Exporting numpy array as csv imported to Tensorflow.js has no shape

Question

I need to export numpy arrays from Python to Javascript for use in Tensorflow.js. I had a look here and tried to do something similar but my imported tensor has no shape, it's not loaded properly.

Export from Python using Pandas dataframe looked like this (simplified):

A = np.array([[1,2,3],[4,5,6],[7,8,9],[10,11,12]])
A_df = pd.DataFrame(A)
A_df.to_csv('tmp.csv')

The tmp.csv looks like this:

,0,1,2
0,1,2,3
1,4,5,6
2,7,8,9
3,10,11,12

In Javascript it's loaded from url like this, when I test it on my local server:

const url = 'https://127.0.0.1/[localurlpath...]/tmp.csv'

const cols = ["", "0", "1", "2"]
let csvConfig = {
  hasHeader: true,
  columnNames: cols
}

const A = tf.data.csv(url, csvConfig)
console.log(A.shape)

Console log says: "undefined"

I have tried:

delimiter: '\n',
delimWhitespace: true

But to no avail. I also tried other column names to avoid the empty position. I get no errors in the Javascript console.

Answer 1

Since the numpy arrays in question aren't that big, aprox. 1000x50 at most, I ended upp using this method instead. CSV seemed overly complicated in the situation. Simplified example:

Export from Python:

A = np.array([[1,2,3],[4,5,6],[7,8,9],[10,11,12]])
with open('mtrx.json', 'w') as outfile:
    outfile.write(json.dumps(np.array(A).tolist()))

Import in Javascript:

const url = 'http://127.0.0.1/[...local filepath...]/mtrx.json'
const res = await fetch(url)
const mtrx = await JSON.parse(await res.text())

It turns out the imported mtrx object can be used in the tf.matMulcalculation I need to perform without any further conversion. I thought it needed to be converted to a tf.tensor. It was simpler than I thought. I still don't understand what's wrong with my first method but I'll save that for a rainy day.