Class: Torchrb::Torch

Inherits:
Lua
  • Object
show all
Defined in:
lib/torchrb/torch.rb

Direct Known Subclasses

Wrapper

Instance Attribute Summary collapse

Attributes inherited from Lua

#debug, #enable_cuda

Instance Method Summary collapse

Methods inherited from Lua

#eval

Constructor Details

#initialize(options = {}) ⇒ Torch

Returns a new instance of Torch.



7
8
9
10
11
12
 
# File 'lib/torchrb/torch.rb', line 7

def initialize options={}
  super
  @network_loaded = false
  @error_rate = Float::NAN
  load_network options[:network_storage_path] unless network_loaded rescue nil
end

Instance Attribute Details

#error_rate ⇒ Object

Returns the value of attribute error_rate.



5
6
7
 
# File 'lib/torchrb/torch.rb', line 5

def error_rate
  @error_rate
end

#network_loaded ⇒ Object

Returns the value of attribute network_loaded.



3
4
5
 
# File 'lib/torchrb/torch.rb', line 3

def network_loaded
  @network_loaded
end

#network_timestamp ⇒ Object

Returns the value of attribute network_timestamp.



4
5
6
 
# File 'lib/torchrb/torch.rb', line 4

def network_timestamp
  @network_timestamp
end

Instance Method Details

#cudify ⇒ Object 



102
103
104
105
106
107
108
109
110
111
112
113
114
115
 
# File 'lib/torchrb/torch.rb', line 102

def cudify
  eval "        -- print(sys.COLORS.red .. '==> using CUDA GPU #' .. cutorch.getDevice() .. sys.COLORS.black)\n        train_set.input = train_set.input:cuda()\n        train_set.label = train_set.label:cuda()\n        test_set.input = test_set.input:cuda()\n        test_set.label = test_set.label:cuda()\n        validation_set.input = validation_set.input:cuda()\n        validation_set.label = validation_set.label:cuda()\n\n        criterion = nn.ClassNLLCriterion():cuda()\n        net = cudnn.convert(net:cuda(), cudnn)\n  EOF\nend\n", __FILE__, __LINE__

#iteration_callback=(callback) ⇒ Object 



14
15
16
17
18
19
20
 
# File 'lib/torchrb/torch.rb', line 14

def iteration_callback= callback
  state.function "iteration_callback" do |trainer, iteration, currentError|
    progress = iteration / state['number_of_iterations']
    self.error_rate = currentError/100.0
    callback.call progress: progress, error_rate: error_rate
  end
end

#load_network(network_storage_path) ⇒ Object 



51
52
53
54
55
56
57
58
59
60
61
62
63
64
 
# File 'lib/torchrb/torch.rb', line 51

def load_network network_storage_path
  raise "Neuronal net not trained yet. Call 'Torch#update_training_data'." unless File.exist?(network_storage_path)
  metadata = eval("      net = torch.load('\#{network_storage_path}')\n      metadata = torch.load('\#{network_storage_path}.meta')\n      classes = metadata[1]\n      timestamp = metadata[3]\n      return metadata[2]\n  EOF\n  self.error_rate = metadata\n  self.network_timestamp = @state['timestamp']\n  puts \"Network with metadata [\#{@state['classes'].to_h}, \#{error_rate}] loaded from \#{network_storage_path} @ \#{network_timestamp}\" if debug\n  self.network_loaded = true\nend\n", __FILE__, __LINE__).to_ruby

#predict(sample, network_storage_path = nil) ⇒ Object 



36
37
38
39
40
41
42
43
44
45
46
47
48
49
 
# File 'lib/torchrb/torch.rb', line 36

def predict sample, network_storage_path=nil
  load_network network_storage_path unless network_loaded

  classes = eval "      \#{sample.to_tensor(\"sample_data\").strip}\n      local prediction = \#{enable_cuda ? \"net:forward(sample_data:cuda()):float()\" : \"net:forward(sample_data)\"}\n      prediction = prediction:exp()\n      confidences = prediction:totable()\n      return classes\n  EOF\n  puts \"predicted \#{@state['confidences'].to_h} based on network @ \#{network_timestamp}\" if debug\n  classes = classes.to_h\n  @state['confidences'].to_h.map { |k, v| {classes[k] => v} }.reduce({}, :merge)\nend\n", __FILE__, __LINE__

#print_results ⇒ Object 



74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
 
# File 'lib/torchrb/torch.rb', line 74

def print_results
  result = eval "      class_performance = torch.LongTensor(#classes):fill(0):totable()\n      test_set_size = test_set:size()\n      for i=1,test_set_size do\n          local groundtruth = test_set.label[i]\n          local prediction = net:forward(test_set.input[i])\n          local confidences, indices = torch.sort(prediction, true)  -- true means sort in descending order\n\n          class_performance[groundtruth] = class_performance[groundtruth] + 1\n\n      end\n\n      local result = {}\n      for i=1,#classes do\n        local confidence = 100*class_performance[i]/test_set_size\n        table.insert(result, { classes[i], confidence } )\n      end\n      return result\n  EOF\n  result = result.to_ruby.map(&:to_ruby)\n  if defined?(DEBUG)\n    puts \"#\" * 80\n    puts \"Results: \#{result.to_h}\"\n    puts \"#\" * 80\n  end\nend\n", __FILE__, __LINE__

#store_network(network_storage_path) ⇒ Object 



66
67
68
69
70
71
72
 
# File 'lib/torchrb/torch.rb', line 66

def store_network network_storage_path
  eval "      torch.save('\#{network_storage_path}', net)\n      torch.save('\#{network_storage_path}.meta', {classes, \#{error_rate}, '\#{network_timestamp}}'} )\n  EOF\n  puts \"Network with metadata [\#{@state['classes'].to_h}, \#{error_rate}] stored in \#{network_storage_path} @ \#{network_timestamp}\" if debug\nend\n", __FILE__, __LINE__

#train ⇒ Object 



22
23
24
25
26
27
28
29
30
31
32
33
34
 
# File 'lib/torchrb/torch.rb', line 22

def train
  eval "      local oldprint = print\n      print = function(...)\n      end\n\n      trainer:train(train_set)\n\n      print = oldprint\n  EOF\n  self.network_loaded = true\n  self.network_timestamp = Time.now\nend\n", __FILE__, __LINE__