Feature/add scratch demo

demo/mxnet/mxnet_demo.py

@@ -1,5 +1,6 @@
-import mxnet as mx
+import logging
 
+import mxnet as mx
 # Here we import LogWriter so that we can write log data while MXNet is training
 from visualdl import LogWriter
 
@@ -45,7 +46,6 @@ def _callback(param):
 # Start to build CNN in MXNet, train MNIST dataset. For more info, check MXNet's official website:
 # https://mxnet.incubator.apache.org/tutorials/python/mnist.html
 
-import logging
 logging.getLogger().setLevel(logging.DEBUG)  # logging to stdout
 
 train_iter = mx.io.NDArrayIter(mnist['train_data'], mnist['train_label'], batch_size, shuffle=True)

demo/paddle/cifar10_image_classification_vgg.py

@@ -2,13 +2,14 @@
 
 import sys
 
+import numpy as np
+from visualdl import LogWriter
+
 import paddle.v2 as paddle
 import paddle.v2.fluid as fluid
 import paddle.v2.fluid.framework as framework
-from paddle.v2.fluid.param_attr import ParamAttr
 from paddle.v2.fluid.initializer import NormalInitializer
-from visualdl import LogWriter
-import numpy as np
+from paddle.v2.fluid.param_attr import ParamAttr
 
 logdir = "./tmp"
 logwriter = LogWriter(logdir, sync_cycle=10)

demo/vdl_scratch.py

@@ -1,11 +1,13 @@
 #!/user/bin/env python
+import math
 import os
-from visualdl import LogWriter, ROOT
+import random
 import subprocess
-from scipy.stats import norm
+
 import numpy as np
-import random
 from PIL import Image
+from scipy.stats import norm
+from visualdl import ROOT, LogWriter
 
 logdir = './scratch_log'
 
@@ -19,22 +21,29 @@
     scalar1 = logger.scalar("scratch/scalar")
 
 # add scalar records.
-for step in range(200):
-    scalar0.add_record(step, step * 1. / 200)
-    scalar1.add_record(step, 1. - step * 1. / 200)
+last_record0 = 0.
+last_record1 = 0.
+for step in range(1, 100):
+    last_record0 += 0.1 * (random.random() - 0.3)
+    last_record1 += 0.1 * (random.random() - 0.7)
+    scalar0.add_record(step, last_record0)
+    scalar1.add_record(step, last_record1)
 
 # create histogram
 with logw.mode('train') as logger:
-    histogram = logger.histogram("scratch/histogram", num_buckets=100)
-    for step in range(100):
+    histogram = logger.histogram("scratch/histogram", num_buckets=200)
+    for step in range(1, 100):
         histogram.add_record(step,
-                             np.random.normal(0.1 + step * 0.01, size=1000))
-
+                             np.random.normal(
+                                 0.1 + step * 0.001,
+                                 200. / (100 + step),
+                                 size=1000))
 # create image
 with logw.mode("train") as logger:
     image = logger.image("scratch/dog", 4,
                          1)  # randomly sample 4 images one pass
     dog_jpg = Image.open(os.path.join(ROOT, 'python/dog.jpg'))
+    dog_jpg = dog_jpg.resize(np.array(dog_jpg.size) / 2)
     shape = [dog_jpg.size[1], dog_jpg.size[0], 3]
 
     for pass_ in xrange(4):
@@ -47,12 +56,13 @@
             right_x = left_x + target_shape[1]
             right_y = left_y + target_shape[0]
 
-            idx = image.is_sample_taken()
-            # a more efficient way to sample images is
+            # a more efficient way to sample images
+            idx = image.is_sample_taken() # check whether this image will be taken by reservoir sampling
             if idx >= 0:
                 data = np.array(
                     dog_jpg.crop((left_x, left_y, right_x,
                                   right_y))).flatten()
+                # add this image to log
                 image.set_sample(idx, target_shape, data)
             # you can also just write followig codes, it is more clear, but need to
             # process image even if it will not be sampled.