Converting PyTorch Tensor to Pandas DataFrame

For converting a PyTorch Tensor to a Pandas DataFrame, first, you need to convert the tensor into a numpy array. Then, after getting a numpy array, you can convert it into a Pandas DataFrame.

So, it is a two-step process. However, these steps can vary depending on the current situation. For example, if your tensors are on the GPU, you need to first move to the CPU because the numpy array works on the CPU, not the GPU.

General conversion steps

1. If your tensor is on the GPU, move to the CPU using the .cpu() method. Avoid this step if you are already working on the CPU.
2. If gradients are required, you need to detach the tensor from the computation graph using the .detach() method.
3. Convert the tensor into a numpy array using the .numpy() method.
4. Convert that numpy array into a DataFrame using pd.DataFrame() method.

Install the Pandas and PyTorch libraries if you haven’t already.

Converting a 1D tensor into a DataFrame

import torch
import pandas as pd

tensor_1d = torch.tensor([48, 6, 19, 21])

print(tensor_1d)
# Output: tensor([48,  6, 19, 21])

print(type(tensor_1d))
# Output: <class 'torch.Tensor'>

# Converting to a data frame
df = pd.DataFrame(
    tensor_1d.cpu().numpy(),  # Ensures CPU and converts to NumPy
    columns=['Values']    # Explicitly name the column
)

print(df)
# Output:
#           Values
# 0           48
# 1            6
# 2           19
# 3           21

print(type(df))
# Output: <class 'pandas.core.frame.DataFrame'>

And we have a data frame with one column. You can also pass your index if you want custom row labels.

Converting 2D Tensor to Tabular DataFrame

import torch
import pandas as pd

tensor = torch.arange(6).reshape(2, 3)

df2 = pd.DataFrame(
    tensor.numpy(),
    index=["row1", "row2"],
    columns=["col1", "col2", "col3"]
)

print(df2)

# Output:
#        col1   col2   col3
# row1      0     1     2
# row2      3     4     5

Combining multiple tensors

Let’s combine multiple 1D vectors and create a DataFrame out of them.

import torch
import pandas as pd

a = torch.tensor([1, 2, 3])
b = torch.tensor([4, 5, 6])

df_combo = pd.concat(
    [pd.Series(a.numpy(), name="A"), pd.Series(b.numpy(), name="B")],
    axis=1
)
print(df_combo)

# Output:
#     A   B
# 0   1   4
# 1   2   5
# 2   3   6

That’s all!