Skip to content

Commit dba80c6

Browse files
rcmallircmalli
committed
Update README.md
1 parent 9d0f585 commit dba80c6

File tree

1 file changed

+83
-35
lines changed

1 file changed

+83
-35
lines changed

‎README.md

+83-35
Original file line numberDiff line numberDiff line change
@@ -8,10 +8,11 @@
88
<a href="https://black.readthedocs.io/en/stable/"><img alt="Code style: black" src="https://img.shields.io/badge/code%20style-black-000000.svg"></a>
99
</p>
1010

11-
This repository is the reimplementation of [MAML](https://arxiv.org/abs/1703.03400) (Model-Agnostic
12-
Meta-Learning) algorithm. Differentiable
13-
optimizers are handled by [Higher](https://github.com/facebookresearch/higher) library and [NN-template](https://github.com/lucmos/nn-template) is used for
14-
structuring
11+
This repository is the reimplementation
12+
of [MAML](https://arxiv.org/abs/1703.03400) (Model-Agnostic Meta-Learning)
13+
algorithm. Differentiable optimizers are handled
14+
by [Higher](https://github.com/facebookresearch/higher) library
15+
and [NN-template](https://github.com/lucmos/nn-template) is used for structuring
1516
the project. The default settings are used for training on Omniglot (5-way
1617
5-shot) problem. It can be easily extended for other few-shot datasets thanks to
1718
[Torchmeta](https://github.com/tristandeleu/pytorch-meta) library.
@@ -21,30 +22,32 @@ the project. The default settings are used for training on Omniglot (5-way
2122
**On Local Machine**
2223

2324
1. Download and install dependencies
25+
2426
```bash
2527
git clone https://github.com/rcmalli/lightning-maml.git
2628
cd ./lightning-maml/
2729
pip install -r requirements.txt
2830
```
2931

30-
2. Create `.env` file containing the info given below using your own [Wandb.
31-
ai](https://wandb.ai)
32+
2. Create `.env` file containing the info given below using your
33+
own [Wandb. ai](https://wandb.ai)
3234
account to track experiments. You can use `.env.template` file.
3335

3436
```bash
3537
export DATASET_PATH="/your/project/root/data/"
3638
export WANDB_ENTITY="USERNAME"
3739
export WANDB_API_KEY="KEY"
3840
```
41+
3942
3. Run the experiment
43+
4044
```bash
4145
python3 src/run.py train.pl_trainer.gpus=1
4246
```
4347

4448
**On Google Colab**
45-
46-
[![Google Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/rcmalli/lightning-maml/blob/main/notebooks/lightning_maml_pub.ipynb)
4749

50+
[![Google Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/rcmalli/lightning-maml/blob/main/notebooks/lightning_maml_pub.ipynb)
4851

4952
## Results
5053

@@ -53,44 +56,88 @@ python3 src/run.py train.pl_trainer.gpus=1
5356
<table class="tg">
5457
<thead>
5558
<tr>
56-
<th class="tg-0pky" colspan="2"></th>
57-
<th class="tg-7btt" colspan="2">Metatrain</th>
58-
<th class="tg-7btt" colspan="2">Metavalidation</th>
59+
<th colspan="3"></th>
60+
<th colspan="2">Metatrain</th>
61+
<th colspan="2">Metavalidation</th>
62+
</tr>
63+
</thead>
64+
<tbody>
65+
<tr>
66+
<td >Algorithm</td>
67+
<td >Model</td>
68+
<td >inner_steps</td>
69+
<td >inner accuracy</td>
70+
<td ><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
71+
<td ><span style="font-style:normal;text-decoration:none">inner accuracy</span></td>
72+
<td ><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
73+
</tr>
74+
<tr>
75+
<td >MAML</td>
76+
<td >OmniConv</td>
77+
<td >1</td>
78+
<td ></td>
79+
<td ></td>
80+
<td ></td>
81+
<td ></td>
82+
</tr>
83+
<tr>
84+
<td >MAML</td>
85+
<td >OmniConv</td>
86+
<td >5</td>
87+
<td ></td>
88+
<td ></td>
89+
<td ></td>
90+
<td ></td>
91+
</tr>
92+
</tbody>
93+
</table>
94+
95+
### MiniImageNet (5-way 5-shot)
96+
97+
<table class="tg">
98+
<thead>
99+
<tr>
100+
<th colspan="3"></th>
101+
<th colspan="2">Metatrain</th>
102+
<th colspan="2">Metavalidation</th>
59103
</tr>
60104
</thead>
61105
<tbody>
62106
<tr>
63-
<td class="tg-7btt">Algorithm</td>
64-
<td class="tg-7btt">inner_steps</td>
65-
<td class="tg-6ic8">inner accuracy</td>
66-
<td class="tg-6ic8"><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
67-
<td class="tg-4erg"><span style="font-style:normal;text-decoration:none">inner accuracy</span></td>
68-
<td class="tg-fymr"><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
107+
<td >Algorithm</td>
108+
<td >Model</td>
109+
<td >inner_steps</td>
110+
<td >inner accuracy</td>
111+
<td ><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
112+
<td ><span style="font-style:normal;text-decoration:none">inner accuracy</span></td>
113+
<td ><span style="font-style:normal;text-decoration:none">outer accuracy</span></td>
69114
</tr>
70115
<tr>
71-
<td class="tg-c3ow">MAML</td>
72-
<td class="tg-c3ow">1</td>
73-
<td class="tg-dvpl"></td>
74-
<td class="tg-dvpl"></td>
75-
<td class="tg-0pky"></td>
76-
<td class="tg-0pky"></td>
116+
<td >MAML</td>
117+
<td >MiniConv</td>
118+
<td >1</td>
119+
<td ></td>
120+
<td ></td>
121+
<td ></td>
122+
<td ></td>
77123
</tr>
78124
<tr>
79-
<td class="tg-c3ow">MAML</td>
80-
<td class="tg-c3ow">5</td>
81-
<td class="tg-dvpl"></td>
82-
<td class="tg-dvpl"></td>
83-
<td class="tg-0pky"></td>
84-
<td class="tg-0pky"></td>
125+
<td >MAML</td>
126+
<td >MiniConv</td>
127+
<td >5</td>
128+
<td ></td>
129+
<td ></td>
130+
<td ></td>
131+
<td ></td>
85132
</tr>
86133
</tbody>
87134
</table>
88135

89136
## Customization
90137

91-
Inside 'conf' folder, you can change all the settings depending on your
92-
problem or dataset. The default parameters are set for Omniglot dataset.
93-
Here are some examples for customization:
138+
Inside 'conf' folder, you can change all the settings depending on your problem
139+
or dataset. The default parameters are set for Omniglot dataset. Here are some
140+
examples for customization:
94141

95142
### Debug on local machine without GPU
96143

@@ -113,8 +160,8 @@ python3 src/run.py train.pl_trainer.gpus=1 data.datamodule.num_inner_steps=5,10,
113160

114161
### Using different dataset from Torchmeta
115162

116-
If you want to try a different dataset (ex. MiniImageNet), you can copy
117-
default.yaml file inside `conf/data` to `miniimagenet.yaml` and edit these
163+
If you want to try a different dataset (ex. MiniImageNet), you can copy
164+
default.yaml file inside `conf/data` to `miniimagenet.yaml` and edit these
118165
lines :
119166

120167
```yaml
@@ -142,7 +189,9 @@ datamodule:
142189

143190
# you may need to update data augmentation and preprocessing steps also!!!
144191
```
192+
145193
Run the experiment as follows:
194+
146195
```bash
147196
python3 src/run.py data=miniimagenet
148197
```
@@ -176,7 +225,6 @@ pytorch-lightning as high-level library
176225
outer_optimizer.step()
177226
```
178227

179-
180228
## References
181229

182230
- [Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks](https://arxiv.org/abs/1703.03400)

0 commit comments

Comments
 (0)