Skip to content

Instantly share code, notes, and snippets.

@koreyou

koreyou/chainer_reporting_demo.ipynb

Created June 26, 2017 14:29
Show Gist options
  • Save koreyou/dbb07498d0a634b229a0f29754368e8c to your computer and use it in GitHub Desktop.
Save koreyou/dbb07498d0a634b229a0f29754368e8c to your computer and use it in GitHub Desktop.
Download ZIP
Demostration of how we can use chainer's reporing mechanism to monitor variables' metrics
Display the source blob
Display the rendered blob
Raw
chainer_reporting_demo.ipynb
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Chainer Reporting Demo\n",
"\n",
"This snippets demostrate how we can use [chainer](https://chainer.org/)'s reporing mechanism to monitor variables' metrics.\n",
"Codes were taken from [\"the official example\"](https://github.com/chainer/chainer/blob/master/examples/mnist/train_mnist.py) with some modifications.\n"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"from __future__ import print_function\n",
"\n",
"import argparse\n",
"\n",
"import chainer\n",
"import chainer.functions as F\n",
"import chainer.links as L\n",
"from chainer import training\n",
"from chainer.training import extensions"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"# Network definition\n",
"class MLP(chainer.Chain):\n",
"\n",
" def __init__(self, n_units, n_out):\n",
" super(MLP, self).__init__()\n",
" with self.init_scope():\n",
" # the size of the inputs to each layer will be inferred\n",
" self.l1 = L.Linear(None, n_units) # n_in -> n_units\n",
" self.l2 = L.Linear(None, n_units) # n_units -> n_units\n",
" self.l3 = L.Linear(None, n_out) # n_units -> n_out\n",
"\n",
" def __call__(self, x):\n",
" h1 = F.relu(self.l1(x))\n",
" h2 = F.relu(self.l2(h1))\n",
" return self.l3(h2)"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"class DummyParser(object):\n",
" def __init__(self, **args):\n",
" self.__dict__ = dict(**args)\n",
"\n",
"\n",
"args = DummyParser(\n",
" batchsize=100,\n",
" epoch=3, # Default was 20; this is for testing\n",
" frequency=-1,\n",
" gpu=-1,\n",
" out='result',\n",
" resume='',\n",
" unit=200 # Default was1000; this is for testing\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# First experiment with original example implementation"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"epoch main/loss validation/main/loss main/accuracy validation/main/accuracy elapsed_time\n",
"\u001b[J total [##................................................] 5.56%\n",
"this epoch [########..........................................] 16.67%\n",
" 100 iter, 0 epoch / 3 epochs\n",
" inf iters/sec. Estimated time to finish: 0:00:00.\n",
"\u001b[4A\u001b[J total [#####.............................................] 11.11%\n",
"this epoch [################..................................] 33.33%\n",
" 200 iter, 0 epoch / 3 epochs\n",
" 104.22 iters/sec. Estimated time to finish: 0:00:15.352177.\n",
"\u001b[4A\u001b[J total [########..........................................] 16.67%\n",
"this epoch [#########################.........................] 50.00%\n",
" 300 iter, 0 epoch / 3 epochs\n",
" 84.481 iters/sec. Estimated time to finish: 0:00:17.755546.\n",
"\u001b[4A\u001b[J total [###########.......................................] 22.22%\n",
"this epoch [#################################.................] 66.67%\n",
" 400 iter, 0 epoch / 3 epochs\n",
" 83.855 iters/sec. Estimated time to finish: 0:00:16.695458.\n",
"\u001b[4A\u001b[J total [#############.....................................] 27.78%\n",
"this epoch [#########################################.........] 83.33%\n",
" 500 iter, 0 epoch / 3 epochs\n",
" 82.813 iters/sec. Estimated time to finish: 0:00:15.697926.\n",
"\u001b[4A\u001b[J1 0.278771 0.135821 0.92135 0.9598 6.9406 \n",
"\u001b[J total [################..................................] 33.33%\n",
"this epoch [..................................................] 0.00%\n",
" 600 iter, 1 epoch / 3 epochs\n",
" 76.09 iters/sec. Estimated time to finish: 0:00:15.770887.\n",
"\u001b[4A\u001b[J total [###################...............................] 38.89%\n",
"this epoch [########..........................................] 16.67%\n",
" 700 iter, 1 epoch / 3 epochs\n",
" 81.07 iters/sec. Estimated time to finish: 0:00:13.568603.\n",
"\u001b[4A\u001b[J total [######################............................] 44.44%\n",
"this epoch [################..................................] 33.33%\n",
" 800 iter, 1 epoch / 3 epochs\n",
" 80.339 iters/sec. Estimated time to finish: 0:00:12.447324.\n",
"\u001b[4A\u001b[J total [#########################.........................] 50.00%\n",
"this epoch [#########################.........................] 50.00%\n",
" 900 iter, 1 epoch / 3 epochs\n",
" 83.275 iters/sec. Estimated time to finish: 0:00:10.807516.\n",
"\u001b[4A\u001b[J total [###########################.......................] 55.56%\n",
"this epoch [#################################.................] 66.67%\n",
" 1000 iter, 1 epoch / 3 epochs\n",
" 82.692 iters/sec. Estimated time to finish: 0:00:09.674441.\n",
"\u001b[4A\u001b[J total [##############################....................] 61.11%\n",
"this epoch [#########################################.........] 83.33%\n",
" 1100 iter, 1 epoch / 3 epochs\n",
" 82.393 iters/sec. Estimated time to finish: 0:00:08.495911.\n",
"\u001b[4A\u001b[J2 0.104873 0.0970318 0.968783 0.969 14.3603 \n",
"\u001b[J total [#################################.................] 66.67%\n",
"this epoch [..................................................] 0.00%\n",
" 1200 iter, 2 epoch / 3 epochs\n",
" 78.769 iters/sec. Estimated time to finish: 0:00:07.617201.\n",
"\u001b[4A\u001b[J total [####################################..............] 72.22%\n",
"this epoch [########..........................................] 16.67%\n",
" 1300 iter, 2 epoch / 3 epochs\n",
" 79.376 iters/sec. Estimated time to finish: 0:00:06.299094.\n",
"\u001b[4A\u001b[J total [######################################............] 77.78%\n",
"this epoch [################..................................] 33.33%\n",
" 1400 iter, 2 epoch / 3 epochs\n",
" 78.333 iters/sec. Estimated time to finish: 0:00:05.106387.\n",
"\u001b[4A\u001b[J total [#########################################.........] 83.33%\n",
"this epoch [#########################.........................] 50.00%\n",
" 1500 iter, 2 epoch / 3 epochs\n",
" 77.904 iters/sec. Estimated time to finish: 0:00:03.850887.\n",
"\u001b[4A\u001b[J total [############################################......] 88.89%\n",
"this epoch [#################################.................] 66.67%\n",
" 1600 iter, 2 epoch / 3 epochs\n",
" 77.804 iters/sec. Estimated time to finish: 0:00:02.570575.\n",
"\u001b[4A\u001b[J total [###############################################...] 94.44%\n",
"this epoch [#########################################.........] 83.33%\n",
" 1700 iter, 2 epoch / 3 epochs\n",
" 79.165 iters/sec. Estimated time to finish: 0:00:01.263185.\n",
"\u001b[4A\u001b[J3 0.0700185 0.102149 0.978267 0.9693 22.7789 \n",
"\u001b[J total [##################################################] 100.00%\n",
"this epoch [..................................................] 0.00%\n",
" 1800 iter, 3 epoch / 3 epochs\n",
" 75.759 iters/sec. Estimated time to finish: 0:00:00.\n",
"\u001b[4A\u001b[J"
]
}
],
"source": [
"def test_original():\n",
" # Set up a neural network to train\n",
" # Classifier reports softmax cross entropy loss and accuracy at every\n",
" # iteration, which will be used by the PrintReport extension below.\n",
" model = L.Classifier(MLP(args.unit, 10))\n",
" if args.gpu >= 0:\n",
" # Make a specified GPU current\n",
" chainer.cuda.get_device_from_id(args.gpu).use()\n",
" model.to_gpu() # Copy the model to the GPU\n",
"\n",
" # Setup an optimizer\n",
" optimizer = chainer.optimizers.Adam()\n",
" optimizer.setup(model)\n",
"\n",
" # Load the MNIST dataset\n",
" train, test = chainer.datasets.get_mnist()\n",
"\n",
" train_iter = chainer.iterators.SerialIterator(train, args.batchsize)\n",
" test_iter = chainer.iterators.SerialIterator(test, args.batchsize,\n",
" repeat=False, shuffle=False)\n",
"\n",
" # Set up a trainer\n",
" updater = training.StandardUpdater(train_iter, optimizer, device=args.gpu)\n",
" trainer = training.Trainer(updater, (args.epoch, 'epoch'), out=args.out)\n",
"\n",
" # Evaluate the model with the test dataset for each epoch\n",
" trainer.extend(extensions.Evaluator(test_iter, model, device=args.gpu))\n",
"\n",
" # Dump a computational graph from 'loss' variable at the first iteration\n",
" # The \"main\" refers to the target link of the \"main\" optimizer.\n",
" trainer.extend(extensions.dump_graph('main/loss'))\n",
"\n",
" # Take a snapshot for each specified epoch\n",
" frequency = args.epoch if args.frequency == -1 else max(1, args.frequency)\n",
" trainer.extend(extensions.snapshot(), trigger=(frequency, 'epoch'))\n",
"\n",
" # Write a log of evaluation statistics for each epoch\n",
" trainer.extend(extensions.LogReport())\n",
"\n",
" # Save two plot images to the result dir\n",
" if extensions.PlotReport.available():\n",
" trainer.extend(\n",
" extensions.PlotReport(['main/loss', 'validation/main/loss'],\n",
" 'epoch', file_name='loss.png'))\n",
" trainer.extend(\n",
" extensions.PlotReport(\n",
" ['main/accuracy', 'validation/main/accuracy'],\n",
" 'epoch', file_name='accuracy.png'))\n",
"\n",
" # Print selected entries of the log to stdout\n",
" # Here \"main\" refers to the target link of the \"main\" optimizer again, and\n",
" # \"validation\" refers to the default name of the Evaluator extension.\n",
" # Entries other than 'epoch' are reported by the Classifier link, called by\n",
" # either the updater or the evaluator.\n",
" trainer.extend(extensions.PrintReport(\n",
" ['epoch', 'main/loss', 'validation/main/loss',\n",
" 'main/accuracy', 'validation/main/accuracy', 'elapsed_time']))\n",
"\n",
" # Print a progress bar to stdout\n",
" trainer.extend(extensions.ProgressBar())\n",
"\n",
" if args.resume:\n",
" # Resume from a snapshot\n",
" chainer.serializers.load_npz(args.resume, trainer)\n",
"\n",
" # Run the training\n",
" trainer.run()\n",
"\n",
"\n",
"test_original()"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"[{u'elapsed_time': 6.940603971481323,\n",
" u'epoch': 1,\n",
" u'iteration': 600,\n",
" u'main/accuracy': 0.9213500021273892,\n",
" u'main/loss': 0.2787705701092879,\n",
" u'validation/main/accuracy': 0.9598000049591064,\n",
" u'validation/main/loss': 0.13582063710317016},\n",
" {u'elapsed_time': 14.360282897949219,\n",
" u'epoch': 2,\n",
" u'iteration': 1200,\n",
" u'main/accuracy': 0.9687833403547604,\n",
" u'main/loss': 0.10487318117947628,\n",
" u'validation/main/accuracy': 0.9690000033378601,\n",
" u'validation/main/loss': 0.0970317642763257},\n",
" {u'elapsed_time': 22.778912782669067,\n",
" u'epoch': 3,\n",
" u'iteration': 1800,\n",
" u'main/accuracy': 0.9782666764656702,\n",
" u'main/loss': 0.07001853270921857,\n",
" u'validation/main/accuracy': 0.9693000078201294,\n",
" u'validation/main/loss': 0.10214902652136516}]"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import json\n",
"with open(\"result/log\") as fin:\n",
" result = json.load(fin)\n",
"\n",
"result"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Little Modification to Report Parameter Metrics\n",
"\n",
"This time I added `trainer.extend(extensions.ParameterStatistics(model))` to report parameter metrics (e.g. average, min, max, ...).\n"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"epoch main/loss validation/main/loss main/accuracy validation/main/accuracy elapsed_time\n",
"\u001b[J total [##................................................] 5.56%\n",
"this epoch [########..........................................] 16.67%\n",
" 100 iter, 0 epoch / 3 epochs\n",
" inf iters/sec. Estimated time to finish: 0:00:00.\n",
"\u001b[4A\u001b[J total [#####.............................................] 11.11%\n",
"this epoch [################..................................] 33.33%\n",
" 200 iter, 0 epoch / 3 epochs\n",
" 36.245 iters/sec. Estimated time to finish: 0:00:44.144241.\n",
"\u001b[4A\u001b[J total [########..........................................] 16.67%\n",
"this epoch [#########################.........................] 50.00%\n",
" 300 iter, 0 epoch / 3 epochs\n",
" 35.457 iters/sec. Estimated time to finish: 0:00:42.304253.\n",
"\u001b[4A\u001b[J total [###########.......................................] 22.22%\n",
"this epoch [#################################.................] 66.67%\n",
" 400 iter, 0 epoch / 3 epochs\n",
" 34.925 iters/sec. Estimated time to finish: 0:00:40.085701.\n",
"\u001b[4A\u001b[J total [#############.....................................] 27.78%\n",
"this epoch [#########################################.........] 83.33%\n",
" 500 iter, 0 epoch / 3 epochs\n",
" 36.517 iters/sec. Estimated time to finish: 0:00:35.600201.\n",
"\u001b[4A\u001b[J1 0.284049 0.135925 0.9185 0.9596 16.9108 \n",
"\u001b[J total [################..................................] 33.33%\n",
"this epoch [..................................................] 0.00%\n",
" 600 iter, 1 epoch / 3 epochs\n",
" 35.603 iters/sec. Estimated time to finish: 0:00:33.705008.\n",
"\u001b[4A\u001b[J total [###################...............................] 38.89%\n",
"this epoch [########..........................................] 16.67%\n",
" 700 iter, 1 epoch / 3 epochs\n",
" 36.07 iters/sec. Estimated time to finish: 0:00:30.496527.\n",
"\u001b[4A\u001b[J total [######################............................] 44.44%\n",
"this epoch [################..................................] 33.33%\n",
" 800 iter, 1 epoch / 3 epochs\n",
" 35.845 iters/sec. Estimated time to finish: 0:00:27.898093.\n",
"\u001b[4A\u001b[J total [#########################.........................] 50.00%\n",
"this epoch [#########################.........................] 50.00%\n",
" 900 iter, 1 epoch / 3 epochs\n",
" 36.036 iters/sec. Estimated time to finish: 0:00:24.975288.\n",
"\u001b[4A\u001b[J total [###########################.......................] 55.56%\n",
"this epoch [#################################.................] 66.67%\n",
" 1000 iter, 1 epoch / 3 epochs\n",
" 35.513 iters/sec. Estimated time to finish: 0:00:22.527099.\n",
"\u001b[4A\u001b[J total [##############################....................] 61.11%\n",
"this epoch [#########################################.........] 83.33%\n",
" 1100 iter, 1 epoch / 3 epochs\n",
" 36.083 iters/sec. Estimated time to finish: 0:00:19.399488.\n",
"\u001b[4A\u001b[J2 0.112587 0.0972494 0.96555 0.97 34.4807 \n",
"\u001b[J total [#################################.................] 66.67%\n",
"this epoch [..................................................] 0.00%\n",
" 1200 iter, 2 epoch / 3 epochs\n",
" 34.733 iters/sec. Estimated time to finish: 0:00:17.274399.\n",
"\u001b[4A\u001b[J total [####################################..............] 72.22%\n",
"this epoch [########..........................................] 16.67%\n",
" 1300 iter, 2 epoch / 3 epochs\n",
" 35.052 iters/sec. Estimated time to finish: 0:00:14.264700.\n",
"\u001b[4A\u001b[J total [######################################............] 77.78%\n",
"this epoch [################..................................] 33.33%\n",
" 1400 iter, 2 epoch / 3 epochs\n",
" 35.583 iters/sec. Estimated time to finish: 0:00:11.241423.\n",
"\u001b[4A\u001b[J total [#########################################.........] 83.33%\n",
"this epoch [#########################.........................] 50.00%\n",
" 1500 iter, 2 epoch / 3 epochs\n",
" 35.862 iters/sec. Estimated time to finish: 0:00:08.365285.\n",
"\u001b[4A\u001b[J total [############################################......] 88.89%\n",
"this epoch [#################################.................] 66.67%\n",
" 1600 iter, 2 epoch / 3 epochs\n",
" 35.639 iters/sec. Estimated time to finish: 0:00:05.611860.\n",
"\u001b[4A\u001b[J total [###############################################...] 94.44%\n",
"this epoch [#########################################.........] 83.33%\n",
" 1700 iter, 2 epoch / 3 epochs\n",
" 36.06 iters/sec. Estimated time to finish: 0:00:02.773170.\n",
"\u001b[4A\u001b[J3 0.0734751 0.079766 0.976883 0.9748 50.3524 \n",
"\u001b[J total [##################################################] 100.00%\n",
"this epoch [..................................................] 0.00%\n",
" 1800 iter, 3 epoch / 3 epochs\n",
" 35.806 iters/sec. Estimated time to finish: 0:00:00.\n",
"\u001b[4A\u001b[J"
]
}
],
"source": [
"def test_parameter_reporting():\n",
" # Set up a neural network to train\n",
" # Classifier reports softmax cross entropy loss and accuracy at every\n",
" # iteration, which will be used by the PrintReport extension below.\n",
" model = L.Classifier(MLP(args.unit, 10))\n",
" if args.gpu >= 0:\n",
" # Make a specified GPU current\n",
" chainer.cuda.get_device_from_id(args.gpu).use()\n",
" model.to_gpu() # Copy the model to the GPU\n",
"\n",
" # Setup an optimizer\n",
" optimizer = chainer.optimizers.Adam()\n",
" optimizer.setup(model)\n",
"\n",
" # Load the MNIST dataset\n",
" train, test = chainer.datasets.get_mnist()\n",
"\n",
" train_iter = chainer.iterators.SerialIterator(train, args.batchsize)\n",
" test_iter = chainer.iterators.SerialIterator(test, args.batchsize,\n",
" repeat=False, shuffle=False)\n",
"\n",
" # Set up a trainer\n",
" updater = training.StandardUpdater(train_iter, optimizer, device=args.gpu)\n",
" trainer = training.Trainer(updater, (args.epoch, 'epoch'), out=args.out)\n",
"\n",
" # Evaluate the model with the test dataset for each epoch\n",
" trainer.extend(extensions.Evaluator(test_iter, model, device=args.gpu))\n",
"\n",
" # Dump a computational graph from 'loss' variable at the first iteration\n",
" # The \"main\" refers to the target link of the \"main\" optimizer.\n",
" trainer.extend(extensions.dump_graph('main/loss'))\n",
"\n",
" # Take a snapshot for each specified epoch\n",
" frequency = args.epoch if args.frequency == -1 else max(1, args.frequency)\n",
" trainer.extend(extensions.snapshot(), trigger=(frequency, 'epoch'))\n",
"\n",
" ##### THIS IS WHERE I ADDED THE CODE #####\n",
" trainer.extend(extensions.ParameterStatistics(model))\n",
" \n",
" # Write a log of evaluation statistics for each epoch\n",
" trainer.extend(extensions.LogReport())\n",
"\n",
" # Save two plot images to the result dir\n",
" if extensions.PlotReport.available():\n",
" trainer.extend(\n",
" extensions.PlotReport(['main/loss', 'validation/main/loss'],\n",
" 'epoch', file_name='loss.png'))\n",
" trainer.extend(\n",
" extensions.PlotReport(\n",
" ['main/accuracy', 'validation/main/accuracy'],\n",
" 'epoch', file_name='accuracy.png'))\n",
"\n",
" # Print selected entries of the log to stdout\n",
" # Here \"main\" refers to the target link of the \"main\" optimizer again, and\n",
" # \"validation\" refers to the default name of the Evaluator extension.\n",
" # Entries other than 'epoch' are reported by the Classifier link, called by\n",
" # either the updater or the evaluator.\n",
" trainer.extend(extensions.PrintReport(\n",
" ['epoch', 'main/loss', 'validation/main/loss',\n",
" 'main/accuracy', 'validation/main/accuracy', 'elapsed_time']))\n",
"\n",
" # Print a progress bar to stdout\n",
" trainer.extend(extensions.ProgressBar())\n",
"\n",
" if args.resume:\n",
" # Resume from a snapshot\n",
" chainer.serializers.load_npz(args.resume, trainer)\n",
"\n",
" # Run the training\n",
" trainer.run()\n",
"\n",
"\n",
"test_parameter_reporting()"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"[{u'None/predictor/l1/W/data/max': 0.18769985591371854,\n",
" u'None/predictor/l1/W/data/mean': 0.0006860141372822189,\n",
" u'None/predictor/l1/W/data/min': -0.21658104345202445,\n",
" u'None/predictor/l1/W/data/percentile/0': -0.1320047355272498,\n",
" u'None/predictor/l1/W/data/percentile/1': -0.08497818301255008,\n",
" u'None/predictor/l1/W/data/percentile/2': -0.04122352957670082,\n",
" u'None/predictor/l1/W/data/percentile/3': 0.0008963784146650747,\n",
" u'None/predictor/l1/W/data/percentile/4': 0.0428067545834066,\n",
" u'None/predictor/l1/W/data/percentile/5': 0.08460525170120099,\n",
" u'None/predictor/l1/W/data/percentile/6': 0.1288499908778201,\n",
" u'None/predictor/l1/W/data/std': 0.04230530080075065,\n",
" u'None/predictor/l1/W/data/zeros': 0.0,\n",
" u'None/predictor/l1/W/grad/max': 0.013363681441017737,\n",
" u'None/predictor/l1/W/grad/mean': -2.5457882616741755e-06,\n",
" u'None/predictor/l1/W/grad/min': -0.013039940825353065,\n",
" u'None/predictor/l1/W/grad/percentile/0': -0.007410150919836411,\n",
" u'None/predictor/l1/W/grad/percentile/1': -0.0032792949646553034,\n",
" u'None/predictor/l1/W/grad/percentile/2': -0.00044474147839619377,\n",
" u'None/predictor/l1/W/grad/percentile/3': 0.0,\n",
" u'None/predictor/l1/W/grad/percentile/4': 0.00041870361433874256,\n",
" u'None/predictor/l1/W/grad/percentile/5': 0.003310361570617339,\n",
" u'None/predictor/l1/W/grad/percentile/6': 0.007486925448873247,\n",
" u'None/predictor/l1/W/grad/std': 0.001357260803342797,\n",
" u'None/predictor/l1/W/grad/zeros': 64320.0,\n",
" u'None/predictor/l1/b/data/max': 0.0548983522199948,\n",
" u'None/predictor/l1/b/data/mean': 0.012055125747569945,\n",
" u'None/predictor/l1/b/data/min': -0.029791952808154748,\n",
" u'None/predictor/l1/b/data/percentile/0': -0.028762824200974896,\n",
" u'None/predictor/l1/b/data/percentile/1': -0.016089606840499066,\n",
" u'None/predictor/l1/b/data/percentile/2': -0.005493110008535476,\n",
" u'None/predictor/l1/b/data/percentile/3': 0.012852825987016937,\n",
" u'None/predictor/l1/b/data/percentile/4': 0.027909995105618788,\n",
" u'None/predictor/l1/b/data/percentile/5': 0.03809976473372724,\n",
" u'None/predictor/l1/b/data/percentile/6': 0.053035637028444944,\n",
" u'None/predictor/l1/b/data/std': 0.0152910965756746,\n",
" u'None/predictor/l1/b/data/zeros': 0.0,\n",
" u'None/predictor/l1/b/grad/max': 0.013211307424741487,\n",
" u'None/predictor/l1/b/grad/mean': -7.848533850525049e-05,\n",
" u'None/predictor/l1/b/grad/min': -0.013068354073135804,\n",
" u'None/predictor/l1/b/grad/percentile/0': -0.012563953940001484,\n",
" u'None/predictor/l1/b/grad/percentile/1': -0.008738801820065204,\n",
" u'None/predictor/l1/b/grad/percentile/2': -0.004217391109127184,\n",
" u'None/predictor/l1/b/grad/percentile/3': -9.326697650503609e-05,\n",
" u'None/predictor/l1/b/grad/percentile/4': 0.004075311267449891,\n",
" u'None/predictor/l1/b/grad/percentile/5': 0.008706802054390925,\n",
" u'None/predictor/l1/b/grad/percentile/6': 0.012666157135854028,\n",
" u'None/predictor/l1/b/grad/std': 0.004354175603560482,\n",
" u'None/predictor/l1/b/grad/zeros': 7.0,\n",
" u'None/predictor/l2/W/data/max': 0.3254947525759538,\n",
" u'None/predictor/l2/W/data/mean': 0.003950303411500803,\n",
" u'None/predictor/l2/W/data/min': -0.32085575138529143,\n",
" u'None/predictor/l2/W/data/percentile/0': -0.23240646572480841,\n",
" u'None/predictor/l2/W/data/percentile/1': -0.14860894085486717,\n",
" u'None/predictor/l2/W/data/percentile/2': -0.07275652346237013,\n",
" u'None/predictor/l2/W/data/percentile/3': 0.0038656487325230654,\n",
" u'None/predictor/l2/W/data/percentile/4': 0.08040970983579382,\n",
" u'None/predictor/l2/W/data/percentile/5': 0.15971063587005932,\n",
" u'None/predictor/l2/W/data/percentile/6': 0.24027507231731624,\n",
" u'None/predictor/l2/W/data/std': 0.07691252761830886,\n",
" u'None/predictor/l2/W/data/zeros': 0.0,\n",
" u'None/predictor/l2/W/grad/max': 0.015284300476778298,\n",
" u'None/predictor/l2/W/grad/mean': -1.8781126353412483e-05,\n",
" u'None/predictor/l2/W/grad/min': -0.015172957402343552,\n",
" u'None/predictor/l2/W/grad/percentile/0': -0.00934184331272488,\n",
" u'None/predictor/l2/W/grad/percentile/1': -0.004634502316668396,\n",
" u'None/predictor/l2/W/grad/percentile/2': -0.001432369026768807,\n",
" u'None/predictor/l2/W/grad/percentile/3': -4.3381341113664836e-07,\n",
" u'None/predictor/l2/W/grad/percentile/4': 0.0013815956475082766,\n",
" u'None/predictor/l2/W/grad/percentile/5': 0.004639800911789835,\n",
" u'None/predictor/l2/W/grad/percentile/6': 0.009400696555757971,\n",
" u'None/predictor/l2/W/grad/std': 0.002028256375051569,\n",
" u'None/predictor/l2/W/grad/zeros': 5744.0,\n",
" u'None/predictor/l2/b/data/max': 0.04791020446437566,\n",
" u'None/predictor/l2/b/data/mean': 0.010086397930419934,\n",
" u'None/predictor/l2/b/data/min': -0.032200952287918576,\n",
" u'None/predictor/l2/b/data/percentile/0': -0.03167514712518647,\n",
" u'None/predictor/l2/b/data/percentile/1': -0.019231082678096045,\n",
" u'None/predictor/l2/b/data/percentile/2': -0.007152004249883097,\n",
" u'None/predictor/l2/b/data/percentile/3': 0.011697621105352786,\n",
" u'None/predictor/l2/b/data/percentile/4': 0.025173387602675477,\n",
" u'None/predictor/l2/b/data/percentile/5': 0.037360551587680896,\n",
" u'None/predictor/l2/b/data/percentile/6': 0.04640533133965915,\n",
" u'None/predictor/l2/b/data/std': 0.015411058925868323,\n",
" u'None/predictor/l2/b/data/zeros': 0.0,\n",
" u'None/predictor/l2/b/grad/max': 0.007923950732608015,\n",
" u'None/predictor/l2/b/grad/mean': -8.564058850538459e-05,\n",
" u'None/predictor/l2/b/grad/min': -0.008049897506910687,\n",
" u'None/predictor/l2/b/grad/percentile/0': -0.0077712604418969725,\n",
" u'None/predictor/l2/b/grad/percentile/1': -0.005579872188582551,\n",
" u'None/predictor/l2/b/grad/percentile/2': -0.0026059530557865306,\n",
" u'None/predictor/l2/b/grad/percentile/3': -3.509546817892062e-05,\n",
" u'None/predictor/l2/b/grad/percentile/4': 0.0024047468276375064,\n",
" u'None/predictor/l2/b/grad/percentile/5': 0.005453195185978501,\n",
" u'None/predictor/l2/b/grad/percentile/6': 0.007656289307926475,\n",
" u'None/predictor/l2/b/grad/std': 0.0026661352358253983,\n",
" u'None/predictor/l2/b/grad/zeros': 20.0,\n",
" u'None/predictor/l3/W/data/max': 0.24597564838826658,\n",
" u'None/predictor/l3/W/data/mean': -0.006622560692982613,\n",
" u'None/predictor/l3/W/data/min': -0.32619385354220864,\n",
" u'None/predictor/l3/W/data/percentile/0': -0.2306768275239463,\n",
" u'None/predictor/l3/W/data/percentile/1': -0.17660742681579786,\n",
" u'None/predictor/l3/W/data/percentile/2': -0.09317452937077603,\n",
" u'None/predictor/l3/W/data/percentile/3': -0.005512705517830909,\n",
" u'None/predictor/l3/W/data/percentile/4': 0.07775105796712511,\n",
" u'None/predictor/l3/W/data/percentile/5': 0.15502845389973113,\n",
" u'None/predictor/l3/W/data/percentile/6': 0.21997718072928502,\n",
" u'None/predictor/l3/W/data/std': 0.08410330789784591,\n",
" u'None/predictor/l3/W/data/zeros': 0.0,\n",
" u'None/predictor/l3/W/grad/max': 0.0734366792657723,\n",
" u'None/predictor/l3/W/grad/mean': 4.1129688657196943e-10,\n",
" u'None/predictor/l3/W/grad/min': -0.07636503320497771,\n",
" u'None/predictor/l3/W/grad/percentile/0': -0.06427242998719027,\n",
" u'None/predictor/l3/W/grad/percentile/1': -0.036685596814023974,\n",
" u'None/predictor/l3/W/grad/percentile/2': -0.00975563835099175,\n",
" u'None/predictor/l3/W/grad/percentile/3': 9.153801069754393e-05,\n",
" u'None/predictor/l3/W/grad/percentile/4': 0.009530000375109747,\n",
" u'None/predictor/l3/W/grad/percentile/5': 0.03502504599237309,\n",
" u'None/predictor/l3/W/grad/percentile/6': 0.06173668931481425,\n",
" u'None/predictor/l3/W/grad/std': 0.01510179663853099,\n",
" u'None/predictor/l3/W/grad/zeros': 205.0,\n",
" u'None/predictor/l3/b/data/max': 0.02937242121338689,\n",
" u'None/predictor/l3/b/data/mean': 0.0006898459542465693,\n",
" u'None/predictor/l3/b/data/min': -0.03381775257643312,\n",
" u'None/predictor/l3/b/data/percentile/0': -0.033618681625444825,\n",
" u'None/predictor/l3/b/data/percentile/1': -0.03032635435910036,\n",
" u'None/predictor/l3/b/data/percentile/2': -0.010998366487167227,\n",
" u'None/predictor/l3/b/data/percentile/3': -0.0001515522909911245,\n",
" u'None/predictor/l3/b/data/percentile/4': 0.015128768134392312,\n",
" u'None/predictor/l3/b/data/percentile/5': 0.026865516303920516,\n",
" u'None/predictor/l3/b/data/percentile/6': 0.029229483652759437,\n",
" u'None/predictor/l3/b/data/std': 0.01670785871508997,\n",
" u'None/predictor/l3/b/data/zeros': 0.0,\n",
" u'None/predictor/l3/b/grad/max': 0.020958167066176734,\n",
" u'None/predictor/l3/b/grad/mean': 3.363035551400227e-10,\n",
" u'None/predictor/l3/b/grad/min': -0.021081253852850448,\n",
" u'None/predictor/l3/b/grad/percentile/0': -0.02098587781309286,\n",
" u'None/predictor/l3/b/grad/percentile/1': -0.019408504847871797,\n",
" u'None/predictor/l3/b/grad/percentile/2': -0.01083751097976544,\n",
" u'None/predictor/l3/b/grad/percentile/3': -1.5249256520064592e-05,\n",
" u'None/predictor/l3/b/grad/percentile/4': 0.010781639548587657,\n",
" u'None/predictor/l3/b/grad/percentile/5': 0.01927716390611354,\n",
" u'None/predictor/l3/b/grad/percentile/6': 0.020862320394769613,\n",
" u'None/predictor/l3/b/grad/std': 0.012271623574585343,\n",
" u'None/predictor/l3/b/grad/zeros': 0.0,\n",
" u'elapsed_time': 16.910752058029175,\n",
" u'epoch': 1,\n",
" u'iteration': 600,\n",
" u'main/accuracy': 0.9185000030696392,\n",
" u'main/loss': 0.28404941281303764,\n",
" u'validation/main/accuracy': 0.9596000021696091,\n",
" u'validation/main/loss': 0.13592462537344546},\n",
" {u'None/predictor/l1/W/data/max': 0.2295598153024912,\n",
" u'None/predictor/l1/W/data/mean': -0.0010978691087802873,\n",
" u'None/predictor/l1/W/data/min': -0.3337414373457432,\n",
" u'None/predictor/l1/W/data/percentile/0': -0.1715770532056913,\n",
" u'None/predictor/l1/W/data/percentile/1': -0.10327599270020436,\n",
" u'None/predictor/l1/W/data/percentile/2': -0.049214466670726716,\n",
" u'None/predictor/l1/W/data/percentile/3': -0.0005298784545751308,\n",
" u'None/predictor/l1/W/data/percentile/4': 0.04737412391026749,\n",
" u'None/predictor/l1/W/data/percentile/5': 0.09677141055891911,\n",
" u'None/predictor/l1/W/data/percentile/6': 0.15023443747842113,\n",
" u'None/predictor/l1/W/data/std': 0.049587591936190924,\n",
" u'None/predictor/l1/W/data/zeros': 0.0,\n",
" u'None/predictor/l1/W/grad/max': 0.01430551103898324,\n",
" u'None/predictor/l1/W/grad/mean': 1.9707583105225503e-06,\n",
" u'None/predictor/l1/W/grad/min': -0.013791593328739206,\n",
" u'None/predictor/l1/W/grad/percentile/0': -0.007352665172329152,\n",
" u'None/predictor/l1/W/grad/percentile/1': -0.0026086697971421536,\n",
" u'None/predictor/l1/W/grad/percentile/2': -0.00021726978253297155,\n",
" u'None/predictor/l1/W/grad/percentile/3': 0.0,\n",
" u'None/predictor/l1/W/grad/percentile/4': 0.00016954874927197587,\n",
" u'None/predictor/l1/W/grad/percentile/5': 0.00274226690423861,\n",
" u'None/predictor/l1/W/grad/percentile/6': 0.007588650849498316,\n",
" u'None/predictor/l1/W/grad/std': 0.0011635373037764414,\n",
" u'None/predictor/l1/W/grad/zeros': 67089.0,\n",
" u'None/predictor/l1/b/data/max': 0.08538756048927705,\n",
" u'None/predictor/l1/b/data/mean': 0.013272578072113295,\n",
" u'None/predictor/l1/b/data/min': -0.047418129605551564,\n",
" u'None/predictor/l1/b/data/percentile/0': -0.043574936308885356,\n",
" u'None/predictor/l1/b/data/percentile/1': -0.023837999054799473,\n",
" u'None/predictor/l1/b/data/percentile/2': -0.008634803174209161,\n",
" u'None/predictor/l1/b/data/percentile/3': 0.012941058898965518,\n",
" u'None/predictor/l1/b/data/percentile/4': 0.03444813090563782,\n",
" u'None/predictor/l1/b/data/percentile/5': 0.05307643705038971,\n",
" u'None/predictor/l1/b/data/percentile/6': 0.08173325167713376,\n",
" u'None/predictor/l1/b/data/std': 0.021116428058594466,\n",
" u'None/predictor/l1/b/data/zeros': 0.0,\n",
" u'None/predictor/l1/b/grad/max': 0.013569926251657307,\n",
" u'None/predictor/l1/b/grad/mean': -9.288730958392222e-06,\n",
" u'None/predictor/l1/b/grad/min': -0.01307839693656812,\n",
" u'None/predictor/l1/b/grad/percentile/0': -0.01237880468803853,\n",
" u'None/predictor/l1/b/grad/percentile/1': -0.007426289065470298,\n",
" u'None/predictor/l1/b/grad/percentile/2': -0.0031788241997480304,\n",
" u'None/predictor/l1/b/grad/percentile/3': -5.736688619763915e-05,\n",
" u'None/predictor/l1/b/grad/percentile/4': 0.003198311038519045,\n",
" u'None/predictor/l1/b/grad/percentile/5': 0.007663695951051998,\n",
" u'None/predictor/l1/b/grad/percentile/6': 0.012872878613070614,\n",
" u'None/predictor/l1/b/grad/std': 0.003696984149767862,\n",
" u'None/predictor/l1/b/grad/zeros': 6.0,\n",
" u'None/predictor/l2/W/data/max': 0.38595819850762686,\n",
" u'None/predictor/l2/W/data/mean': 0.004569140146486461,\n",
" u'None/predictor/l2/W/data/min': -0.39474050198992094,\n",
" u'None/predictor/l2/W/data/percentile/0': -0.2631317454326082,\n",
" u'None/predictor/l2/W/data/percentile/1': -0.16407296296592544,\n",
" u'None/predictor/l2/W/data/percentile/2': -0.07859283381623271,\n",
" u'None/predictor/l2/W/data/percentile/3': 0.00472891669603996,\n",
" u'None/predictor/l2/W/data/percentile/4': 0.08707444573851172,\n",
" u'None/predictor/l2/W/data/percentile/5': 0.17526727512900028,\n",
" u'None/predictor/l2/W/data/percentile/6': 0.28157698171670975,\n",
" u'None/predictor/l2/W/data/std': 0.08440262224525213,\n",
" u'None/predictor/l2/W/data/zeros': 0.0,\n",
" u'None/predictor/l2/W/grad/max': 0.011583656925940886,\n",
" u'None/predictor/l2/W/grad/mean': -2.105507710060313e-06,\n",
" u'None/predictor/l2/W/grad/min': -0.01143122955535849,\n",
" u'None/predictor/l2/W/grad/percentile/0': -0.006480507606540278,\n",
" u'None/predictor/l2/W/grad/percentile/1': -0.0029765916828615184,\n",
" u'None/predictor/l2/W/grad/percentile/2': -0.0008427810466160833,\n",
" u'None/predictor/l2/W/grad/percentile/3': -1.6187956525482346e-07,\n",
" u'None/predictor/l2/W/grad/percentile/4': 0.0008328083104854974,\n",
" u'None/predictor/l2/W/grad/percentile/5': 0.0030833940541642947,\n",
" u'None/predictor/l2/W/grad/percentile/6': 0.0066814565342781995,\n",
" u'None/predictor/l2/W/grad/std': 0.0013231629846510865,\n",
" u'None/predictor/l2/W/grad/zeros': 4243.0,\n",
" u'None/predictor/l2/b/data/max': 0.05736853959038854,\n",
" u'None/predictor/l2/b/data/mean': 0.014847257463261486,\n",
" u'None/predictor/l2/b/data/min': -0.04029602174957593,\n",
" u'None/predictor/l2/b/data/percentile/0': -0.038776279838063095,\n",
" u'None/predictor/l2/b/data/percentile/1': -0.019918942003246426,\n",
" u'None/predictor/l2/b/data/percentile/2': -0.004548643746025865,\n",
" u'None/predictor/l2/b/data/percentile/3': 0.01614985933760181,\n",
" u'None/predictor/l2/b/data/percentile/4': 0.033526092198897815,\n",
" u'None/predictor/l2/b/data/percentile/5': 0.04903571015953027,\n",
" u'None/predictor/l2/b/data/percentile/6': 0.05653972272616058,\n",
" u'None/predictor/l2/b/data/std': 0.018573332748686274,\n",
" u'None/predictor/l2/b/data/zeros': 0.0,\n",
" u'None/predictor/l2/b/grad/max': 0.005730223531524341,\n",
" u'None/predictor/l2/b/grad/mean': -1.6921611055451535e-05,\n",
" u'None/predictor/l2/b/grad/min': -0.005733378459893477,\n",
" u'None/predictor/l2/b/grad/percentile/0': -0.005532953142384715,\n",
" u'None/predictor/l2/b/grad/percentile/1': -0.003881969137598451,\n",
" u'None/predictor/l2/b/grad/percentile/2': -0.0018137920477178915,\n",
" u'None/predictor/l2/b/grad/percentile/3': -1.8478656669386102e-05,\n",
" u'None/predictor/l2/b/grad/percentile/4': 0.0017994521200937993,\n",
" u'None/predictor/l2/b/grad/percentile/5': 0.003973955448431132,\n",
" u'None/predictor/l2/b/grad/percentile/6': 0.005549185480232125,\n",
" u'None/predictor/l2/b/grad/std': 0.0019191197817660093,\n",
" u'None/predictor/l2/b/grad/zeros': 13.0,\n",
" u'None/predictor/l3/W/data/max': 0.2618614660203457,\n",
" u'None/predictor/l3/W/data/mean': -0.010291073985087374,\n",
" u'None/predictor/l3/W/data/min': -0.445046594440937,\n",
" u'None/predictor/l3/W/data/percentile/0': -0.2895010172459134,\n",
" u'None/predictor/l3/W/data/percentile/1': -0.21029580955842156,\n",
" u'None/predictor/l3/W/data/percentile/2': -0.11003875367748367,\n",
" u'None/predictor/l3/W/data/percentile/3': -0.008205772105914852,\n",
" u'None/predictor/l3/W/data/percentile/4': 0.086189404697402,\n",
" u'None/predictor/l3/W/data/percentile/5': 0.17154898492981963,\n",
" u'None/predictor/l3/W/data/percentile/6': 0.2427794595418794,\n",
" u'None/predictor/l3/W/data/std': 0.09699305151899655,\n",
" u'None/predictor/l3/W/data/zeros': 0.0,\n",
" u'None/predictor/l3/W/grad/max': 0.047409058484093596,\n",
" u'None/predictor/l3/W/grad/mean': 1.141019166601397e-09,\n",
" u'None/predictor/l3/W/grad/min': -0.0504266367907015,\n",
" u'None/predictor/l3/W/grad/percentile/0': -0.04303241542093122,\n",
" u'None/predictor/l3/W/grad/percentile/1': -0.023907562216809845,\n",
" u'None/predictor/l3/W/grad/percentile/2': -0.005392467489394444,\n",
" u'None/predictor/l3/W/grad/percentile/3': 0.00013196134112376448,\n",
" u'None/predictor/l3/W/grad/percentile/4': 0.005287904573399991,\n",
" u'None/predictor/l3/W/grad/percentile/5': 0.022380537163379257,\n",
" u'None/predictor/l3/W/grad/percentile/6': 0.04061083738548267,\n",
" u'None/predictor/l3/W/grad/std': 0.009535577121035506,\n",
" u'None/predictor/l3/W/grad/zeros': 135.0,\n",
" u'None/predictor/l3/b/data/max': 0.026389154453451434,\n",
" u'None/predictor/l3/b/data/mean': -0.0002436420226024912,\n",
" u'None/predictor/l3/b/data/min': -0.029270565885429582,\n",
" u'None/predictor/l3/b/data/percentile/0': -0.029142855942780484,\n",
" u'None/predictor/l3/b/data/percentile/1': -0.027030729968198917,\n",
" u'None/predictor/l3/b/data/percentile/2': -0.013837539287600236,\n",
" u'None/predictor/l3/b/data/percentile/3': -0.0006820206931130694,\n",
" u'None/predictor/l3/b/data/percentile/4': 0.015336115136258799,\n",
" u'None/predictor/l3/b/data/percentile/5': 0.02557801643028732,\n",
" u'None/predictor/l3/b/data/percentile/6': 0.02634290535563945,\n",
" u'None/predictor/l3/b/data/std': 0.015980343559446435,\n",
" u'None/predictor/l3/b/data/zeros': 0.0,\n",
" u'None/predictor/l3/b/grad/max': 0.013331305620376953,\n",
" u'None/predictor/l3/b/grad/mean': 1.002950659222008e-09,\n",
" u'None/predictor/l3/b/grad/min': -0.013989534929278307,\n",
" u'None/predictor/l3/b/grad/percentile/0': -0.013920330119702479,\n",
" u'None/predictor/l3/b/grad/percentile/1': -0.012775789038256071,\n",
" u'None/predictor/l3/b/grad/percentile/2': -0.006642490836205779,\n",
" u'None/predictor/l3/b/grad/percentile/3': 0.00030572686263478016,\n",
" u'None/predictor/l3/b/grad/percentile/4': 0.0064399208388081575,\n",
" u'None/predictor/l3/b/grad/percentile/5': 0.012200348811597216,\n",
" u'None/predictor/l3/b/grad/percentile/6': 0.013266821240928988,\n",
" u'None/predictor/l3/b/grad/std': 0.007767713066617337,\n",
" u'None/predictor/l3/b/grad/zeros': 0.0,\n",
" u'elapsed_time': 34.48065805435181,\n",
" u'epoch': 2,\n",
" u'iteration': 1200,\n",
" u'main/accuracy': 0.9655500056346258,\n",
" u'main/loss': 0.11258724196814,\n",
" u'validation/main/accuracy': 0.9700000059604644,\n",
" u'validation/main/loss': 0.09724940008949488},\n",
" {u'None/predictor/l1/W/data/max': 0.27185197348395984,\n",
" u'None/predictor/l1/W/data/mean': -0.0018086677351190398,\n",
" u'None/predictor/l1/W/data/min': -0.3958187406758467,\n",
" u'None/predictor/l1/W/data/percentile/0': -0.20385727424958108,\n",
" u'None/predictor/l1/W/data/percentile/1': -0.11698534017020969,\n",
" u'None/predictor/l1/W/data/percentile/2': -0.05451850489750008,\n",
" u'None/predictor/l1/W/data/percentile/3': -0.0009897575474557621,\n",
" u'None/predictor/l1/W/data/percentile/4': 0.05137862113049197,\n",
" u'None/predictor/l1/W/data/percentile/5': 0.10710877471574153,\n",
" u'None/predictor/l1/W/data/percentile/6': 0.16940773534787262,\n",
" u'None/predictor/l1/W/data/std': 0.0551770847166578,\n",
" u'None/predictor/l1/W/data/zeros': 0.0,\n",
" u'None/predictor/l1/W/grad/max': 0.013656101867284937,\n",
" u'None/predictor/l1/W/grad/mean': 9.098795635746152e-07,\n",
" u'None/predictor/l1/W/grad/min': -0.013285819670224252,\n",
" u'None/predictor/l1/W/grad/percentile/0': -0.007165272038014878,\n",
" u'None/predictor/l1/W/grad/percentile/1': -0.0022454656175137882,\n",
" u'None/predictor/l1/W/grad/percentile/2': -0.00012885856785527783,\n",
" u'None/predictor/l1/W/grad/percentile/3': 0.0,\n",
" u'None/predictor/l1/W/grad/percentile/4': 8.493975607754234e-05,\n",
" u'None/predictor/l1/W/grad/percentile/5': 0.0023623810969903517,\n",
" u'None/predictor/l1/W/grad/percentile/6': 0.007440176921659513,\n",
" u'None/predictor/l1/W/grad/std': 0.0010541691524849738,\n",
" u'None/predictor/l1/W/grad/zeros': 68023.0,\n",
" u'None/predictor/l1/b/data/max': 0.09409072152028482,\n",
" u'None/predictor/l1/b/data/mean': 0.014366240447076658,\n",
" u'None/predictor/l1/b/data/min': -0.05377215398475528,\n",
" u'None/predictor/l1/b/data/percentile/0': -0.04946540089750718,\n",
" u'None/predictor/l1/b/data/percentile/1': -0.0262206041258089,\n",
" u'None/predictor/l1/b/data/percentile/2': -0.010352576444571432,\n",
" u'None/predictor/l1/b/data/percentile/3': 0.013620692878806342,\n",
" u'None/predictor/l1/b/data/percentile/4': 0.03883927839730615,\n",
" u'None/predictor/l1/b/data/percentile/5': 0.06370427280726525,\n",
" u'None/predictor/l1/b/data/percentile/6': 0.09198039733007803,\n",
" u'None/predictor/l1/b/data/std': 0.02443187476756672,\n",
" u'None/predictor/l1/b/data/zeros': 0.0,\n",
" u'None/predictor/l1/b/grad/max': 0.012916438297446197,\n",
" u'None/predictor/l1/b/grad/mean': -6.758792552924812e-06,\n",
" u'None/predictor/l1/b/grad/min': -0.012597657297543871,\n",
" u'None/predictor/l1/b/grad/percentile/0': -0.011882684694883723,\n",
" u'None/predictor/l1/b/grad/percentile/1': -0.006773838302144701,\n",
" u'None/predictor/l1/b/grad/percentile/2': -0.002665127662558216,\n",
" u'None/predictor/l1/b/grad/percentile/3': -6.465573986620398e-05,\n",
" u'None/predictor/l1/b/grad/percentile/4': 0.0026809942840123457,\n",
" u'None/predictor/l1/b/grad/percentile/5': 0.007103103777146238,\n",
" u'None/predictor/l1/b/grad/percentile/6': 0.012212283092132662,\n",
" u'None/predictor/l1/b/grad/std': 0.0033267931569328844,\n",
" u'None/predictor/l1/b/grad/zeros': 6.0,\n",
" u'None/predictor/l2/W/data/max': 0.42594017311930654,\n",
" u'None/predictor/l2/W/data/mean': 0.004585100904417535,\n",
" u'None/predictor/l2/W/data/min': -0.4193602364261945,\n",
" u'None/predictor/l2/W/data/percentile/0': -0.2870031386104177,\n",
" u'None/predictor/l2/W/data/percentile/1': -0.17638268876893315,\n",
" u'None/predictor/l2/W/data/percentile/2': -0.08273029080755269,\n",
" u'None/predictor/l2/W/data/percentile/3': 0.004919045708763102,\n",
" u'None/predictor/l2/W/data/percentile/4': 0.09155897223883613,\n",
" u'None/predictor/l2/W/data/percentile/5': 0.18563907284932304,\n",
" u'None/predictor/l2/W/data/percentile/6': 0.3057549727208191,\n",
" u'None/predictor/l2/W/data/std': 0.08973173742492993,\n",
" u'None/predictor/l2/W/data/zeros': 0.0,\n",
" u'None/predictor/l2/W/grad/max': 0.010738343312016998,\n",
" u'None/predictor/l2/W/grad/mean': -1.1086801492895167e-06,\n",
" u'None/predictor/l2/W/grad/min': -0.010554489806527272,\n",
" u'None/predictor/l2/W/grad/percentile/0': -0.00569238985479338,\n",
" u'None/predictor/l2/W/grad/percentile/1': -0.0024711209500458595,\n",
" u'None/predictor/l2/W/grad/percentile/2': -0.0006279451677817872,\n",
" u'None/predictor/l2/W/grad/percentile/3': -1.1413112943836623e-07,\n",
" u'None/predictor/l2/W/grad/percentile/4': 0.0006171325545753931,\n",
" u'None/predictor/l2/W/grad/percentile/5': 0.002569888059629768,\n",
" u'None/predictor/l2/W/grad/percentile/6': 0.0059066941723352456,\n",
" u'None/predictor/l2/W/grad/std': 0.0010985501808075545,\n",
" u'None/predictor/l2/W/grad/zeros': 3916.0,\n",
" u'None/predictor/l2/b/data/max': 0.07187967089936137,\n",
" u'None/predictor/l2/b/data/mean': 0.01782079283768932,\n",
" u'None/predictor/l2/b/data/min': -0.043728755054374534,\n",
" u'None/predictor/l2/b/data/percentile/0': -0.04198552798283009,\n",
" u'None/predictor/l2/b/data/percentile/1': -0.021558611168259138,\n",
" u'None/predictor/l2/b/data/percentile/2': -0.006673424679421003,\n",
" u'None/predictor/l2/b/data/percentile/3': 0.018262709815365572,\n",
" u'None/predictor/l2/b/data/percentile/4': 0.04141569238068903,\n",
" u'None/predictor/l2/b/data/percentile/5': 0.06002165412134422,\n",
" u'None/predictor/l2/b/data/percentile/6': 0.07119944974335653,\n",
" u'None/predictor/l2/b/data/std': 0.02276027821935713,\n",
" u'None/predictor/l2/b/data/zeros': 0.0,\n",
" u'None/predictor/l2/b/grad/max': 0.0049160390564550955,\n",
" u'None/predictor/l2/b/grad/mean': -9.985572270139186e-06,\n",
" u'None/predictor/l2/b/grad/min': -0.004835934512666427,\n",
" u'None/predictor/l2/b/grad/percentile/0': -0.004664398826965725,\n",
" u'None/predictor/l2/b/grad/percentile/1': -0.003293416322356983,\n",
" u'None/predictor/l2/b/grad/percentile/2': -0.0015120295459652303,\n",
" u'None/predictor/l2/b/grad/percentile/3': -1.5014628050688102e-05,\n",
" u'None/predictor/l2/b/grad/percentile/4': 0.0015077708413442064,\n",
" u'None/predictor/l2/b/grad/percentile/5': 0.0033866817279028477,\n",
" u'None/predictor/l2/b/grad/percentile/6': 0.004757825256751487,\n",
" u'None/predictor/l2/b/grad/std': 0.0016174230916900947,\n",
" u'None/predictor/l2/b/grad/zeros': 11.0,\n",
" u'None/predictor/l3/W/data/max': 0.30353682513038316,\n",
" u'None/predictor/l3/W/data/mean': -0.012781197633594275,\n",
" u'None/predictor/l3/W/data/min': -0.48483591253558794,\n",
" u'None/predictor/l3/W/data/percentile/0': -0.34600310326281686,\n",
" u'None/predictor/l3/W/data/percentile/1': -0.23641301063910153,\n",
" u'None/predictor/l3/W/data/percentile/2': -0.12205532338128369,\n",
" u'None/predictor/l3/W/data/percentile/3': -0.00738427418245313,\n",
" u'None/predictor/l3/W/data/percentile/4': 0.09309118541500833,\n",
" u'None/predictor/l3/W/data/percentile/5': 0.1800933897688586,\n",
" u'None/predictor/l3/W/data/percentile/6': 0.25322595016557464,\n",
" u'None/predictor/l3/W/data/std': 0.10624750250329575,\n",
" u'None/predictor/l3/W/data/zeros': 0.0,\n",
" u'None/predictor/l3/W/grad/max': 0.03892977581902717,\n",
" u'None/predictor/l3/W/grad/mean': 2.3498230971360362e-09,\n",
" u'None/predictor/l3/W/grad/min': -0.0421846509918881,\n",
" u'None/predictor/l3/W/grad/percentile/0': -0.03579130570738624,\n",
" u'None/predictor/l3/W/grad/percentile/1': -0.0190631178547456,\n",
" u'None/predictor/l3/W/grad/percentile/2': -0.0036464376532477265,\n",
" u'None/predictor/l3/W/grad/percentile/3': 9.100759447644958e-05,\n",
" u'None/predictor/l3/W/grad/percentile/4': 0.0036814866487753072,\n",
" u'None/predictor/l3/W/grad/percentile/5': 0.01758430899790021,\n",
" u'None/predictor/l3/W/grad/percentile/6': 0.03288057050147398,\n",
" u'None/predictor/l3/W/grad/std': 0.0074729266262147575,\n",
" u'None/predictor/l3/W/grad/zeros': 119.0,\n",
" u'None/predictor/l3/b/data/max': 0.02353643614798784,\n",
" u'None/predictor/l3/b/data/mean': -0.0014332280014059506,\n",
" u'None/predictor/l3/b/data/min': -0.036389247166613736,\n",
" u'None/predictor/l3/b/data/percentile/0': -0.036124117515093375,\n",
" u'None/predictor/l3/b/data/percentile/1': -0.031739280970717466,\n",
" u'None/predictor/l3/b/data/percentile/2': -0.011910545684384266,\n",
" u'None/predictor/l3/b/data/percentile/3': -0.0017472865573942423,\n",
" u'None/predictor/l3/b/data/percentile/4': 0.013693675600731977,\n",
" u'None/predictor/l3/b/data/percentile/5': 0.02219700478505716,\n",
" u'None/predictor/l3/b/data/percentile/6': 0.023460065061504942,\n",
" u'None/predictor/l3/b/data/std': 0.016169575022843975,\n",
" u'None/predictor/l3/b/data/zeros': 0.0,\n",
" u'None/predictor/l3/b/grad/max': 0.010434650448538984,\n",
" u'None/predictor/l3/b/grad/mean': 2.182454817514258e-09,\n",
" u'None/predictor/l3/b/grad/min': -0.010983652017118099,\n",
" u'None/predictor/l3/b/grad/percentile/0': -0.010925223681504266,\n",
" u'None/predictor/l3/b/grad/percentile/1': -0.009958908900198573,\n",
" u'None/predictor/l3/b/grad/percentile/2': -0.004830245970443916,\n",
" u'None/predictor/l3/b/grad/percentile/3': 0.00018675856216331018,\n",
" u'None/predictor/l3/b/grad/percentile/4': 0.004771847658956327,\n",
" u'None/predictor/l3/b/grad/percentile/5': 0.009501020630374713,\n",
" u'None/predictor/l3/b/grad/percentile/6': 0.010381417169433128,\n",
" u'None/predictor/l3/b/grad/std': 0.005975008951014995,\n",
" u'None/predictor/l3/b/grad/zeros': 0.0,\n",
" u'elapsed_time': 50.35235905647278,\n",
" u'epoch': 3,\n",
" u'iteration': 1800,\n",
" u'main/accuracy': 0.9768833428621292,\n",
" u'main/loss': 0.0734751478745602,\n",
" u'validation/main/accuracy': 0.9748000061511993,\n",
" u'validation/main/loss': 0.07976596570777474}]"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import json\n",
"with open(\"result/log\") as fin:\n",
" result = json.load(fin)\n",
"\n",
"result"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 2",
"language": "python",
"name": "python2"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 2
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython2",
"version": "2.7.12"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment