anasanchezf commited on
Commit
debdc6e
·
verified ·
1 Parent(s): 273e475

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +78 -0
README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Getting started
2
+
3
+ # Environment setup
4
+ Create a conda environment using the following command:
5
+ ```
6
+ conda env create -f environment.yml
7
+ conda activate plt
8
+ ```
9
+ You might need to adjust the CUDA versions.
10
+
11
+ # Evaluate your own images
12
+
13
+ We have uploaded an example folder in Hugging Face to ease playing around with our models and as a reference for how to format the data. But if you have your own images that wou would like to test, you can also use them!
14
+ You just need to have a folder with the TIFF files and a parquet index file.
15
+
16
+ ```bash
17
+ hf download anasanchezf/plate_effects example.tar.gz --type=dataset --local-dir="."
18
+ tar -xzf example.tar.gz
19
+
20
+ ```
21
+
22
+
23
+ ```python
24
+ import pandas as pd
25
+ from sklearn.metrics import accuracy_score
26
+ from sklearn.preprocessing import LabelEncoder
27
+ from plt.model.pretrained import PreTrainedARMBN
28
+
29
+ # COMPOUND ENCODING
30
+ # 'JCP2022_012818': 0 TC-S-7004
31
+ # 'JCP2022_025848': 1 dexamethasone
32
+ # 'JCP2022_035095': 2 LY2109761
33
+ # 'JCP2022_037716': 3 AMG900
34
+ # 'JCP2022_046054': 4 FK-866
35
+ # 'JCP2022_050797': 5 quinidine
36
+ # 'JCP2022_064022': 6 NVS-PAK1-1
37
+ # 'JCP2022_085227': 7 aloxistatin
38
+
39
+
40
+ perturbed = pd.read_parquet("example/plate_index.pq")
41
+ controls = pd.read_parquet("example/plate_index_controls.pq")
42
+
43
+ perturbed_and_controls = pd.concat([perturbed, controls])
44
+
45
+ model = PreTrainedCSARMBN()
46
+
47
+ model.adapt(perturbed_and_controls)
48
+ probs, uncertainty = model.predict(perturbed)
49
+
50
+ label_encoder = LabelEncoder()
51
+ labels = label_encoder.fit_transform(perturbed["Metadata_JCP2022"])
52
+ accuracy_score(labels, probs.argmax(dim=1).cpu())
53
+
54
+ ```
55
+
56
+ # Train your own adaptable models
57
+
58
+ ## Download the data
59
+ ``` bash
60
+ preprocess/images/01-download_indices.sh
61
+ preprocess/images/02-download_metadata.sh
62
+ python preprocess/images/03-create_indices.py
63
+ python preprocess/images/04-download_images_aws.py
64
+ ```
65
+
66
+ ## Run CS-ARM-BN
67
+ ``` bash
68
+ scripts/run_cs_arm_bn.sh
69
+ ```
70
+
71
+ ## Additional dependencies
72
+ To use the Channel Agnostic Masked Autoencoder baseline you will have to export or install the package as follows.
73
+
74
+ ``` bash
75
+ git clone https://github.com/recursionpharma/maes_microscopy.git
76
+ git clone https://huggingface.co/recursionpharma/OpenPhenom
77
+ export PYTHONPATH="${PYTHONPATH}:${PWD}/maes_microscopy"
78
+ ```