#!/usr/bin/env python
# coding: utf-8

# # Dogs and Cats

# In[1]:


import conx as cx
import os
import glob
from tqdm import tqdm


# First, you'll need to have an account at Kaggle.com. Got over there and log in.
# 
# Next, let's get the data. Go to https://www.kaggle.com/c/dogs-vs-cats/data and download the test1.zip and train.zip files into this folder.

# In[2]:


cx.download("https://www.kaggle.com/c/3362/download/test1.zip", verbose=0)


# In[3]:


cx.download("https://www.kaggle.com/c/3362/download/train.zip", verbose=0)


# In[4]:


get_ipython().run_cell_magic('time', '', 'if not os.path.exists("test_data.h5"):\n    def read_data(directory):\n        for filename in tqdm(glob.glob(directory)):\n            target = [0,0] # we don\'t know what these are\n            yield cx.image_to_array(filename, (64,64), raw=True)\n    cx.save_data("test_data.h5", read_data("./test1/*.jpg"))\ntest_data = cx.load_data("test_data.h5", return_h5=True)\n')


# In[5]:


get_ipython().run_cell_magic('time', '', 'if True or not os.path.exists("train_data.h5"):\n    targets = []\n    def read_data(directory):\n        for filename in tqdm(glob.glob(directory)):\n            yield [0,1] if "dog" in filename else [1, 0] # cat\n            #yield cx.image_to_array(filename, (64,64), raw=True)\n    #cx.save_data("train_data.h5", read_data("./train/*.jpg"))\n    targets = [x for x in read_data("./train/*.jpg")]\ntrain_data = cx.load_data("train_data.h5", return_h5=True)\n')


# In[6]:


cx.shape(train_data["data"]), cx.shape(test_data["data"]), cx.shape(targets)


# In[7]:


net = cx.Network("Dogs and Cats")
net.add(
    cx.ImageLayer("input", (64,64), 3),
    cx.Convolution2DLayer("conv", 16, (4,4)),
    cx.FlattenLayer("flatten"),
    cx.Layer("hidden", 32, activation="relu"),
    cx.Layer("output", 2, activation="softmax")
)
net.connect()
net.compile(error="binary_crossentropy", optimizer="rmsprop")


# In[8]:


net.dataset._targets = [cx.np.array(targets)]
net.dataset._inputs = [train_data]
net.dataset._labels = [None] * 25000
net.dataset._cache_values()


# In[9]:


cx.shape(net.dataset.inputs[0])


# In[10]:


net.dataset.targets[0]


# In[11]:


net.dataset.set_labels_from_function(lambda i: "Cat" if net.dataset.targets[i] == [1, 0] else "Dog")


# In[12]:


net.dashboard()


# In[ ]:


net.train(100, accuracy=.98)


# In[11]:


net.evaluate(select=100)


# In[ ]:


data = net.evaluate_and_label()


# In[14]:


[x[0] for x in data]


# In[12]:


import pandas as pd


# In[23]:


df = pd.DataFrame(data=[net.dataset._labels])


# In[25]:


get_ipython().run_line_magic('pinfo', 'net.dataset.set_inputs_from_targets')


# In[2]:


mnist_hp = cx.Dataset.get("mnist_h5")


# In[3]:


mnist_hp.save_to_disk("MNIST.conx")


# In[2]:


ds = cx.Dataset()

ds.load_from_disk("MNIST.conx")


# In[4]:


ds._inputs[0].dtype


# In[5]:


net


# In[ ]: