phi-grib / flameweb Goto Github PK

It would be great to select various models and then retrieve a table with quality parameters. Even if it could be possible a bar plot showing the performance of the models, plus applicability domain.

It would be nice to get a radar plot for external validation (prediction) similar to the one obtained in model cross-validation.

Thanks!

Setting learning imbalance to none produces an error

PLS models must show a "conformal" control.

Some models do not show complete prediction results

Here is the python code for radial plot:

def radialplot(df, tnames, simcol,dilicol):
    ''' eg. radialplot(adf, 'atc_names_train', 'Tanimoto_fp', 'H_HT_class_train')'''
    labels = df[tnames].values
    stats = df[simcol].values
    dili = df[dilicol].values
    angles = np.linspace(0.07, 2*np.pi, len(labels), endpoint=False)
    stats = np.concatenate((stats,[stats[0]]))
    angles = np.concatenate((angles,[angles[0]]))
    fig=plt.figure(figsize=(10,20))

    ax = fig.add_subplot(111, polar=True)
    ax.plot(angles, stats, 'o',  linewidth=2, color='red')
    ax.fill(angles, stats, alpha=0.25, color='red')
    ax.set_thetagrids(angles * 180/np.pi, labels)
    ax.tick_params(pad=35)
    ax.set_rmax(1)
    ax.tick_params(axis='y', which='major', labelsize=14)

    ax.set_rticks([round(i,1) for i in np.linspace(0,1,11)])
    ax.set_rlabel_position(0)
    # ax.set_title('Reference: '+refname)
    ax.tick_params(axis='x', which='major', labelsize=18, grid_linestyle='-')

    ax.grid(True)
    sns.set_style("white")

In some cases the model parameters are not shown and an empty form, non-editable, is shown

Is there any way to give a different activity name than the one used for building the model in order to make a prediction and compare it with the experimental one? Then I will be able to obtain quality values such as sensitivity, specificity, ... and plots to validate my test sets.

The information regarding the percentage of samples inside the applicability domain is not shown in the web interface.

I would like to have a way to filter my models by qualitative and sort them by MCC.

It would be very useful to have a Venn diagram showing the overlapping between two models selected, my model and my test set, ...

When a model is being built, the summary shown at the model list must be cleared. Right now the results of the last model are shown and this is confussing

I cannot select various models and export them at once

when TSV is selected, the output is saved in flame_API folder. I suggest to give the power to the enduser to save and name the file wherever he/she wants.