Cannot unserialize dataset without analysis

[ElodieENSTA]

The method Dataset.from_dict doesn't work since I upgraded from v0.2.5 to v0.4.0.

Here is an exemple of json that doesnt work:

{
  "datasets": {
    "original": {
      "class": "AudioDataset",
      "json": "[...]\\dataset\\small_dataset\\data\\audio\\original\\original.json"
    }
  },
  "depth": 200,
  "folder": "[...]\\dataset\\small_dataset",
  "gps_coordinates": [
    -49.47,
    70.09
  ],
  "instrument": {
    "end_to_end_db": 150.0,
    "gain_db": 0.0,
    "peak_voltage": 1.0,
    "sensitivity": 1.0
  },
  "strptime_format": "%y%m%d%H%M%S",
  "timezone": "UTC"
}

The error is:

KeyError: 'analysis'

In public_api/dataset.py line 727

[Gautzilla]

The analysis key has been added to link the exported core datasets to the analysis (this is usefull to grab matching audio/spectro datasets from a same analysis, or to rename/delete analyses).

Do you need me to work on the retrocompability of the unserialization? I didn't because the feature was added at a realtively early stage, but I can add a default analysis name that takes the name of the dataset or whatever if needed.

Otherwise, you just have to add the field as such:

{
  "datasets": {
    "original": {
      "analysis": "original",
      "class": "AudioDataset",
      "json": "[...]\\dataset\\small_dataset\\data\\audio\\original\\original.json"
    }

and for you analysis datasets:

{
  "datasets": {
    "original": {
      "analysis": "original",
      "class": "AudioDataset",
      "json": "[...]\\dataset\\small_dataset\\data\\audio\\original\\original.json"
    },
    "analsyis_1": {
      "analysis": "analsyis_1",
      "class": "AudioDataset",
      "json": "[...]\\dataset\\small_dataset\\data\\audio\\analsyis_1\\analsyis_1.json"
    }

[ElodieENSTA]

I came into a a second mismatch : the normalization and normalization_values that were added in audio data (both in audio and spectro dataset json).

I manually updated my json and it works again :)