TypeError: '<=' not supported between instances of 'list' and 'float'

[mhorlacher]

Getting the above error when running MoDisco.

Minimal example to reproduce the error:

import modisco
import numpy as np
import tensorflow as tf

one_hot = tf.one_hot(np.random.randint(low=0, high=3, size=100*50).reshape(50, 100), depth=4).numpy()
print(one_hot.shape)

impscores = tf.random.uniform(shape=one_hot.shape).numpy()
print(impscores.shape)

null_per_pos_scores = modisco.coordproducers.LaplaceNullDist(num_to_samp=500)

tfmodisco_results = modisco.tfmodisco_workflow.workflow.TfModiscoWorkflow(
            target_seqlet_fdr=0.25,
            seqlets_to_patterns_factory=
                modisco.tfmodisco_workflow.seqlets_to_patterns.TfModiscoSeqletsToPatternsFactory(
                    initclusterer_factory=modisco.clusterinit.memeinit.MemeInitClustererFactory(    
                        meme_command="meme", base_outdir="meme_out",   
                        #max_num_seqlets_to_use specifies the maximum number of seqlets to use
                        # with MEME (this is to speed up MEME in the cases where the number of seqlets is
                        #  very large)
                        max_num_seqlets_to_use=10000,
                        nmotifs=10,
                        n_jobs=4),
                    use_louvain=False,
                    #Adjust trim_to_window_size and initial_flank_to_add
                    # according to how big you expect
                    # the core motif to be; default value is 10
                    trim_to_window_size=8,
                    initial_flank_to_add=2,
                    final_flank_to_add=5,
                    final_min_cluster_size=20
            ),
       )(
    #There is only one task, so we just call this 'task0'
    task_names=["task0"],
    contrib_scores={'task0': impscores},                
    hypothetical_contribs={'task0': impscores},
    one_hot=one_hot,
    null_per_pos_scores=null_per_pos_scores)

Any help welcome.

[halasadi]

I get the same issue sometimes, seems to be dependent on input data.

[AvantiShri]

Hi, sorry for the slow response (I've graduated and am no longer actively involved with this project) - you may want to refer to the TF-MoDISco lite repository (https://github.com/jmschrei/tfmodisco-lite/) for a more efficient, actively maintained, and easier-to-use version of the same algorithm.

[AvantiShri]

But to answer the question to the best of my ability (easier with a stack trace/output logs): when random or low-quality importance scores are used, the number of seqlets passing the filtering steps tends to become very small, and my guess is that the error is due to that

[gz-mg]

Hello, how did you solve it in the end?