APILASTRI
/
reformer_fastai
forked from Dean/reformer_fastai


  
1

	
2

	
3

	
4

	
5

	
6

	
7

	
8

	
9

	
10

	
11

	
12

	
13

	
14

	
15

	
16

	
17

	
18

	
19

	
20

	
21

	
22

	
23

	
24

	
25

	
26

	
27

	
28

	
29

	
30

	
31

	
32

	
33

	
34

	
35

	
36

	
37

	
38

	
39

	
40

	
41

	
42

	
43

	
44

	
45

	
46

	
47

	
48

	
49

	
50

	
51

	
52

	
53

	
54

	
55

	
56

	
57

	
58

	
59

	
60

	
61

	
62

	
63

	
64

	
65

	
66

	
67

	
68

	
69

	
70

	
71

	
72

	
73

	
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

	
101

	
102

	
103

	
104

	
105

	
106

	
107

	
108

	
109

	
110

	
111

	
112

	
113

	
114

	
115

	
116

	
117

	
118

	
119

	
120

	
121

	
122

	
123

	
124

	
125

	
126

	
127

	
128

	
129

	
130

	
131

	
132

	
133

	
134

	
135

	
136

	
137

	
138

	
139

	
140

	
141

	
142

	
143

	
144

	
145

	
146

	
147

	
148

	
149

	
150

	
151

	
152

	
153

	
154

	
155

	
156

	
157

	
158

	
159

	
160

	
161

	
162

	
163

	
164

	
165

	
166

	
167

	
168

	
169

	
170

	
171

	
172

	
173

	
174

	
175

	
176

	
177

	
178

	
179

	
180

	
181

	
182

	
183

	
184

	
185

	
186

	
187

	
188

	
189

	
190

	
191

	
192

	
193

	
            ---

title: Optimizers


keywords: fastai
sidebar: home_sidebar


nb_path: "nbs/08_optimizers.ipynb"
---
<!--

#################################################
### THIS FILE WAS AUTOGENERATED! DO NOT EDIT! ###
#################################################
# file to edit: nbs/08_optimizers.ipynb
# command to build the docs after a change: nbdev_build_docs

-->

<div class="container" id="notebook-container">
        
    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">

</div>
    {% endraw %}

    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">

</div>
    {% endraw %}

    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">

<div class="output_wrapper">
<div class="output">

<div class="output_area">


<div class="output_markdown rendered_html output_subarea ">
<h2 id="Adafactor" class="doc_header"><code>class</code> <code>Adafactor</code><a href="https://github.com/arampacha/reformer_fastai/tree/master/reformer_fastai/optimizers.py#L10" class="source_link" style="float:right">[source]</a></h2><blockquote><p><code>Adafactor</code>(<strong><code>params</code></strong>, <strong><code>lr</code></strong>=<em><code>None</code></em>, <strong><code>eps</code></strong>=<em><code>(1e-30, 0.001)</code></em>, <strong><code>clip_threshold</code></strong>=<em><code>1.0</code></em>, <strong><code>decay_rate</code></strong>=<em><code>-0.8</code></em>, <strong><code>mom</code></strong>=<em><code>None</code></em>, <strong><code>weight_decay</code></strong>=<em><code>0.0</code></em>, <strong><code>scale_parameter</code></strong>=<em><code>True</code></em>, <strong><code>relative_step</code></strong>=<em><code>True</code></em>, <strong><code>warmup_init</code></strong>=<em><code>False</code></em>) :: <code>Optimizer</code></p>
</blockquote>
<p>Base class for all optimizers.</p>
<p>.. warning::
    Parameters need to be specified as collections that have a deterministic
    ordering that is consistent between runs. Examples of objects that don't
    satisfy those properties are sets and iterators over values of dictionaries.</p>
<p>Arguments:
    params (iterable): an iterable of :class:<code>torch.Tensor</code> s or
        :class:<code>dict</code> s. Specifies what Tensors should be optimized.
    defaults: (dict): a dict containing default values of optimization
        options (used when a parameter group doesn't specify them).</p>

</div>

</div>

</div>
</div>

</div>
    {% endraw %}

    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">

</div>
    {% endraw %}

<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Implements Adafactor algorithm.</p>
<p>This implementation is based on: <code>Adafactor: Adaptive Learning Rates with Sublinear Memory Cost</code> (see <a href="https://arxiv.org/abs/1804.04235">https://arxiv.org/abs/1804.04235</a>)</p>
<p>Note that this optimizer internally adjusts the learning rate depending on the <em>scale_parameter</em>, <em>relative_step</em> and <em>warmup_init</em> options. To use a manual (external) learning rate schedule you should set <code>scale_parameter=False</code> and <code>relative_step=False</code>.</p>
<p><strong>Arguments</strong></p>

<pre><code>`params` (iterable): iterable of parameters to optimize or dicts defining parameter groups

`lr` (float, optional): external learning rate (default: None)

`eps` (tuple[float, float]): regularization constans for square gradient and parameter scale respectively (default: (1e-30, 1e-3))

`clip_threshold` (float): threshold of root mean square of final gradient update (default: 1.0)

`decay_rate` (float): coefficient used to compute running averages of square gradient (default: -0.8)

`mom` (float): coefficient used for computing running averages of gradient (default: None)

`weight_decay` (float, optional): weight decay (L2 penalty) (default: 0)

`scale_parameter` (bool): if True, learning rate is scaled by root mean square of parameter (default: True)

`relative_step` (bool): if True, time-dependent learning rate is computed instead of external learning rate (default: True)

`warmup_init` (bool): time-dependent learning rate computation depends on whether warm-up initialization is being used (default: False)</code></pre>

</div>
</div>
</div>
    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">

<div class="output_wrapper">
<div class="output">

<div class="output_area">


<div class="output_markdown rendered_html output_subarea ">
<h4 id="adafactor" class="doc_header"><code>adafactor</code><a href="https://github.com/arampacha/reformer_fastai/tree/master/reformer_fastai/optimizers.py#L182" class="source_link" style="float:right">[source]</a></h4><blockquote><p><code>adafactor</code>(<strong><code>param_groups</code></strong>, <strong><code>lr</code></strong>=<em><code>None</code></em>, <strong><code>eps</code></strong>=<em><code>(1e-30, 0.001)</code></em>, <strong><code>clip_threshold</code></strong>=<em><code>1.0</code></em>, <strong><code>decay_rate</code></strong>=<em><code>-0.8</code></em>, <strong><code>mom</code></strong>=<em><code>None</code></em>, <strong><code>weight_decay</code></strong>=<em><code>0.0</code></em>, <strong><code>scale_parameter</code></strong>=<em><code>True</code></em>, <strong><code>relative_step</code></strong>=<em><code>True</code></em>, <strong><code>warmup_init</code></strong>=<em><code>False</code></em>)</p>
</blockquote>

</div>

</div>

</div>
</div>

</div>
    {% endraw %}

    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">
<div class="input">

<div class="inner_cell">
    <div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="nd">@delegates</span><span class="p">(</span><span class="n">Adafactor</span><span class="o">.</span><span class="fm">__init__</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">adafactor</span><span class="p">(</span><span class="n">param_groups</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
    <span class="k">return</span> <span class="n">OptimWrapper</span><span class="p">(</span><span class="n">Adafactor</span><span class="p">([{</span><span class="s1">&#39;params&#39;</span><span class="p">:</span> <span class="n">ps</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">}</span> <span class="k">for</span> <span class="n">ps</span> <span class="ow">in</span> <span class="n">param_groups</span><span class="p">]))</span>
</pre></div>

    </div>
</div>
</div>

</div>
    {% endraw %}

<div class="cell border-box-sizing text_cell rendered"><div class="inner_cell">
<div class="text_cell_render border-box-sizing rendered_html">
<p>Wrapping a pytorch optimizer in fastai's <code>OptimWrapper</code> enables its use with fastai</p>

</div>
</div>
</div>
    {% raw %}
    
<div class="cell border-box-sizing code_cell rendered">
<div class="input">

<div class="inner_cell">
    <div class="input_area">
<div class=" highlight hl-ipython3"><pre><span></span><span class="n">ps</span> <span class="o">=</span>  <span class="p">[</span><span class="n">tensor</span><span class="p">([</span><span class="mi">1</span><span class="p">,</span><span class="mi">2</span><span class="p">,</span><span class="mi">3</span><span class="p">])]</span> <span class="c1">#, tensor([4,5,6])]</span>

<span class="n">adaf</span> <span class="o">=</span> <span class="n">Adafactor</span><span class="p">(</span><span class="n">ps</span><span class="p">,</span> <span class="n">mom</span><span class="o">=</span><span class="mf">0.9</span><span class="p">,</span> <span class="n">weight_decay</span><span class="o">=</span><span class="mf">1e-2</span><span class="p">)</span>
<span class="n">test_adaf</span> <span class="o">=</span> <span class="n">adafactor</span><span class="p">(</span><span class="n">param_groups</span><span class="o">=</span><span class="n">ps</span><span class="p">,</span> <span class="n">mom</span><span class="o">=</span><span class="mf">0.9</span><span class="p">,</span> <span class="n">weight_decay</span><span class="o">=</span><span class="mf">1e-2</span><span class="p">)</span>

<span class="c1">#Access to param_groups</span>
<span class="n">test_eq</span><span class="p">(</span><span class="n">test_adaf</span><span class="o">.</span><span class="n">param_lists</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">adaf</span><span class="o">.</span><span class="n">param_groups</span><span class="p">[</span><span class="mi">0</span><span class="p">][</span><span class="s1">&#39;params&#39;</span><span class="p">])</span>
<span class="c1">#Set param_groups</span>
<span class="n">test_adaf</span><span class="o">.</span><span class="n">param_lists</span> <span class="o">=</span> <span class="p">[[</span><span class="n">tensor</span><span class="p">([</span><span class="mi">4</span><span class="p">,</span><span class="mi">5</span><span class="p">,</span><span class="mi">6</span><span class="p">])]]</span>
<span class="n">test_eq</span><span class="p">(</span><span class="n">test_adaf</span><span class="o">.</span><span class="n">opt</span><span class="o">.</span><span class="n">param_groups</span><span class="p">[</span><span class="mi">0</span><span class="p">][</span><span class="s1">&#39;params&#39;</span><span class="p">],</span> <span class="p">[</span><span class="n">tensor</span><span class="p">(</span><span class="mi">4</span><span class="p">,</span><span class="mi">5</span><span class="p">,</span><span class="mi">6</span><span class="p">)])</span>
<span class="c1">#Access to hypers</span>
<span class="c1"># test_eq(test_adaf.hypers, [{**adaf.defaults}])</span>
<span class="c1"># #Set hypers</span>
<span class="n">test_adaf</span><span class="o">.</span><span class="n">set_hyper</span><span class="p">(</span><span class="s1">&#39;mom&#39;</span><span class="p">,</span> <span class="mf">0.95</span><span class="p">)</span>
<span class="n">test_eq</span><span class="p">(</span><span class="n">test_adaf</span><span class="o">.</span><span class="n">opt</span><span class="o">.</span><span class="n">param_groups</span><span class="p">[</span><span class="mi">0</span><span class="p">][</span><span class="s1">&#39;mom&#39;</span><span class="p">],</span> <span class="mf">0.95</span><span class="p">)</span>
</pre></div>

    </div>
</div>
</div>

</div>
    {% endraw %}

</div>