diskmelts 2.1__tar.gz → 2.1.2__tar.gz

{diskmelts-2.1/src/diskmelts.egg-info → diskmelts-2.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diskmelts
-Version: 2.1
+Version: 2.1.2
 Summary: Machine-learning Enabled Line fitting Tool for Spectra of protoplanetary disks
 Author: Chengyan Xie, Dingshan Deng
 License-Expression: MIT
@@ -49,58 +49,30 @@ density (`logN`), and emitting area scaling (`A`) from observed spectra.
 
 ## Installation
 
-Requires Python 3.9 or newer. Once the package is published to PyPI, install
-it from any directory without cloning this repository:
+DiskMELTS is available on [PyPI](https://pypi.org/project/diskmelts/).
+The C6H6 and C3H4 checkpoints are included starting with version 2.1.2.
+With Python 3.9 or newer, install it from any directory. These commands use
+`python` in an activated environment; the
+[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
+shows how to create one on macOS, Linux, Windows, or Conda:
 
 ```bash
 python -m pip install diskmelts
-# Add [io] for FITS input, or [io,notebooks] for JupyterLab.
+python -c "import diskmelts; print('DiskMELTS is ready')"
 ```
 
-Until the first PyPI release, install directly from GitHub without a clone:
+Optional features can be installed in the same way:
 
 ```bash
-python -m pip install "diskmelts[io] @ git+https://github.com/wddlx/DiskMELTS.git"
+python -m pip install "diskmelts[io]"            # FITS input
+python -m pip install "diskmelts[io,notebooks]"  # FITS input and JupyterLab
 ```
 
-The package includes the production checkpoints and continuum reference tables.
-Full slab grids for training and direct-grid fitting are separate downloads;
-pass their local path as `grid_root`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
-for details.
-
-### Development checkout (macOS, Linux, Windows)
-
-From the repository root, use a native Python 3.10 or 3.11 interpreter to
-create a virtual environment. For FITS input and the example notebooks:
-
-```bash
-python3.11 -m venv .venv
-.venv/bin/python -m pip install -e ".[io,notebooks]"
-.venv/bin/python -c "import diskmelts; print(diskmelts.__name__)"
-```
-
-Use `python3.10` if that is the native Python available on your machine.
-For core fitting and CSV/array inputs only, replace the install command with
-`.venv/bin/python -m pip install -e .`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html) for Windows commands.
-
-### Conda or Mamba
-
-The YAML files use Conda only for Python and pip; pip installs the package and
-its declared dependencies. Choose one from the repository root:
-
-| File | Contents |
-| --- | --- |
-| `environment.yaml` | Core package |
-| `environment-notebooks.yaml` | Core package, FITS support, JupyterLab |
-| `environment-dev.yaml` | Notebook environment, tests, documentation tools |
-
-For the notebook setup, run `mamba env create --file environment-notebooks.yaml`
-then `conda activate diskmelts-notebooks`. `conda env create --file` works with
-the same files when using a current Conda release. On Apple Silicon, use a
-native `osx-arm64` installation such as Miniforge; an older Intel Conda can
-still spend a long time solving even these smaller files.
+The pip package includes the production checkpoints and continuum reference
+tables. Full slab grids for training and direct-grid fitting are separate
+downloads; pass their local path as `grid_root`.
+Example notebooks and spectra can be downloaded from the repository when
+needed; they are not installed by pip.
 
 ## Runtime dependencies
 
@@ -158,34 +130,44 @@ special regions, FITS input, saving CSVs, and uncertainty diagnostics.
 
 ## Quick fitting example
 
+Use a continuum-subtracted CSV with a header and `wave,flux` as its first two
+columns, with wavelength in µm and flux in Jy. For a runnable example without
+cloning the repository, download the
+[J16120505 spectrum](https://github.com/wddlx/DiskMELTS/blob/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv)
+as `j16120505_contsub.csv`, or run:
+
+```bash
+curl -L -o j16120505_contsub.csv https://raw.githubusercontent.com/wddlx/DiskMELTS/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv
+```
+
 ```python
-from diskmelts import load_fitting_models, load_observed_spectrum, fit_stage
+from diskmelts import fit_stage, load_fitting_models, load_observed_spectrum
 
-pretrained = load_fitting_models(['H2O'])
+models = load_fitting_models('C2H2')
 
 obs_wav, obs_flux = load_observed_spectrum(
-    'Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv'
+    'j16120505_contsub.csv'
 )
 
 fit = fit_stage(
     obs_wav,
     obs_flux,
-    molecule='H2O',
-    wavelength_range=(11.0, 19.0),
-    pretrained=pretrained,
-    fit_ranges=[(11.0, 12.0), (16.5, 18.5)],
+    molecule='C2H2',
+    wavelength_range=(13.0, 15.0),
+    pretrained=models,
+    fit_ranges=(13.5, 14.1),
     distance_pc=122.5,
-    n_samples=20000,
-    n_refine=32,
-    n_top=20,
-    sigma=0.001,
+    n_samples=1024,
+    n_refine=4,
+    n_top=4,
 )
 
-print(fit['params'])
+print(fit['params']['C2H2'])
 ```
 
 You can also read spectra with your own code. DiskMELTS only requires
-one-dimensional `obs_wav` and `obs_flux` arrays.
+one-dimensional `obs_wav` and `obs_flux` arrays. Replace the sample CSV with
+your own file to fit your source; update `distance_pc` and fit intervals.
 
 For a complete staged fit-and-subtract workflow from a checkout, edit and run:
 
@@ -215,9 +197,9 @@ plotting.
 
 ## Training and validation data
 
-`Model_grids/` and `Pretrain_grid/` are intentionally ignored by Git because
-they contain the full local training data. The generated checkpoint directories
-under the top-level `Trained_model/` are also optional; the package bundles
+Generated data under `Model_grids/` and `Pretrain_grid/` are intentionally
+ignored by Git because they contain the full local training data. Generated
+checkpoint directories under the top-level `Trained_model/` are also optional; the package bundles
 only the production checkpoints needed for v2 fitting. To run
 `examples/dev_v2_pt_validation.py` or
 `notebooks/Example_Training_Validation.ipynb`, place the complete model grids
@@ -228,7 +210,8 @@ The maintained v2 workflow uses a hybrid design after restricting the expanded
 IRIS grids to the intended physical domain. H2O uses five overlapping
 wavelength checkpoints covering 4.9–25.0 µm, all sharing `T=100–1400 K` and
 `logN=13–19`. Primary carbon-bearing and rarer molecules use one checkpoint
-each, the same temperature range, and `logN=13–19.5`.
+each and `logN=13–19.5`; C6H6 is limited to `T=100–675 K`, while the others
+use `T=100–1400 K`.
 The new C2HD checkpoint spans 11.5–20.5 µm to include its 19 µm band;
 CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 11.5–17 µm. To retrain these four grids, run:
@@ -237,6 +220,13 @@ CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 DISKMELTS_MOLECULES=13C12CH2,C2HD,CO17O,CO18O python examples/dev_v2_pt_validation.py
 ```
 
+C6H6 and C3H4 also have bundled fitting checkpoints. The C6H6 checkpoint
+covers 13–17 µm and 100–675 K; C3H4 covers 15–18 µm and 100–1400 K. Load
+either by name with `load_fitting_models`. Direct grid fitting
+uses the same names with local `Model_grids/C6H6/` and `Model_grids/C3H4/`
+files. See the [training guide](https://diskmelts.readthedocs.io/en/latest/training.html)
+for peak audits and retraining commands.
+
 Five to ten percent of complete `(T, logN)` grid spectra are excluded before
 PCA or neural-network training. They are used to select the PCA dimension and
 to validate spectral reconstruction and full parameter fitting. The

{diskmelts-2.1 → diskmelts-2.1.2}/README.md

@@ -10,58 +10,30 @@ density (`logN`), and emitting area scaling (`A`) from observed spectra.
 
 ## Installation
 
-Requires Python 3.9 or newer. Once the package is published to PyPI, install
-it from any directory without cloning this repository:
+DiskMELTS is available on [PyPI](https://pypi.org/project/diskmelts/).
+The C6H6 and C3H4 checkpoints are included starting with version 2.1.2.
+With Python 3.9 or newer, install it from any directory. These commands use
+`python` in an activated environment; the
+[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
+shows how to create one on macOS, Linux, Windows, or Conda:
 
 ```bash
 python -m pip install diskmelts
-# Add [io] for FITS input, or [io,notebooks] for JupyterLab.
+python -c "import diskmelts; print('DiskMELTS is ready')"
 ```
 
-Until the first PyPI release, install directly from GitHub without a clone:
+Optional features can be installed in the same way:
 
 ```bash
-python -m pip install "diskmelts[io] @ git+https://github.com/wddlx/DiskMELTS.git"
+python -m pip install "diskmelts[io]"            # FITS input
+python -m pip install "diskmelts[io,notebooks]"  # FITS input and JupyterLab
 ```
 
-The package includes the production checkpoints and continuum reference tables.
-Full slab grids for training and direct-grid fitting are separate downloads;
-pass their local path as `grid_root`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
-for details.
-
-### Development checkout (macOS, Linux, Windows)
-
-From the repository root, use a native Python 3.10 or 3.11 interpreter to
-create a virtual environment. For FITS input and the example notebooks:
-
-```bash
-python3.11 -m venv .venv
-.venv/bin/python -m pip install -e ".[io,notebooks]"
-.venv/bin/python -c "import diskmelts; print(diskmelts.__name__)"
-```
-
-Use `python3.10` if that is the native Python available on your machine.
-For core fitting and CSV/array inputs only, replace the install command with
-`.venv/bin/python -m pip install -e .`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html) for Windows commands.
-
-### Conda or Mamba
-
-The YAML files use Conda only for Python and pip; pip installs the package and
-its declared dependencies. Choose one from the repository root:
-
-| File | Contents |
-| --- | --- |
-| `environment.yaml` | Core package |
-| `environment-notebooks.yaml` | Core package, FITS support, JupyterLab |
-| `environment-dev.yaml` | Notebook environment, tests, documentation tools |
-
-For the notebook setup, run `mamba env create --file environment-notebooks.yaml`
-then `conda activate diskmelts-notebooks`. `conda env create --file` works with
-the same files when using a current Conda release. On Apple Silicon, use a
-native `osx-arm64` installation such as Miniforge; an older Intel Conda can
-still spend a long time solving even these smaller files.
+The pip package includes the production checkpoints and continuum reference
+tables. Full slab grids for training and direct-grid fitting are separate
+downloads; pass their local path as `grid_root`.
+Example notebooks and spectra can be downloaded from the repository when
+needed; they are not installed by pip.
 
 ## Runtime dependencies
 
@@ -119,34 +91,44 @@ special regions, FITS input, saving CSVs, and uncertainty diagnostics.
 
 ## Quick fitting example
 
+Use a continuum-subtracted CSV with a header and `wave,flux` as its first two
+columns, with wavelength in µm and flux in Jy. For a runnable example without
+cloning the repository, download the
+[J16120505 spectrum](https://github.com/wddlx/DiskMELTS/blob/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv)
+as `j16120505_contsub.csv`, or run:
+
+```bash
+curl -L -o j16120505_contsub.csv https://raw.githubusercontent.com/wddlx/DiskMELTS/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv
+```
+
 ```python
-from diskmelts import load_fitting_models, load_observed_spectrum, fit_stage
+from diskmelts import fit_stage, load_fitting_models, load_observed_spectrum
 
-pretrained = load_fitting_models(['H2O'])
+models = load_fitting_models('C2H2')
 
 obs_wav, obs_flux = load_observed_spectrum(
-    'Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv'
+    'j16120505_contsub.csv'
 )
 
 fit = fit_stage(
     obs_wav,
     obs_flux,
-    molecule='H2O',
-    wavelength_range=(11.0, 19.0),
-    pretrained=pretrained,
-    fit_ranges=[(11.0, 12.0), (16.5, 18.5)],
+    molecule='C2H2',
+    wavelength_range=(13.0, 15.0),
+    pretrained=models,
+    fit_ranges=(13.5, 14.1),
     distance_pc=122.5,
-    n_samples=20000,
-    n_refine=32,
-    n_top=20,
-    sigma=0.001,
+    n_samples=1024,
+    n_refine=4,
+    n_top=4,
 )
 
-print(fit['params'])
+print(fit['params']['C2H2'])
 ```
 
 You can also read spectra with your own code. DiskMELTS only requires
-one-dimensional `obs_wav` and `obs_flux` arrays.
+one-dimensional `obs_wav` and `obs_flux` arrays. Replace the sample CSV with
+your own file to fit your source; update `distance_pc` and fit intervals.
 
 For a complete staged fit-and-subtract workflow from a checkout, edit and run:
 
@@ -176,9 +158,9 @@ plotting.
 
 ## Training and validation data
 
-`Model_grids/` and `Pretrain_grid/` are intentionally ignored by Git because
-they contain the full local training data. The generated checkpoint directories
-under the top-level `Trained_model/` are also optional; the package bundles
+Generated data under `Model_grids/` and `Pretrain_grid/` are intentionally
+ignored by Git because they contain the full local training data. Generated
+checkpoint directories under the top-level `Trained_model/` are also optional; the package bundles
 only the production checkpoints needed for v2 fitting. To run
 `examples/dev_v2_pt_validation.py` or
 `notebooks/Example_Training_Validation.ipynb`, place the complete model grids
@@ -189,7 +171,8 @@ The maintained v2 workflow uses a hybrid design after restricting the expanded
 IRIS grids to the intended physical domain. H2O uses five overlapping
 wavelength checkpoints covering 4.9–25.0 µm, all sharing `T=100–1400 K` and
 `logN=13–19`. Primary carbon-bearing and rarer molecules use one checkpoint
-each, the same temperature range, and `logN=13–19.5`.
+each and `logN=13–19.5`; C6H6 is limited to `T=100–675 K`, while the others
+use `T=100–1400 K`.
 The new C2HD checkpoint spans 11.5–20.5 µm to include its 19 µm band;
 CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 11.5–17 µm. To retrain these four grids, run:
@@ -198,6 +181,13 @@ CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 DISKMELTS_MOLECULES=13C12CH2,C2HD,CO17O,CO18O python examples/dev_v2_pt_validation.py
 ```
 
+C6H6 and C3H4 also have bundled fitting checkpoints. The C6H6 checkpoint
+covers 13–17 µm and 100–675 K; C3H4 covers 15–18 µm and 100–1400 K. Load
+either by name with `load_fitting_models`. Direct grid fitting
+uses the same names with local `Model_grids/C6H6/` and `Model_grids/C3H4/`
+files. See the [training guide](https://diskmelts.readthedocs.io/en/latest/training.html)
+for peak audits and retraining commands.
+
 Five to ten percent of complete `(T, logN)` grid spectra are excluded before
 PCA or neural-network training. They are used to select the PCA dimension and
 to validate spectral reconstruction and full parameter fitting. The

{diskmelts-2.1 → diskmelts-2.1.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "diskmelts"
-version = "2.1"
+version = "2.1.2"
 description = "Machine-learning Enabled Line fitting Tool for Spectra of protoplanetary disks"
 readme = "README.md"
 requires-python = ">=3.9"

{diskmelts-2.1 → diskmelts-2.1.2}/src/diskmelts/fitting.py

@@ -519,6 +519,9 @@ def _build_components(molecules, h2o_components=1, component_names=None):
             raise ValueError('component_names must match the number of fitted components')
         for comp, label in zip(components, component_names):
             comp['label'] = label
+    labels = [comp['label'] for comp in components]
+    if len(labels) != len(set(labels)):
+        raise ValueError('component labels must be unique; name repeated molecules with component_names')
     return components
 
 
@@ -697,8 +700,9 @@ def fit_molecules(
             molecule names or component labels.
         logN_bounds (tuple or dict or None): default (14, 19).  A dict can use
             molecule names or component labels.
-        loga_bounds (tuple or None): bounds on log10(A).  None means A >= 0
-            with no upper bound; otherwise bounded least squares is used.
+        loga_bounds (tuple, dict, or None): bounds on log10(A), optionally
+            keyed by component label or molecule. None means A >= 0 with no
+            upper bound. Shared-area groups use the intersection of bounds.
         shared_area_groups (list of sequences or None): component labels whose
             emitting areas must be equal, e.g. [('C2H2', '13C12CH2')]. Each
             component may appear in at most one group. Unlisted components
@@ -914,10 +918,22 @@ def fit_molecules(
                 )
         return physical
 
-    if loga_bounds is None:
-        A_bounds = (0.0, np.inf)
-    else:
-        A_bounds = (10.0 ** float(loga_bounds[0]), 10.0 ** float(loga_bounds[1]))
+    component_area_bounds = []
+    for comp in components:
+        pair = _bounds_for(comp, loga_bounds, None)
+        if pair is None:
+            component_area_bounds.append((0.0, np.inf))
+            continue
+        if (len(pair) != 2 or not np.all(np.isfinite(pair)) or
+                pair[0] >= pair[1] or not 0 < 10.0 ** pair[0] < 10.0 ** pair[1]):
+            raise ValueError(f'{comp["label"]} needs finite increasing loga_bounds')
+        component_area_bounds.append((10.0 ** pair[0], 10.0 ** pair[1]))
+    area_lower = np.array([max(component_area_bounds[i][0] for i in members)
+                           for members in area_group_members])
+    area_upper = np.array([min(component_area_bounds[i][1] for i in members)
+                           for members in area_group_members])
+    if np.any(area_lower >= area_upper):
+        raise ValueError('shared area group has no overlap in loga_bounds')
 
     def _matrix_for(params):
         cols = []
@@ -939,12 +955,11 @@ def fit_molecules(
         flux_scale = max(float(np.max(np.abs(obs_fit))),
                          float(np.max(np.abs(M_group))), np.finfo(float).tiny)
         M_scaled, obs_scaled = M_group / flux_scale, obs_fit / flux_scale
-        if np.isfinite(A_bounds[1]):
+        if np.any(area_lower > 0) or np.any(np.isfinite(area_upper)):
             res = scipy.optimize.lsq_linear(
                 M_scaled,
                 obs_scaled,
-                bounds=(np.full(len(area_group_members), A_bounds[0]),
-                        np.full(len(area_group_members), A_bounds[1])),
+                bounds=(area_lower, area_upper),
             )
             A_group = np.asarray(res.x, dtype=np.float64)
         else:
@@ -1332,14 +1347,14 @@ def save_fitted_comparison(csv_path, source_name, all_fits):
 
     def _v(comp, key):
         if comp not in all_params:
-            return ''
+            return np.nan
         return round(all_params[comp][key], 4)
 
     def _e(comp, key, which):
         if comp not in all_unc:
-            return ''
+            return np.nan
         v = all_unc[comp].get(key, {}).get(which, float('nan'))
-        return round(v, 4) if np.isfinite(v) else ''
+        return round(v, 4) if np.isfinite(v) else np.nan
 
     row = {
         'Source':               source_name,

{diskmelts-2.1 → diskmelts-2.1.2}/src/diskmelts/plotting.py

@@ -43,6 +43,8 @@ _MOL_LABEL = {
     'C2HD':      r'C$_2$HD',
     'CO17O':     r'C$^{16}$O$^{17}$O',
     'CO18O':     r'C$^{16}$O$^{18}$O',
+    'C6H6':      r'C$_6$H$_6$',
+    'C3H4':      r'C$_3$H$_4$',
 }
 
 _MOL_COLOR = {
@@ -64,6 +66,8 @@ _MOL_COLOR = {
     'C2HD':     'teal',
     'CO17O':    'slateblue',
     'CO18O':    'goldenrod',
+    'C6H6':     'darkorange',
+    'C3H4':     'purple',
 }
 
 # ---------------------------------------------------------------------------
@@ -270,7 +274,8 @@ def plot_fit(
         ax_bot.text(0.99, 0.92, rf'$\chi^2_{{\rm red}} = {chi2_red:.2f}$',
                     transform=ax_bot.transAxes, ha='right', va='top', fontsize=9)
 
-        plt.tight_layout()
+        fig.subplots_adjust(left=0.11, right=0.98, bottom=0.10, top=0.92,
+                            hspace=0.08)
         if show:
             plt.show()
         if save_path is not None:

{diskmelts-2.1 → diskmelts-2.1.2}/src/diskmelts/v2_fitting.py

@@ -34,7 +34,8 @@ CARBON_RANGES = {
     'HC3N': (13.0, 18.0), 'C2H6': (9.0, 15.0),
     'C2H4': (8.0, 14.0), 'CH4': (5.0, 9.0),
     'C2HD': (11.5, 20.5), 'CO17O': (13.0, 18.0),
-    'CO18O': (13.0, 18.0),
+    'CO18O': (13.0, 18.0), 'C6H6': (13.0, 17.0),
+    'C3H4': (15.0, 18.0),
 }
 MODEL_SPECS = {
     'H2O': {
@@ -49,9 +50,9 @@ MODEL_SPECS = {
         mol: {
             'path': f'Trained_model/{mol}_single_v2/net_{mol}_forward.pt',
             'wavelength_ranges': (wav_range,),
-            'T_bounds': (100.0, 1400.0),
+            'T_bounds': (100.0, 675.0) if mol == 'C6H6' else (100.0, 1400.0),
             'logN_bounds': (13.0, 19.5),
-            'n_pca': 30,
+            'n_pca': 75 if mol == 'C3H4' else 30,
             'hidden': (64, 128, 64),
         }
         for mol, wav_range in CARBON_RANGES.items()

{diskmelts-2.1 → diskmelts-2.1.2}/src/diskmelts/v2_grid_fitting.py

@@ -58,8 +58,8 @@ def _stage_ranges(fit_ranges, components):
 
 def _components(molecule, h2o_components, component_names):
     molecules = [molecule] if isinstance(molecule, str) else list(molecule)
-    if not molecules or len(molecules) != len(set(molecules)):
-        raise ValueError('molecule must contain distinct molecular names')
+    if not molecules:
+        raise ValueError('molecule must contain at least one molecular name')
     if h2o_components not in (1, 2):
         raise ValueError('h2o_components must be 1 or 2')
     parts = []
@@ -281,10 +281,6 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
     noise_scale = float(sigma) * to_reference if sigma is not None else max(
         float(np.std(target)), np.finfo(float).tiny,
     )
-    lo_A, hi_A = 10.0 ** float(loga_bounds[0]), 10.0 ** float(loga_bounds[1])
-    if not 0 < lo_A < hi_A:
-        raise ValueError('loga_bounds must be increasing and finite')
-
     if n_refine < 1:
         raise ValueError('n_refine must be positive')
 
@@ -301,6 +297,14 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
     bounds = [(bound_for(T_bounds, part, (100.0, 1400.0)),
                bound_for(logN_bounds, part, default_logN(part)))
               for part in parts]
+    area_bounds = [bound_for(loga_bounds, part, (-2.0, 2.0))
+                   for part in parts]
+    if any(len(pair) != 2 or not np.all(np.isfinite(pair)) or
+           pair[0] >= pair[1] or not 0 < 10.0 ** pair[0] < 10.0 ** pair[1]
+           for pair in area_bounds):
+        raise ValueError('loga_bounds must be increasing and finite')
+    area_lower = np.array([10.0 ** pair[0] for pair in area_bounds])
+    area_upper = np.array([10.0 ** pair[1] for pair in area_bounds])
     molecules = list(dict.fromkeys(part['model_mol'] for part in parts))
     libraries = {}
     for mol in molecules:
@@ -330,7 +334,7 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
     def profile_amplitudes(nonlin):
         if (ordered_water and
                 nonlin[2 * water_indices[0]] > nonlin[2 * water_indices[1]]):
-            return np.inf, np.full(n_parts, lo_A)
+            return np.inf, area_lower.copy()
         matrix = templates(nonlin)
         scale = max(float(np.max(np.abs(target))),
                     float(np.max(np.abs(matrix))), np.finfo(float).tiny)
@@ -338,12 +342,12 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
             column = matrix[:, 0] / scale
             denom = float(np.dot(column, column))
             amplitude = (float(np.clip(np.dot(column, target / scale) / denom,
-                                       lo_A, hi_A)) if denom > 0 else lo_A)
+                                       area_lower[0], area_upper[0]))
+                         if denom > 0 else area_lower[0])
             amplitudes = np.array([amplitude])
         else:
             solution = lsq_linear(matrix / scale, target / scale,
-                                  bounds=(np.full(n_parts, lo_A),
-                                          np.full(n_parts, hi_A)))
+                                  bounds=(area_lower, area_upper))
             amplitudes = solution.x
         residual = matrix @ amplitudes - target
         return float(np.dot(residual, residual)), amplitudes
@@ -358,8 +362,8 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
                         key=lambda x: profile_amplitudes(x)[0])
     if not candidates:
         raise RuntimeError('grid search found no feasible component ordering')
-    solution_bounds = [entry for pair in bounds
-                       for entry in (pair[0], pair[1], loga_bounds)]
+    solution_bounds = [entry for k, pair in enumerate(bounds)
+                       for entry in (pair[0], pair[1], area_bounds[k])]
     lower = np.array([item[0] for item in solution_bounds], dtype=float)
     upper = np.array([item[1] for item in solution_bounds], dtype=float)
     scales = np.array([
@@ -445,8 +449,8 @@ def fit_grid_stage(obs_wav, obs_flux, molecule, wavelength_range, fit_ranges, *,
                                np.isclose(T, bounds[k][0][1], rtol=0, atol=1e-5)),
             'logN_at_bound': bool(np.isclose(logN, bounds[k][1][0], rtol=0, atol=1e-6) or
                                   np.isclose(logN, bounds[k][1][1], rtol=0, atol=1e-6)),
-            'log10A_at_bound': bool(np.isclose(logA, loga_bounds[0], rtol=0, atol=1e-6) or
-                                    np.isclose(logA, loga_bounds[1], rtol=0, atol=1e-6)),
+            'log10A_at_bound': bool(np.isclose(logA, area_bounds[k][0], rtol=0, atol=1e-6) or
+                                   np.isclose(logA, area_bounds[k][1], rtol=0, atol=1e-6)),
         }
     print(f'  Grid fit: {len(parts)} components, weighted SSE={best_loss:.4g}')
     return {

{diskmelts-2.1 → diskmelts-2.1.2/src/diskmelts.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: diskmelts
-Version: 2.1
+Version: 2.1.2
 Summary: Machine-learning Enabled Line fitting Tool for Spectra of protoplanetary disks
 Author: Chengyan Xie, Dingshan Deng
 License-Expression: MIT
@@ -49,58 +49,30 @@ density (`logN`), and emitting area scaling (`A`) from observed spectra.
 
 ## Installation
 
-Requires Python 3.9 or newer. Once the package is published to PyPI, install
-it from any directory without cloning this repository:
+DiskMELTS is available on [PyPI](https://pypi.org/project/diskmelts/).
+The C6H6 and C3H4 checkpoints are included starting with version 2.1.2.
+With Python 3.9 or newer, install it from any directory. These commands use
+`python` in an activated environment; the
+[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
+shows how to create one on macOS, Linux, Windows, or Conda:
 
 ```bash
 python -m pip install diskmelts
-# Add [io] for FITS input, or [io,notebooks] for JupyterLab.
+python -c "import diskmelts; print('DiskMELTS is ready')"
 ```
 
-Until the first PyPI release, install directly from GitHub without a clone:
+Optional features can be installed in the same way:
 
 ```bash
-python -m pip install "diskmelts[io] @ git+https://github.com/wddlx/DiskMELTS.git"
+python -m pip install "diskmelts[io]"            # FITS input
+python -m pip install "diskmelts[io,notebooks]"  # FITS input and JupyterLab
 ```
 
-The package includes the production checkpoints and continuum reference tables.
-Full slab grids for training and direct-grid fitting are separate downloads;
-pass their local path as `grid_root`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html)
-for details.
-
-### Development checkout (macOS, Linux, Windows)
-
-From the repository root, use a native Python 3.10 or 3.11 interpreter to
-create a virtual environment. For FITS input and the example notebooks:
-
-```bash
-python3.11 -m venv .venv
-.venv/bin/python -m pip install -e ".[io,notebooks]"
-.venv/bin/python -c "import diskmelts; print(diskmelts.__name__)"
-```
-
-Use `python3.10` if that is the native Python available on your machine.
-For core fitting and CSV/array inputs only, replace the install command with
-`.venv/bin/python -m pip install -e .`. See the
-[installation guide](https://diskmelts.readthedocs.io/en/latest/installation.html) for Windows commands.
-
-### Conda or Mamba
-
-The YAML files use Conda only for Python and pip; pip installs the package and
-its declared dependencies. Choose one from the repository root:
-
-| File | Contents |
-| --- | --- |
-| `environment.yaml` | Core package |
-| `environment-notebooks.yaml` | Core package, FITS support, JupyterLab |
-| `environment-dev.yaml` | Notebook environment, tests, documentation tools |
-
-For the notebook setup, run `mamba env create --file environment-notebooks.yaml`
-then `conda activate diskmelts-notebooks`. `conda env create --file` works with
-the same files when using a current Conda release. On Apple Silicon, use a
-native `osx-arm64` installation such as Miniforge; an older Intel Conda can
-still spend a long time solving even these smaller files.
+The pip package includes the production checkpoints and continuum reference
+tables. Full slab grids for training and direct-grid fitting are separate
+downloads; pass their local path as `grid_root`.
+Example notebooks and spectra can be downloaded from the repository when
+needed; they are not installed by pip.
 
 ## Runtime dependencies
 
@@ -158,34 +130,44 @@ special regions, FITS input, saving CSVs, and uncertainty diagnostics.
 
 ## Quick fitting example
 
+Use a continuum-subtracted CSV with a header and `wave,flux` as its first two
+columns, with wavelength in µm and flux in Jy. For a runnable example without
+cloning the repository, download the
+[J16120505 spectrum](https://github.com/wddlx/DiskMELTS/blob/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv)
+as `j16120505_contsub.csv`, or run:
+
+```bash
+curl -L -o j16120505_contsub.csv https://raw.githubusercontent.com/wddlx/DiskMELTS/main/Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv
+```
+
 ```python
-from diskmelts import load_fitting_models, load_observed_spectrum, fit_stage
+from diskmelts import fit_stage, load_fitting_models, load_observed_spectrum
 
-pretrained = load_fitting_models(['H2O'])
+models = load_fitting_models('C2H2')
 
 obs_wav, obs_flux = load_observed_spectrum(
-    'Realobs_data/Consub_data/j16120505_v9.0_contsub_RVcorr.csv'
+    'j16120505_contsub.csv'
 )
 
 fit = fit_stage(
     obs_wav,
     obs_flux,
-    molecule='H2O',
-    wavelength_range=(11.0, 19.0),
-    pretrained=pretrained,
-    fit_ranges=[(11.0, 12.0), (16.5, 18.5)],
+    molecule='C2H2',
+    wavelength_range=(13.0, 15.0),
+    pretrained=models,
+    fit_ranges=(13.5, 14.1),
     distance_pc=122.5,
-    n_samples=20000,
-    n_refine=32,
-    n_top=20,
-    sigma=0.001,
+    n_samples=1024,
+    n_refine=4,
+    n_top=4,
 )
 
-print(fit['params'])
+print(fit['params']['C2H2'])
 ```
 
 You can also read spectra with your own code. DiskMELTS only requires
-one-dimensional `obs_wav` and `obs_flux` arrays.
+one-dimensional `obs_wav` and `obs_flux` arrays. Replace the sample CSV with
+your own file to fit your source; update `distance_pc` and fit intervals.
 
 For a complete staged fit-and-subtract workflow from a checkout, edit and run:
 
@@ -215,9 +197,9 @@ plotting.
 
 ## Training and validation data
 
-`Model_grids/` and `Pretrain_grid/` are intentionally ignored by Git because
-they contain the full local training data. The generated checkpoint directories
-under the top-level `Trained_model/` are also optional; the package bundles
+Generated data under `Model_grids/` and `Pretrain_grid/` are intentionally
+ignored by Git because they contain the full local training data. Generated
+checkpoint directories under the top-level `Trained_model/` are also optional; the package bundles
 only the production checkpoints needed for v2 fitting. To run
 `examples/dev_v2_pt_validation.py` or
 `notebooks/Example_Training_Validation.ipynb`, place the complete model grids
@@ -228,7 +210,8 @@ The maintained v2 workflow uses a hybrid design after restricting the expanded
 IRIS grids to the intended physical domain. H2O uses five overlapping
 wavelength checkpoints covering 4.9–25.0 µm, all sharing `T=100–1400 K` and
 `logN=13–19`. Primary carbon-bearing and rarer molecules use one checkpoint
-each, the same temperature range, and `logN=13–19.5`.
+each and `logN=13–19.5`; C6H6 is limited to `T=100–675 K`, while the others
+use `T=100–1400 K`.
 The new C2HD checkpoint spans 11.5–20.5 µm to include its 19 µm band;
 CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 11.5–17 µm. To retrain these four grids, run:
@@ -237,6 +220,13 @@ CO17O and CO18O span 13–18 µm. The updated 13C12CH2 checkpoint keeps
 DISKMELTS_MOLECULES=13C12CH2,C2HD,CO17O,CO18O python examples/dev_v2_pt_validation.py
 ```
 
+C6H6 and C3H4 also have bundled fitting checkpoints. The C6H6 checkpoint
+covers 13–17 µm and 100–675 K; C3H4 covers 15–18 µm and 100–1400 K. Load
+either by name with `load_fitting_models`. Direct grid fitting
+uses the same names with local `Model_grids/C6H6/` and `Model_grids/C3H4/`
+files. See the [training guide](https://diskmelts.readthedocs.io/en/latest/training.html)
+for peak audits and retraining commands.
+
 Five to ten percent of complete `(T, logN)` grid spectra are excluded before
 PCA or neural-network training. They are used to select the PCA dimension and
 to validate spectral reconstruction and full parameter fitting. The

{diskmelts-2.1 → diskmelts-2.1.2}/src/diskmelts.egg-info/SOURCES.txt

@@ -34,7 +34,9 @@ src/diskmelts/models/Trained_model/C2H2_single_v2/net_C2H2_forward.pt
 src/diskmelts/models/Trained_model/C2H4_single_v2/net_C2H4_forward.pt
 src/diskmelts/models/Trained_model/C2H6_single_v2/net_C2H6_forward.pt
 src/diskmelts/models/Trained_model/C2HD_single_v2/net_C2HD_forward.pt
+src/diskmelts/models/Trained_model/C3H4_single_v2/net_C3H4_forward.pt
 src/diskmelts/models/Trained_model/C4H2_single_v2/net_C4H2_forward.pt
+src/diskmelts/models/Trained_model/C6H6_single_v2/net_C6H6_forward.pt
 src/diskmelts/models/Trained_model/CH4_single_v2/net_CH4_forward.pt
 src/diskmelts/models/Trained_model/CO17O_single_v2/net_CO17O_forward.pt
 src/diskmelts/models/Trained_model/CO18O_single_v2/net_CO18O_forward.pt

{diskmelts-2.1 → diskmelts-2.1.2}/tests/test_dev_v2_fitting_grids.py

@@ -14,12 +14,13 @@ from diskmelts.v2_grid_fitting import (
 )
 
 
-def _small_grid(tmp_path):
-    grid_dir = tmp_path / 'grids' / 'C2H2'
+def _small_grid(tmp_path, molecule='C2H2', wavelength_range=(11.0, 13.0)):
+    grid_dir = tmp_path / 'grids' / molecule
     grid_dir.mkdir(parents=True)
-    wav = np.linspace(11.0, 13.0, 60)
+    lo, hi = wavelength_range
+    wav = np.linspace(lo, hi, 60)
     basis = np.stack([np.exp(-((wav - center) / 0.09) ** 2)
-                      for center in (11.3, 12.0, 12.7)])
+                      for center in (lo + 0.3, (lo + hi) / 2, hi - 0.3)])
 
     def spectrum(T, logN):
         return 0.01 * (basis[0] + (1 + (T - 100) / 25) * basis[1]
@@ -33,6 +34,25 @@ def _small_grid(tmp_path):
     return wav, spectrum
 
 
+@pytest.mark.parametrize('molecule,wavelength_range', [
+    ('C6H6', (13.0, 17.0)),
+    ('C3H4', (15.0, 18.0)),
+])
+def test_new_molecules_fit_from_direct_grids(tmp_path, molecule,
+                                             wavelength_range):
+    wav, spectrum = _small_grid(tmp_path, molecule, wavelength_range)
+    observed = spectrum(113.0, 13.07)
+    fit = fit_grid_stage(
+        wav, observed, molecule, wavelength_range, wavelength_range,
+        grid_root=tmp_path / 'grids', T_bounds=(100, 150),
+        logN_bounds=(13, 13.25), sigma=1e-4,
+        maxiter=20, popsize=5, n_refine=2,
+    )
+    assert molecule in fit['params']
+    assert np.isfinite(fit['model_flux']).all()
+    assert np.linalg.norm(fit['residual']) < np.linalg.norm(observed)
+
+
 def test_grid_fit_and_saved_errors(tmp_path):
     wav, spectrum = _small_grid(tmp_path)
     distance_pc = 177.0
@@ -114,3 +134,42 @@ def test_grid_stage_rejects_unapplied_isotope_constraints(tmp_path):
             output_dir=tmp_path / 'results', detection_screening=False,
         )
     assert not (tmp_path / 'results').exists()
+
+
+def test_repeated_molecule_has_independent_component_bounds(tmp_path):
+    wav, spectrum = _small_grid(tmp_path)
+    observed = 0.7 * spectrum(110, 13.04) + 1.4 * spectrum(140, 13.20)
+    stage = {
+        'name': 'acetylene', 'mol': ['C2H2', 'C2H2'],
+        'component_names': ['C2H2_thin', 'C2H2_thick'],
+        'wavelength_range': (11, 13), 'fit_ranges': (11, 13),
+        'T_bounds': (100, 150),
+        'logN_bounds': {'C2H2_thin': (13.0, 13.1),
+                        'C2H2_thick': (13.15, 13.25)},
+        'loga_bounds': {'C2H2_thin': (-1.0, 0.0),
+                        'C2H2_thick': (0.0, 1.0)},
+    }
+    input_path = tmp_path / 'observation.csv'
+    pd.DataFrame({'wave': wav, 'flux': observed}).to_csv(input_path, index=False)
+    result = fit_observation_grids(
+        input_path, [stage], grid_root=tmp_path / 'grids',
+        output_dir=tmp_path / 'results', sigma_noise=1e-4,
+        detection_screening=False, maxiter=12, popsize=5, n_refine=2,
+    )
+    params = result['fits']['acetylene']['params']
+    assert set(params) == {'C2H2_thin', 'C2H2_thick'}
+    assert 13.0 <= params['C2H2_thin']['logN'] <= 13.1
+    assert 13.15 <= params['C2H2_thick']['logN'] <= 13.25
+    assert 0.1 <= params['C2H2_thin']['A'] <= 1.0
+    assert 1.0 <= params['C2H2_thick']['A'] <= 10.0
+    spectrum_csv = pd.read_csv(result['products']['acetylene']['spectrum'])
+    assert {'C2H2_thin_model', 'C2H2_thick_model'} <= set(spectrum_csv.columns)
+
+
+def test_repeated_grid_molecule_requires_unique_labels(tmp_path):
+    wav, spectrum = _small_grid(tmp_path)
+    with pytest.raises(ValueError, match='component labels must be unique'):
+        fit_grid_stage(
+            wav, spectrum(125, 13.125), ['C2H2', 'C2H2'],
+            (11, 13), (11, 13), grid_root=tmp_path / 'grids',
+        )

{diskmelts-2.1 → diskmelts-2.1.2}/tests/test_io.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import warnings
+
 import numpy as np
 import pandas as pd
 import pytest
@@ -53,3 +55,26 @@ def test_save_running_spectrum(tmp_path):
     assert list(df.columns) == ["wave", "residual", "cumulative_model"]
     assert len(df) == 50
     assert df["wave"].iloc[0] == pytest.approx(11.0)
+
+
+def test_save_fitted_comparison_updates_missing_numeric_fields(tmp_path):
+    pytest.importorskip("torch")
+    from diskmelts import save_fitted_comparison
+
+    path = tmp_path / "comparison.csv"
+    water = {"water": {"params": {
+        "H2O_warm": {"logN": 17.0, "T": 400.0, "log10A": 1.0},
+    }}}
+    carbon = {"carbon": {"params": {
+        "C2H2": {"logN": 18.0, "T": 600.0, "log10A": 2.0},
+    }}}
+    save_fitted_comparison(path, "source", water)
+    with warnings.catch_warnings():
+        warnings.simplefilter("error", FutureWarning)
+        save_fitted_comparison(path, "source", carbon)
+
+    df = pd.read_csv(path)
+    assert len(df) == 1
+    assert pd.isna(df.loc[0, "H2O_N_warm"])
+    assert pd.isna(df.loc[0, "C2H2_N_err_low"])
+    assert df.loc[0, "C2H2_N"] == pytest.approx(18.0)

{diskmelts-2.1 → diskmelts-2.1.2}/tests/test_shared_area_fitting.py

@@ -8,6 +8,44 @@ pytest.importorskip('torch')
 from diskmelts import fitting
 
 
+def test_repeated_model_molecule_uses_component_bounds(monkeypatch):
+    wave = np.linspace(13.0, 14.0, 41)
+    bases = (np.exp(-((wave - 13.3) / 0.1) ** 2),
+             np.exp(-((wave - 13.7) / 0.1) ** 2))
+
+    def fake_flux(model, T, logN, A, obs_wav):
+        shape = bases[0] + (logN - 14.0) * bases[1]
+        return A * np.interp(obs_wav, wave, shape)
+
+    monkeypatch.setattr(fitting, '_mol_flux_on_wav', fake_flux)
+    observed = fake_flux(None, 300, 14.5, 0.8, wave) + fake_flux(None, 500, 16.5, 1.5, wave)
+    fit = fitting.fit_molecules(
+        wave, observed, ['C2H2', 'C2H2'], {'C2H2': {'mol': 'C2H2'}},
+        (13.0, 14.0), component_names=['C2H2_thin', 'C2H2_thick'],
+        T_bounds=(200, 600),
+        logN_bounds={'C2H2_thin': (14.0, 15.0),
+                     'C2H2_thick': (16.0, 17.0)},
+        loga_bounds={'C2H2_thin': (-1.0, 0.0),
+                     'C2H2_thick': (0.0, 1.0)},
+        n_samples=16, n_refine=2, n_top=2, seed=7, verbose=False,
+    )
+    params = fit['params']
+    assert set(params) == {'C2H2_thin', 'C2H2_thick'}
+    assert 14.0 <= params['C2H2_thin']['logN'] <= 15.0
+    assert 16.0 <= params['C2H2_thick']['logN'] <= 17.0
+    assert 0.1 <= params['C2H2_thin']['A'] <= 1.0
+    assert 1.0 <= params['C2H2_thick']['A'] <= 10.0
+
+
+def test_repeated_model_molecule_requires_unique_labels():
+    with pytest.raises(ValueError, match='component labels must be unique'):
+        fitting.fit_molecules(
+            np.array([13.0, 14.0]), np.array([1.0, 1.0]),
+            ['C2H2', 'C2H2'], {'C2H2': {}}, (13.0, 14.0),
+            verbose=False,
+        )
+
+
 def test_shared_area_groups_use_one_amplitude_in_joint_fit(monkeypatch):
     wave = np.linspace(13.0, 16.0, 80)
     profiles = {

{diskmelts-2.1 → diskmelts-2.1.2}/tests/test_v2_package_assets.py

@@ -23,3 +23,22 @@ def test_bundled_model_fit_without_grids(tmp_path, monkeypatch):
     assert fit['params']['C2H2']['logN'] == pytest.approx(16, abs=0.02)
     with pytest.raises(ValueError, match='grid_root is required'):
         fit_grid_stage(wave, flux, 'C2H2', (13.5, 14.1), (13.5, 14.1))
+
+
+@pytest.mark.parametrize('molecule,window', [
+    ('C6H6', (14.5, 15.2)),
+    ('C3H4', (15.68, 15.85)),
+])
+def test_new_bundled_molecules_fit_without_grids(tmp_path, monkeypatch,
+                                                  molecule, window):
+    monkeypatch.chdir(tmp_path)
+    models = load_fitting_models(molecule)
+    wave = np.linspace(*window, 100)
+    flux = generate_spectrum(500, 16, 1.0, models[molecule], obs_wav=wave)
+    fit = fit_stage(
+        wave, flux, molecule, window, window, models,
+        n_samples=64, n_refine=2, n_top=2, seed=4,
+    )
+    assert molecule in fit['params']
+    assert np.isfinite(fit['model_flux']).all()
+    assert np.linalg.norm(fit['residual']) < np.linalg.norm(flux)

{diskmelts-2.1 → diskmelts-2.1.2}/tests/test_workflows.py

@@ -30,6 +30,14 @@ def test_notebook_code_cells_compile(path):
             compile("".join(cell["source"]), f"{path.name}:cell-{index}", "exec")
 
 
+@pytest.mark.parametrize("path", [FITTING_NOTEBOOK, TRAINING_NOTEBOOK])
+def test_notebook_rejects_incompatible_sklearn(path, monkeypatch):
+    sklearn = pytest.importorskip("sklearn")
+    monkeypatch.setattr(sklearn, "__version__", "0.0")
+    with pytest.raises(RuntimeError, match="require scikit-learn 1.8.0"):
+        exec(_code_cell(_notebook(path), 1), {})
+
+
 @pytest.mark.parametrize("start_dir", [ROOT, ROOT / "notebooks"])
 def test_consub_notebook_reads_public_usco_csv(monkeypatch, start_dir):
     raw_dir = ROOT / "Realobs_data" / "Raw_data"
@@ -46,6 +54,9 @@ def test_consub_notebook_reads_public_usco_csv(monkeypatch, start_dir):
 @pytest.mark.parametrize("start_dir", [ROOT, ROOT / "notebooks"])
 def test_fitting_notebook_resolves_committed_assets(monkeypatch, start_dir):
     pytest.importorskip("torch")
+    sklearn = pytest.importorskip("sklearn")
+    # This test checks paths and stage settings; the version guard is tested above.
+    monkeypatch.setattr(sklearn, "__version__", "1.8.0")
     notebook = _notebook(FITTING_NOTEBOOK)
     namespace = {}
 
@@ -65,6 +76,8 @@ def test_fitting_notebook_resolves_committed_assets(monkeypatch, start_dir):
 @pytest.mark.parametrize("start_dir", [ROOT, ROOT / "notebooks"])
 def test_training_notebook_resolves_local_data_locations(monkeypatch, start_dir):
     pytest.importorskip("torch")
+    sklearn = pytest.importorskip("sklearn")
+    monkeypatch.setattr(sklearn, "__version__", "1.8.0")
     notebook = _notebook(TRAINING_NOTEBOOK)
     namespace = {}
 
@@ -83,11 +96,11 @@ def test_production_checkpoint_files_are_bundled():
     from diskmelts import MODEL_SPECS
     from diskmelts.v2_fitting import MODEL_ROOT
 
-    assert len(MODEL_SPECS) == 14
+    assert len(MODEL_SPECS) == 16
     for molecule, spec in MODEL_SPECS.items():
         path = MODEL_ROOT / spec['path']
         assert path.is_file(), f'missing bundled checkpoint for {molecule}: {path}'
-    assert len(list(MODEL_ROOT.rglob('*.pt'))) == 18
+    assert len(list(MODEL_ROOT.rglob('*.pt'))) == 20
 
 
 def test_new_grid_checkpoints_load_and_generate_finite_spectra():
@@ -95,7 +108,7 @@ def test_new_grid_checkpoints_load_and_generate_finite_spectra():
     from diskmelts import load_fitting_models
     from diskmelts.fitting import generate_spectrum
 
-    molecules = ('13C12CH2', 'C2HD', 'CO17O', 'CO18O')
+    molecules = ('13C12CH2', 'C2HD', 'CO17O', 'CO18O', 'C6H6', 'C3H4')
     models = load_fitting_models(molecules)
     for molecule in molecules:
         flux = generate_spectrum(400., 16., 1., models[molecule])
@@ -111,7 +124,7 @@ def test_gitignore_separates_fitting_assets_from_training_data():
         if line.strip() and not line.lstrip().startswith("#")
     }
 
-    for pattern in ('/Model_grids/', '/Pretrain_grid/', '/Trained_model/',
+    for pattern in ('/Model_grids/*', '/Pretrain_grid/', '/Trained_model/',
                     '/0-Consub/',
                     '/docs/_build/', '/realobs_results/', '/figures/*'):
         assert pattern in patterns