regular_commands.py

'''
Set forwards hooks in PyTorch
'''
activation = {}
def getActivation(name):
    def hook(model, input, output):
        print(input.shape)
        if name not in activation:
            if torch.is_tensor(output):
                activation[name] = output.detach()
            else:
                activation[name] = output[0].detach()
    return hook
layer_interested_in = 9 
model.model.layers[layer_interested_in].self_attn.q_proj.register_forward_hook(getActivation(f"q{layer_interested_in}"))

'''
Check GPU memory usage
'''
mem_usage = [x/1e+9 for x in torch.cuda.mem_get_info(device=0)]
print("Memory consumption",mem_usage[1] - mem_usage[0])

'''
Your custom print function
'''
import builtins
f = open("tmp.txt", "w")
save_to_file = True

def print(*args):
    builtins.print(*args)
    if save_to_file:
        for strr in args:
            f.write(f"{strr} ")
        f.write("\n")
        f.flush()

'''
Put text using PIL with white box
'''
from PIL import Image, ImageDraw, ImageFont
image = Image.open(path)
draw = ImageDraw.Draw(image)
text = f"Some text goes here"
font = ImageFont.truetype("FreeMono.ttf", 64)
bbox = draw.textbbox((0, 0), text, font=font)
text_width, text_height = bbox[2] - bbox[0], bbox[3] - bbox[1]
x, y = 10, 10  # 10 pixels from the right edge
draw.rectangle([x, y, x + text_width, y + text_height], fill="white")
draw.text((x, y), text, font=font, fill="black")
image.save("img.jpg")

'''
Helpful terminal commands
'''
grep -rl "example" /home/user/documents
# find all files in the /home/user/documents directory that contain the word "example"

rsync -av --include='july11*/' --exclude='*' /source/path login-server:/destination/path/

cp -r `ls -A | grep -v "dir2"` /home/sk/backup/
# The command lists all files and directories in the current directory, excluding “dir2”.
# It then copies all the listed files and directories (except “dir2”) recursively to /home/sk/backup/                                                    

find . -type f -name '*.jpg' -delete
# searches all its subdirectories for *.jpg and deltes them

sudo kill -9 <pid> 
ps -u -p $pid
# Tells the username

du -sh .
# disk usage

df -h
# display information about the file system disk space usage

chmod -R 777 /home/jack/

watch -n 1 nvidia-smi
# displays nvidia-smi every 1 sec


'''
Conda commands
'''
conda info --envs
conda create -p /anaconda/envs/py171 --clone myenv
conda list
conda remove --name myenv --all

conda env export > environment.yaml
conda env create -f environment.yaml

conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
# where cuda needs to match torch.version.cuda
# For updated list see https://anaconda.org/nvidia/cuda-toolkit


'''
Pandas commands
'''
pd.read_csv(file, delimiter="|",quoting = csv.QUOTE_NONE)
# no characters will be considered as quote characters

mask = df["Column"].str.contains("id1|id2", case=False, na=False)
mask = csvFile_train["FNo"].isin([4,3])


df['Column'] = df['Column'].apply(lambda x: f"PREFIX{x}")

combine = pd.concat([df1, df2], axis=0, ignore_index=True).reset_index(drop=True)

df = df.assign(C=df['B'])
#Replicate column 'B' and assign it to a new column 'C'

df2.at[idx, "GT"] = df1.loc[num, "GT"]

df.insert(PosIndex, 'NewColumn', NewValue)

df['Column'] = df['Column'].str.replace('str1','str2')

'''
HuggingFace
'''
huggingface-cli download bert-base-uncased
# execute this command in terminal to download the specified HF model

'''
Python House Keeping
'''
shutil.rmtree(title,ignore_errors=True)

class Example:
    def __init__(self, x, y):
        self.x = x
        self.y = y

obj = Example(10, 20)

print("dir(obj):", dir(obj))
print("vars(obj):", vars(obj))
#dir(obj): ['__class__', '__delattr__', '__dict__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', '__weakref__', 'x', 'y']
#vars(obj): {'x': 10, 'y': 20}


'''
How to set up a infinite dataloader in PyTorch
'''
def cycle(dl):
    while True:
        for batch in dl:
            yield batch

self.dl = DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=2, drop_last=True)

dl = cycle(self.dl)
for ind in range(self.num_train_steps):
    data = next(dl)
    # inputs, targets = data
    # outputs = model(inputs)
    # loss = criterion(outputs, targets)
    # loss.backward()
    # optimizer.step()

'''
How to convert a batch(or more dim!) of PyTorch Images to a single large image
something like torchutils.make_grid() but with more flexibity in dimensions
'''
images.shape = (16,3,32,32)

from einops import rearrange, repeat
image = rearrange(images, '(row col) c h w -> c (row h) (col w)', row = desired_number_of_rows)