name}_conv2')(x)
+        x = Conv2D(F2, (3, 3), activation=None, padding='same', name='{}_conv2'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = MaxPooling2D((2, 2), strides=(2, 2), name=f'{name}_pool')(x)
+        x = MaxPooling2D((2, 2), strides=(2, 2), name='{}_pool'.format(name))(x)
 
         return x
 
     def three_conv_pool(x, F1, F2, F3, name):
-        x = Conv2D(F1, (3, 3), activation=None, padding='same', name=f'{name}_conv1')(x)
+        x = Conv2D(F1, (3, 3), activation=None, padding='same', name='{}_conv1'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F2, (3, 3), activation=None, padding='same', name=f'{name}_conv2')(x)
+        x = Conv2D(F2, (3, 3), activation=None, padding='same', name='{}_conv2'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F3, (3, 3), activation=None, padding='same', name=f'{name}_conv3')(x)
+        x = Conv2D(F3, (3, 3), activation=None, padding='same', name='{}_conv3'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = MaxPooling2D((2, 2), strides=(2, 2), name=f'{name}_pool')(x)
+        x = MaxPooling2D((2, 2), strides=(2, 2), name='{}_pool'.format(name))(x)
 
         return x
 

vgg5.py

@@ -23,27 +23,27 @@ def vgg(input_tensor):
     y : softmax output tensor
     """
     def two_conv_pool(x, F1, F2, name):
-        x = Conv2D(F1, (3, 3), activation=None, padding='same', name=f'{name}_conv1')(x)
+        x = Conv2D(F1, (3, 3), activation=None, padding='same', name='{}_conv1'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F2, (3, 3), activation=None, padding='same', name=f'{name}_conv2')(x)
+        x = Conv2D(F2, (3, 3), activation=None, padding='same', name='{}_conv2'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = MaxPooling2D((2, 2), strides=(2, 2), name=f'{name}_pool')(x)
+        x = MaxPooling2D((2, 2), strides=(2, 2), name='{}_pool'.format(name))(x)
 
         return x
 
     def three_conv_pool(x, F1, F2, F3, name):
-        x = Conv2D(F1, (3, 3), activation=None, padding='same', name=f'{name}_conv1')(x)
+        x = Conv2D(F1, (3, 3), activation=None, padding='same', name='{}_conv1'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F2, (3, 3), activation=None, padding='same', name=f'{name}_conv2')(x)
+        x = Conv2D(F2, (3, 3), activation=None, padding='same', name='{}_conv2'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = Conv2D(F3, (3, 3), activation=None, padding='same', name=f'{name}_conv3')(x)
+        x = Conv2D(F3, (3, 3), activation=None, padding='same', name='{}_conv3'.format(name))(x)
         x = BatchNormalization()(x)
         x = Activation('relu')(x)
-        x = MaxPooling2D((2, 2), strides=(2, 2), name=f'{name}_pool')(x)
+        x = MaxPooling2D((2, 2), strides=(2, 2), name='{}_pool'.format(name))(x)
 
         return x